Improving EEG: From Simulations to Applications

Inaugural-Dissertation zur Erlangung der Doktorwürde der Fakultät für Biologie der Albert-Ludwigs-Universität Freiburg im Breisgau

vorgelegt von Lukas Dominique Josef Fiederer geboren in Meyrin, Schweiz

Freiburg im Breisgau Juli 2018 Angefertigt an der Uniklinik Freiburg

The research presented in this thesis was carried out at the Institute for Biology I of the Albert-Ludwig-University Freiburg, in collaboration with the University Hospital of the University of Freiburg, from September 2012 to August 2017.

Dekanin der Fakultät für Biologie: Prof. Dr. Bettina Warscheid Promotionsvorsitzender: Prof. Dr. Andreas Hiltbrunner

Betreuer der Arbeit: PD Dr.med. Tonio Ball

Referent: PD Dr. med. Tonio Ball Koreferent: Drittprüfer:

Datum der mündlichen Prüfung:

1 Table of Contents

TABLE OF CONTENTS

ACKNOWLEDGEMENTS ...... 3 DECLARATION ...... 4

ABSTRACT ...... 6

I. INTRODUCTION ...... 9

1) ...... 9 LECTROPHYSIOLOGY A. 2) E ...... & stereotactic electroencephalography ...... 10. 9 3) Local field potentials ...... 11 4) Multi-unit & single-unit activity ...... 12

B. MAGNETIC RESONANCE IMAGING ...... 12 OLUME ONDUCTOR EAD ODELING C. V C H M ...... 14 D. SOMATOSENSORY EVOKED POTENTIALS ...... 16 II. SUMMARYBJECT OF OF THIS THE HESIS RESULTS ...... 18 E. O T ...... 16

A. SPATIAL AND FREQUENCY-DOMAIN CHARACTERISTICS OF INTRACRANIALLY-MEASURED EMG ...... 18 NDOGENOUS RAIN TIMULATION THROUGH B. E B S EMG ...... 18 C. VOLUME CONDUCTION MODELING OF BLOOD VESSELS ...... 18 III. DISCUSSIONODELING µ...... O EURONAL OURCES BASED ON PAPERS 19 D. M EC G N S ( 2 ) ...... 18

A. MODEL-GUIDED IMPROVEMENTS OF INTRACRANIAL ELECTROPHYSIOLOGICAL MEASUREMENTS ...... 19 HEWING ELATED NDOGENOUS RAIN TIMULATION URTHER TEPS TO ROVE YPOTHESIS B. C -R E B S : F S P H ...... 21 C. CHEWING-RELATED ENDOGENOUS BRAIN STIMULATION– : BENEFICIAL OR DETRIMENTAL? THOUGHTS ON PPLICATIONS TO VOLUTION AND OOD NTAKE A E F I ...... 22 D. BLOOD VESSELS AS WINDOWS TO THE BRAIN A RECORDING AND STIMULATION PERSPECTIVE...... 23 IV. REFERENCESEPTH OCALIZATION ...... OF EURONAL CTIVITY 27 E. D L N A ...... 25 V. SPATIAL AND FREQUENCY-DOMAIN CHARACTERISTICS OF INTRACRANIALLY- MEASURED EMG ...... 35 VI. ELECTRICAL STIMULATION OF THE HUMAN CEREBRAL CORTEX BY EXTRACRANIAL MUSCLE ACTIVITY: EFFECT QUANTIFICATION WITH INTRACRANIAL EEG AND FEM SIMULATIONS ...... 51 VII. THE ROLE OF BLOOD VESSELS IN HIGH-RESOLUTION VOLUME CONDUCTOR HEAD MODELING OF EEG ...... 65 VIII. MAPPING THE FINE STRUCTURE OF CORTICAL ACTIVITY WITH DIFFERENT MICRO-ECOG ELECTRODE ARRAY GEOMETRIES ...... 83 IX. CORTICAL-DEPTH AND FREQUENCY-BAND DEPENDENCY OF THE SPATIAL EXTENT OF THE GENERATORS UNDERLYING µECOG RECORDINGS ...... 107

2 Acknowledgements

ACKNOWLEDGEMENTS I would like to thank my supervisor, Tonio Ball, for his continued support and valuable guidance. Without him and the possibilities he created this work would not have been possible and I would probably not be the same ;-) Thank you Tonio! I would like to thank my colleagues for their support, for all the more or less fruitful discussions and for the great times spent together :-D You know who you are! I would like to thank my parents who have made my studies possible and who infused me with the moral and perseverance I needed to accomplish my PhD. Without them I would have sought much quicker ‘satisfaction’. Merci maman, danke Papa! <3 I would like to thank the love of my life, who incidentally became my wife during this work, and my children, born during this work, for lighting up my life. Without them life would be much more … dull … (and quiet). Merci mes chéries! <3 <3 <3 Lastly, I would like to once again thank my wife, this time for all her loving support and for leveraging the tremendous of work all her ‘children’ require of her. Danke mein Schatz, ich liebe dich!

3 Declaration

DECLARATION The detailed results of selected work performed during the time of the thesis can be found in the appended publications and manuscripts. The contributions of the individual authors are declared below.

THE ROLE OF BLOOD VESSELS IN HIGH-RESOLUTION VOLUME CONDUCTOR HEAD MODELING OF EEG Fiederer L.D.J., Vorwerk J., Lucka F., Dannhauer M., Yang S., Dümpelmann M., Schulze- Bonhage A., Aertsen A., Speck O., Wolters C.H., Ball T.

T. Ball and L.D.J. Fiederer devised the study. L.D.J. Fiederer programed and performed the segmentations, created the models, devised, ran and evaluated the simulations. J. Vorwerk, F. Lucka, M. Dannhauer, M. Dümpelmann and C.H. Wolters introduced L.D.J. Fiederer to the simulation and visualization software, provided support and counseling. S. Yang acquired the 7T MRI data and performed initial skull stripping and brain segmentation using Freesurfer, under the supervision of O. Speck. L.D.J. Fiederer and T. Ball wrote the paper with the help of J. Vorwerk, F. Lucka, M. Dannhauer, S. Yang, M. Dümpelmann, A. Schulze-Bonhage, A. Aertsen, O. Speck and C.H. Wolters. L.D.J. Fiederer created the visualizations, figures and designed the cover image.

SPATIAL AND FREQUENCY-DOMAIN CHARACTERISTICS OF INTRACRANIALLY-MEASURED EMG J. Lahr, L.D.J. Fiederer, O. Iljina, A. Aertsen, A. Schulze-Bonhage, T. Ball

T. Ball and J. Lahr devised the study. J. Lahr evaluated the chewing data and performed the data analysis. O. Iljina evaluated the linguistic data. J. Lahr, L.D.J. Fiederer and T. Ball wrote the paper with the help of O. Iljina, A. Aertsen and A. Schulze-Bonhage. J. Lahr created the visualizations. J. Lahr and L.D.J. Fiederer created the figures.

ELECTRICAL STIMULATION OF THE HUMAN CEREBRAL CORTEX BY EXTRACRANIAL MUSCLE ACTIVITY: EFFECT QUANTIFICATION WITH INTRACRANIAL EEG AND FEM SIMULATIONS Lukas Dominique Josef Fiederer, Jacob Lahr, Johannes Vorwerk, Felix Lucka, Ad Aertsen, Carsten Hermann Wolters, Andreas Schulze-Bonhage, Tonio Ball

T. Ball, L.D.J. Fiederer and J. Lahr devised the study. L.D.J. Fiederer programed and performed the segmentations, created the models, devised, ran and evaluated the simulations, measured and analyzed the non-invasive EEG data. J. Lahr analyzed the invasive EEG data. J. Vorwerk, F. Lucka, and C.H. Wolters provided support and counseling. A. Schulze-Bonhage provided access to the patients data. L.D.J. Fiederer, T. Ball and J. Lahr wrote the paper with the help of J. Vorwerk, F. Lucka, A. Aertsen, C.H. Wolters and A. Schulze-Bonhage. L.D.J. Fiederer and J. Lahr created the visualizations and figures, L.D.J. Fiederer created the highlight animations.

MAPPING THE FINE STRUCTURE OF CORTICAL ACTIVITY WITH µECOG Xi Wang, C. Alexis Gkogkidis, Olga Iljina, Lukas D.J. Fiederer, Christian Henle, Irina Mader, Jan Kaminsky, Thomas Stieglitz, Mortimer Gierthmuehlen, Tonio Ball

4 Declaration

T. Ball and M. Gierthmühlen devised the study. C.A. Gkogkidis and X. Wang performed the experiments. C. Henle created the µECoG arrays under the supervision of T. Stieglitz. I. Mader, J. Kaminsky and M. Gierthmuehlen performed the surgery on the minipigs. O. Iljina, X. Wang and T. Ball wrote the paper with the help of C.A. Gkogkidis and L.D.J. Fiederer. X. Wang and L.D.J. Fiederer created the visualizations. X. Wang created the figures.

CORTICAL-DEPTH AND FREQUENCY-BAND DEPENDENCY OF THE SPATIAL EXTENT OF THE GENERATORS UNDERLYING µECOG RECORDINGS Xi Wang, Lukas D.J. Fiederer, C. Alexis Gkogkidis, Christian Henle, Irina Mader, Jan Kaminsky, Mortimer Gierthmuehlen, Thomas Stieglitz, Tonio Ball

T. Ball, X. Wang and L.D.J. Fiederer devised the study. C.A. Gkogkidis and X. Wang performed the experiments. C. Henle created the µECoG arrays under the supervision of T. Stieglitz. I. Mader, J. Kaminsky and M. Gierthmuehlen performed the surgery on the minipigs. X. Wang analyzed the data. X. Wang ran the simulations under the supervision of L.D.J. Fiederer. L.D.J. Fiederer programed and performed the segmentations and created the models. X. Wang wrote the manuscript with the help of L.D.J. Fiederer and T. Ball. X. Wang and L.D.J. Fiederer created the visualizations. X. Wang created the figures.

5 Abstract

ABSTRACT Nearly a century has elapsed since the discovery of the electroencephalogram (EEG) by Hans Berger. In this time, electroencephalography (EEG) has installed itself as clinical standard for the diagnosis of multiple neurological disorders as well as a relatively cheap and versatile research tool. Over the last 90 years, EEG has continuously been improved by researches, clinicians and engineers. One notable improvement, on which this thesis builds, is the use of an ever-increasing number of spatial measurement points (electrodes). EEG started being measured with only a handful of electrodes targeting specific brain locations. Clinicians expanded the EEG montage to ~20 electrodes covering the whole scalp and defined nomenclatures and positions in the 10-20 system. While the 10-20 system is still clinical standard, in research this system has been extended to ~74 electrodes (10-10 system) and >300 electrodes (10-5 system). The push for more electrodes has been driven by the advent of modern day computing making it possible to digitize and store large amounts of data, combined with new mathematical methods capable of disentangling the sources of the ongoing brain activity and localizing them in the brain, namely source localization. Increasing computational power made it progressively possible to refine the level of detail included in the head models and the forward solutions underlying the source localization (also called inverse solution). Including these details made the forward solutions, and thus the inverse solutions too, more accurate and create the possibility of investigating the effect of individual anatomy on electromagnetic volume conduction. Despite the progress of EEG and source localization, EEG still not close to the gold standard for measuring and localizing neuronal activity in the brain. The gold standard is still to measure the activity directly on the brain surface or even directly within the brain. These methods are commonly called invasive or intracranial EEG (both iEEG). Because of the superior spatial and signal resolution of iEEG, these methods have seen relatively few attempts at improvements. Thus, the aim of this thesis was to further improve the spatial and signal resolution EEG to drive it as close as possible to iEEG, and to improve iEEG itself. The improvements achieved during this thesis are reflected both in the write-up of the thesis as well as in publications and manuscripts available online. Some of these were not included in the write-up in an attempt to keep the scope focused on one topic, namely the spatial simulation of EEG and iEEG. For methodological and experimental improvements the interested reader is referred to my scholar profile. In the following the results included in the write-up are briefly presented. In chapter V, a shortcoming of electrocorticography (ECoG), an iEEG method, is addressed. In EEG, the description of the spatial, temporal and frequency characteristics of the electric activity of muscles (electromyogram, electromyography, EMG) contaminating EEG has been a necessity to separate it from brain activity. That ECoG, and iEEG in general, is also contaminated by EMG, albeit to a lesser extent than EEG, has been largely disregarded. Thus, a first step to improve iEEG is to create guidelines to identify EMG and differentiate it from brain activity. One major source of EMG artifacts in iEEG recordings is the chewing musculature. In chapter V we therefore describe the spatial, temporal and frequency characteristics of chewing-related (ChR) EMG signals recorded with ECoG. We show that ChR EMG can be differentiated from brain activity based on spatial, temporal and frequency characteristics. Its spatially wide distribution, which bridges anatomical borders, is in stark contrast to the mostly focal spatial extend of brain activity, which is also restricted to cortical structures with relatively sharp borders. The repetitive temporal characteristic of ChR EMG can also be used to differentiate it from brain activity, which is usually either transient or sustained. Lastly, the frequency profile of ChR EMG has a much more broadband energy distribution than brain activity, which is distributed across different narrow frequency bands, some of which behave in an anti-correlated fashion. The insights won in chapter V can now

6 Abstract

be applied to either discard iEEG data contaminated by EMG or to design methods to attempt to clean the contaminated data, improving the quality and yield of iEEG. In chapter VI, we present results based on the data described in chapter V extended using volume conductor simulations. The results suggest that the cortical electric fields occurring during strong mastication could be steep enough to entrain neuronal activity. This claim is based on multiple recent findings showing that neurons passively influence the membrane potential of their neighbors because of the changing ionic concentration of the extracellular medium during ongoing activity. This influence creates synchronous fluctuations of the membrane potential of local neuronal populations, making the whole population synchronously more, or less, receptive to incoming information. We argue that this bidirectional mechanism is potentially active in a unidirectional fashion, from ChR muscles to cortical neurons, during strong mastication. Based on the current literature we discuss the possibility that this interaction might influence our cognition on both short-term and long- term timescales. In chapter VII, we improve our understanding of the influence of mm-scale anatomical detail on EEG volume conduction using cutting edge imaging data and simulations. Although anatomy is increasingly being taken into account when simulating volume conduction, many details are being disregarded under the assumption that their influence on volume conduction is negligible. We show that blood vessels, which are quite difficult to model and have thus up to now always been disregarded, do have a non-negligible influence of volume conduction. Source localization errors > 20 mm could be observed in the vicinity of the major cerebral arteries. While minor arteries did not skew source localization considerably (< 5 mm), in densely vascularized brain areas, like the insula and the medial temporal lobe, spatial accumulation of minor arteries also led to source localization errors > 20 mm. These brain areas are targets of clinical source localization and play important roles for the normal function of the brain. Clinical EEG source localization could thus be further improved if more attention would be payed to small details. Disregarding seemingly negligible details could potentially bear large consequences and should thus be investigated as soon as methodologically possible. While the accuracy of most models is sufficient for current research questions, emerging topics like layer specificity of neuronal activity do need a sub- mm localization accuracy. In chapter VIII, we show that increasing the spatial sampling of ECoG leads to significant improvements in the spatio-frequential content of recordings. In last couple of decades, spatial sampling of EEG has been drastically increased. The spatial sampling of iEEG however, because of clinical certification constraints, has stayed constant. The resent push in research to improve ECoG has triggered the development of µECoG where contact size and intervals are in the µm range. In chapter VIII we not only show that µECoG has a richer spatio- frequential content than standard ECoG, but also that the spatio-frequential content further increases the smaller the µECoG grids get. Combined with recent findings that neuronal spikes can be recorded from the cortical surface, our findings put in question the traditional assumptions regarding the limitations of the optimal spatial sampling of EEG and iEEG recordings. In chapter IX, we present results based on the data described in chapter VIII, extended using volume conductor simulations, showing that µECoG recordings have the potential to be used to perform cortical layer specific source localization of neuronal activity. The laminar or laminated recordings currently used to investigate the laminar specificity of neuronal activity are a tedious and technically challenging procedure. Moreover, such recordings are, in the clinical context of medically intractable , which is the main source of iEEG data in humans, ethically questionable and bear strong regulatory constraints. The medical and scientific prospect of having access to this information using superficial recordings performed

7 Abstract

in the clinical context is thus huge. Our results show that it is possible to generate a frequency-resolved depth mapping of µECoG activity matching the current penetrating electrode literature using simple methods based on spatial properties. This represents the first step towards layer specific source localization of neuronal activity and leads the way for further analysis using more sophisticated methods. Concluding, this thesis shows that volume conductor simulations of electromagnetic potentials are a potent tool for improving EEG and exploring new possibilities. Simulations can be used to localize the sources of the measured activity, to infer data at point where no measurement took place, to investigate how the measured data is influenced by the underlying anatomy and to test new ideas and hypotheses. Although the methodological and experimental aspects are crucial for improving EEG, simulations also opened the doors for further improvements. Some of which, as described in this thesis, could probably not have been envisioned without the use of simulations.

8 Introduction

I. INTRODUCTION A. Electrophysiology Electrophysiology is a broad research field where the electrical activity elicited by live tissue is measured and recorded to investigate the underlying functions. One of the hallmarks of electrophysiology is its close to unlimited temporal resolution, which is only restricted by the processes it measures. Electrophysiological measurements span multiple orders of spatial magnitude, with measurement methods most appropriate for each spatial scale. Figure 1 depicts these scales, with their measurement techniques, neuronal coverage, spatial resolution and invasiveness. This thesis will focus on the largest spatial scales used to measure electrical brain activity, electroencephalography (EEG) and electrocorticography (ECoG). For completeness sake, we will briefly introduce the major electrophysiological scales and methods.

Figure 1: An overview of the measurement scales of the electrical activity of the brain. Pyramidal cells are depicted in orange with penetrating recording electrodes represented as graded gray-black cones. Surface electrodes are represented by a black bar. Pyramidal cells are located in the neocortex, which is therefore also represented in orange. Courtesy of Tonio Ball, modified from [1]

1) Electroencephalography EEG was first described by Hans Berger in 1929 [2]. Figure 2 shows one of the first recordings of EEG. A golden age of EEG ensued during which EEG instated itself as the clinical standard for measurements of brain activity. But the advent of (MEG) and functional magnetic resonance imaging (fMRI), which both have higher spatial resolutions, in the late 1980ies diverted some of the attention from EEG [3]. None the less, because of its simplicity and relative cheapness, EEG remained the main workhorse in the clinical practice. In the late 1990ies and early 2000s the introduction of brain-computer interfaces (BCIs) again renewed the interest in EEG. Now, close to 100 years after the discovery of EEG, the age of digitalization, miniaturization and mass marketing,

9 Introduction

combined with the advent of dry EEG, has made EEG available to the general public, dramatically increasing its impact on society. EEG measures the potential difference at multiple points of the scalp, relative to a reference using circular electrodes. The electric fields generating these potentials are generated by the apical dendrite input to neocortical pyramidal cells [4]. As these fields have dipolar properties [4], their amplitude is diminished by the square of the distance they have traveled from source to electrode [4]. This attenuation is further increased by the low electrical conductivity of the skull [4]. Because of this drastic attenuation EEG can only measure brain activity with large amplitudes. Such brain activity is usually generated by a patch of neocortex with an area of multiple cm2 [5]. But a large spatial extent of synchronous activity is not a guarantee for measurable EEG. During both volume conduction and referential recording, the activity can be attenuated, distorted or cancelled out by electric fields of neuronal or artifactual origin. Taken together, these factors make it difficult to measure and interpret EEG. For the initial identification and isolation of true EEG patterns, concurrent intracranial measurements are a potent safeguard against misinterpretations (see Figure 3 for an example).

Figure 2: EEG trace published by Hans Berger in 1929. Upper trace, EEG recorded on the scalp surface of a 15 years old boy with clearly visible alpha rhythm (10 Hz). Lower trace, 100 ms (10 Hz) oscillation used for time reference. The EEG was measured between the forehead and the occipitus using lead strip electrodes and a double coil galvanometer. Reproduced from [2].

2) Electrocorticography & stereotactic electroencephalography Intracranial measurements in live humans can be roughly split into two categories. The electrocorticogram (ECoG), which measures brain activity at the pial surface of the neocortex, and the stereotactic EEG (SEEG), which measures brain activity below the pial surface. Both methods are routinely used for pre-neurosurgical diagnosis in medically intractable epilepsy. ECoG electrodes are usually arranged in regularly spaced, silicone embedded, stripes or grids with geometries ranging from 1x4 to 16x16 electrodes. ECoG grids provide a fair overview of the distribution of electrical activity at easily accessed surfaces of the brain, e.g. prefrontal, sensorimotor, parietal and temporal cortices. ECoG stripes can be effectively used to measure electrical activity from surfaces of the brain too curved or too hard to access directly, e.g. ventral frontal, temporal and occipital areas and poles and the interhemispheric fissure. SEEG electrodes are usually arranged in regularly spaced rods. These rods are inserted into the brain to record from deeper brain areas, e.g. hippocampus, insula and white matter. As EEG, ECoG and SEEG measure the summed activity of synchronously active neuronal populations. Compared to EEG, ECoG and SEEG are much closer to the signal sources and thus can measure from much smaller populations. Moreover, being closer to the signal sources, the signal measured by ECoG and SEEG electrodes are less dominated by the dipolar component and have contributions from tri- and quadru-polar components, making the measured signals more spatially focal. Lastly, ECoG and SEEG electrodes are further away from sources of artifactual electrical activity than EEG and benefit from the shielding effect of the skull, which further increases the signal-to-noise ratio (SNR). Figure 3 illustrates this effect in simultaneous EEG and ECoG recordings during head movements. Despite these tremendous advantages, intracranial measurements typically sample only a small portion of the brain, the one which is relevant for medical diagnosis. Large coverage, even if relevant for diagnosis, is

10 Introduction

usually avoided because of the high risk of complications [6]. Forming a whole-brain picture of a specific task or process is thus a very time-consuming work where patients with different implantation coverages need to be combined. As EEG excels at whole-brain coverage, concurrent intra- and extra-cranial measurements add precious information for both modalities.

100 µV

Figure 3: Concurrent EEG and ECoG recordings during head movements. Head movement artifacts are clearly visible in the EEG. Apart from the spike at ~5 s, no artifacts are visible in the ECoG. But, as we see in chapter V, the artifacts are still present, albeit strongly attenuated and not visible in the unprocessed data. Upper 4 traces, EEG. Lower 6 traces, ECoG. Head movements from ~1.3 s to ~7.5s. EEG and ECoG at same scale. Courtesy of Tonio Ball, modified from [7].

3) Local field potentials Compared to EEG, the electric field induced potentials measured by ECoG and SEEG are much more local. Thus one could say that EEG measures global field potentials and ECoG and SEEG local field potentials (LFP). Generally speaking, the LFP is considered to be the neuronal activity excluding action potentials and is traditionally extracted by low-pass filtering the signal below 400 Hz. The LFP is dominated by the fluctuation of the neuronal membrane potential, mainly driven by dendritic inputs. In addition to ECoG and SEEG electrodes, human LFP can be measured using micro-wire electrode bundles and needle-like metal or glass electrodes, which are stuck into the brain. These electrodes can also be used to concurrently measure neuronal spiking activity by high-pass filtering the signal above 400 Hz.

11 Introduction

4) Multi-unit & single-unit activity When measured with such electrodes, neuronal spiking activity is usually contributed to by multiple neurons and is therefore labeled multi-unit activity (MUA). Thus, the main challenge while measuring MUA is to attribute each spike to the corresponding neuronal unit. This process is called spike sorting and can, since the digitalization of recordings, be done, with some caution, automatically. Sorted MUA is often referred to as single-unit activity (SUA). Definitive SUA, albeit not in live humans, can be measured by using the patch-clamp method in whole-cell configuration, where by default only one cell is recorded from. Patch-clamping can also be used to measure even more detailed neuronal activity from single dendrites or single ion channels. Correlating the neuronal activity measured at each of the described spatial scales, from single ion channels to scalp potentials, is an essential step in our understanding of the function of the . In this thesis, EEG, ECoG and µECoG are used to measure neuronal and muscular electrical activity. This activity is decomposed into its time and frequency components and compared to simulated activity. The simulations are based on anatomical models segmented from magnetic resonance imaging data. B. Magnetic Resonance Imaging Magnetic resonance imaging (MRI) is an imaging method used to perform tomographies of soft tissues. In contrast to x-ray computed tomography (CT), MRI is particularly well suited for this task, even without the use of contrast enhancing agents. MRI has furthermore the advantage of relying on strong magnetic fields and radio-frequency pulses (MHz range) to create images, instead of the focused ion beams used in CT. The latter having been shown to ionize tissues and cause cell degeneration after repeated or prolonged exposure. Nonetheless, radio-frequency pulses heat up tissues, limiting the time tissues can be safely scanned using MRI. Long-term effects of the strong magnetic fields are yet to be described. Generally, MRI is considered to be much safer than CT. MRI has been intensively developed in the early 70ies and is the imaging version of nuclear magnetic resonance (NMR [8]–[10]). The groundwork, which made MRI possible was laid down in the early 50s and 60s by many scientists and pioneered by Erwin Hahn [11] (discovery of the spin echoes) and Herman Carr [12] (one-dimensional NMR). The first 2D image produced using NMR and thus the first MRI scan was published by Paul Lauterbur in 1973 [13], quickly followed by the first in vivo image in 1974 [14] (mouse thorax). The mid and late seventies saw multiple publications featuring first human in vivo images with varying fields of view and resolutions [15]–[17]. However, these images were still being reconstructed by acquiring multiple projections of the imaged object, like in CT, which in MRI is excruciatingly slow and prone to reconstruction artifacts. The introduction of frequency and phase encoding gradients by Paul Lauterbur made it possible to generate tomographies without having to rely on projections. The realization by Richard R. Ernst that the could be used to very quickly reconstruct images further improved and accelerated the generation of MRI images. Once echo planar imaging[18], which significantly speeds up MRI image acquisition, was broadly adopted, MRI developed its full impact on medical imaging. MRI uses magnetic fields and more importantly magnetic gradients combined with radio- frequency pulses to generate tomographies of water proton spin densities. A strong static magnetic field, B0, ranging from 1.5 T to 9.4 T when investigating humans, is used to force all protons within the imaged tissue to align their individual spin along the axis of B0. B0 also modifies the rotation frequency of the spins. This rotation frequency is called the Larmor frequency [19], ω, and is characteristic for every nucleus and is directly proportional to the strength of B0. Radio-frequency pulses with the same frequency as the Larmor frequency of a

12 Introduction

specific nuclei’s protons, in the case of MRI water molecule protons, can be used to selectively transfer energy to the given protons. The heightened energy state of the protons allows them to change their alignment to the static magnetic field. Concurrently the radio- frequency pulse aligns the rotation phase of the targeted protons. These two effects make it possible to record a weak, magnetically- induced compound signal in the direct vicinity of the stimulated, synchronized, protons. Depending on the angle of the change in alignment to B0, ≤ 90° or 180°, the signal respectively decreases or increases over time with a tissue specific time constant, τ. In MRI, the tissue specific time constants are generally known as T1 and T2 relaxation times, respectively. T1 describes the decrease of the magnetization signal transverse to B0, depends strongly on the interaction of the excited proton spins with the environment, the lattice (spin-lattice interaction), and is a relatively long time constant. T2 describes the increase of the magnetization signal longitudinal to B0. It depends strongly on the interaction between proton spins (spin-spin interaction) and is a relatively short time constant. These time constants are used to weight the contrast between tissues, generating T1 or T2 weighted images, depending for which time constant the contrast was optimized. To determine the spatial origin of the measured signal, three magnetic gradients are used. They are used to create a 3D spatial encoding. The first gradient, the slice selection gradient B1, is overlaid parallel to B0 during the application of the radio-frequency pulse to create a spatial gradient of Larmor frequencies. The steepness of B1, combined with the frequency bandwidth of the exciting radio-frequency pulse, dictates the thickness of the slab of excited protons and thus the thickness of the MRI image. The second gradient, the frequency encoding gradient, is used to impose spatially defined Larmor frequencies onto the excited slab orthogonally to B0 and B1 during the readout of the magnetization signal. Each frequency component in the readout signal can thus be unambiguously assigned to a slice of the excited slab. The size of each frequency encoded slice is inversely proportional to the steepness of the frequency encoding gradient and to the number of time samples recorded during signal acquisition. The third gradient, the phase encoding gradient, is applied very shortly to the excited slab just before each signal acquisition. The phase encoding gradient changes the Larmor frequency of the excited protons long enough to induce a spatially defined shift in the rotation phase. The signal arising from each location within the excited slab will thus have a different phase according to the position along the phase encoding gradient. In contrast to the frequency encoding, this information is not enough for unambiguous assignment of location. The phase encoding gradient needs to be applied before a multitude of acquisitions to create a phase- difference encoding. By comparing the phase shift between consecutive acquisitions, the signal can be unambiguously assigned to a slice of the excited slab. The size of each phase- difference encoded slice is inversely proportional to the steepness of the phase encoding gradient and to the number of repeated acquisitions. As the phase encoding gradient is applied orthogonally to B0, B1 and the frequency encoding gradient, the slices of frequency and phase-difference encoding are orthogonal and form pixels (picture elements) of the MRI image. Adding the thickness of the excited slab thus forms the voxels (volume elements) of the MRI volume. This time sequence of magnetic gradients and radio-frequency pluses is called an imaging sequence. The imaging sequence described above corresponds to the echo planar imaging sequence. Other imaging sequences have different pulse and gradient applications but the spatial encoding concept using 3 gradients is universal to all MRI sequences In this thesis MRI is used to acquire tomographic volumes of the head. Both T1 and T2 weighted volumes are used to construct detailed volume models of the head. These models are then used to simulate the spatial spread of electricity in the volume in dependence of the modeled anatomical detail. Such models are called volume conductor head models.

13 Introduction

C. Volume Conductor Head Modeling Head models are important tools in basic and applied . They are routinely used for source localization [20]–[24], model-guided surgery [25], [26], research on traumatology [27]–[29], modeling transcranial magnetic/direct current stimulation [30], [31], electrical impedance tomography (EIT) [32]–[34] and other fields of magneto- & electro- encephalography (MEG, EEG) research. The latter comprises investigations such as the influence of anatomy on field propagation [30], [35], [36] and the optimal spatial sampling of EEG signals [37]–[39]. The present thesis focuses on head models for EEG research, which are used as volume conductor head models (VCHM) for computing the propagation of electromagnetic fields created by electric sources. For such applications, model accuracy is an important topic [40]–[43]. The overarching aim to optimize model accuracy has shaped the historical evolution of VCHMs, which is grossly displayed in Figure 4. Nowadays, VCHMs are typically created by segmentation of anatomical data, mainly from MRI, but also obtained from CT, or even from cryogenic anatomical sections [44], [45]. Before anatomical imaging data became available, VCHMs were constructed as infinite media [46] or as concentric [47]–[50] or eccentric [51] spherical shells. The simple mathematical properties of shell models allow them to be solved analytically and thus established them, until today, as validation standard for numerical methods. The first non-spherical VCHM was manually constructed from a stereotaxic cat brain atlas and solved numerically using the finite difference method (FDM) [52]. Later, Meijs et al. published the first 3D MRI-based four compartment realistic human head VCHM [53] and solved it using the boundary element method (BEM). The BEM is computationally very efficient but considers only the boundaries between concentric closed compartments and has an upper limit on the number of nodes from which the model can be built [54]. The full volume of a model can be modeled with the FDM, which is straightforward to implement. Because of the ever increasing complexity of subsequent models a third method, the finite element method (FEM) was introduced [55], [56], followed by the finite volume method (FVM) [57]. The FDM, FEM and FVM made the integration of complex geometries and anisotropies possible, where particularly FEM excels but has higher computational costs.

Figure 4: Evolution from simple spherical models over simple surface models to realistically-shaped volume models. Surface model courtesy of Matthias Dümpelmann. Volume model courtesy of \Carsten Wolters. These methods, combined with the commercial availability of 1.5T MRI scanners made it possible to push the level of detail included in the VCHMs far beyond that of previous models, thus drastically reducing model errors compared to spherical models. Most notably, model errors produced by the influence of incorrect modeling of skull [42], [42], [43], [58]– [65], anisotropy [59], [62], [66]–[69], the cerebro-spinal fluid (CSF) [38], [58], [65], [70], [71] and model volume [41], [43], [72] have been investigated. Most recent FEM VCHM are

14 Introduction

now based on 3T MRI with an isotropic resolution of 1 mm [43], [73], [74] and have at least isotropic white and gray matter, CSF, skull and skin compartments implemented. Table 1 gives an extensive overview of some recently published VCHMs. With the advent of 7T MRI, sub-millimeter resolution scans with excellent contrast-to-noise ratio (CNR) have become available [75]. Relative to the 1-mm data, more anatomical detail becomes visible, but the number of voxels in the image is increased several fold. This represents a modeling challenge and it is currently not clear whether sub-millimeter forward and inverse FEM modeling is feasible with the available segmentation and solver software tools. What kinds of additional anatomical information can be included in such models and what the consequences for simulations of electrical field propagation such as of the EEG are is also unknown. In this thesis, VCHMs are used to model the signal sources of EEG, ECoG and µECoG measurements. These sources are simulated as dipoles and either represent neuronal populations or muscle fiber populations. The results of the simulations are used to draw conclusions for real applications of EEG, ECoG and µECoG. A prime target for such simulations are somatosensory evoked potentials. Table 1: List of tissues with distinct conductivities implemented in selected VCHMs. Tissues / Models [76] [77] [68] [67] [44] [30] [78] [42] [30] [79] [80] [81] [43] [45] [82] Gray matter V V V V V V V V V V V V V V V Anisotropic gray V V V V matter White matter V V V V V V V V V V V Anisotropic white V V V V V matter Brain General V Spinal cord + V V cerebellum Pia matter V CSF V V V V V V V V V V V V Dura Mater V V Inner bone V V V V V V V V V V V V V V V Soft bone V V V V V V V V V Outer bone V Anisotropic skull V V Teeth V Muscle V V V V V Fat V V V V V V Soft tissue V V V V Eyes V V V V V V V V Skin V V V V V V V V V V V V V V Internal air V V V V V V Major blood vessels V V V V V Detailed blood V vessels 1.5T 3T CT, 3T Cryo- 3&7T 3T 3T 3T 3T 3T 3T Cryo- 7T Imaging MRI MRI MRI MRI MRI section MRI MRI MRI MRI MRI MRI MRI section MRI Method FEM FEM FEM FEM FDM FEM FEM FEM FEM FEM FEM FEM FEM FDM FEM

15 Introduction

D. Somatosensory Evoked Potentials Somatosensory evoked potentials (SEPs) are stereotypical, phase-locked, neuronal potentials measurable after peripheral nerve stimulation. SEPs are an established clinical method and mainly used to diagnose nerve generation diseases [83]. SEPs are, due to their stereotypical nature, prime targets for validating source localization methods. Weak electric pulses are predominantly used as stimulus as these are easy to generate, apply and synchronize with measurements of brain activity. The prime target for electrical SEP measurements in humans is the median nerve, which is stimulated on the inside of the wrist, just before entering the carpal tunnel on its way to innerve the muscles of the thenar eminence (thumb), the first (index) and second (major) lumbricals and the palmar surface of the lateral three and half digits as well as their fingertips. In the late 30ies and early 40ies, multiple groups laid out the groundworks of SEPs by electrically stimulating sensory nerves and invasively recording brain responses in lower mammals [84]–[87]. In 1947, George D. Dawson first described electrically evoked SEPs recorded non-invasively in healthy subjects and myoclonic patients [88], [89]. Dawson’s seminal work was complemented by Lars-Erik Larsson [90] and deepened by Truett Allison in 1962 [91]. From the late 70ies to the late 80ies multiple research groups contributed to the elucidation of the cortical origin of SEPs. Two prominent models were investigated. The single tangential dipole in area 3b model [92]–[94] and the paired radial dipoles in area 1 and 4 model [95]. Through accumulated evidence from cortical surface and intracerebral recordings, neuromagnetic recordings and lesion studies in humans and monkeys, a mixture of both models was proposed as final model. This model is composed of a tangential dipole in area 3b and a radial dipole in area 1 [96]. Upon cathodal electrical stimulation of a nerve, the nerve fibers are depolarized close to the cathode (negative pole, injecting anions), leading to axonal action potentials (APs). In theory, the APs propagate along the nerve away from the anode (positive pole, up-tacking anions), as the nerve fibers below the anode are hyperpolarized. In practice, I have not observed this effect, which is corroborated in literature [97], [98]. The APs then propagate along the nerve and follow the dorsal column-medial lemniscal pathway to the primary somatosensory cortex. Specifically, the first-order neurons synapse on the second-order neurons in the dorsal column nuclei in the medulla oblongata. The second-order neurons cross to the contralateral side of the brain in the decussation of the medial lemniscus and synapse on the third-order neurons in the ventral posterior lateral nucleus of the thalamus. The third-order neurons relay the stimulation to the somatosensory cortex. The thalamic afferents terminate upon pyramidal cells in layer III-IV of Brodmann area 3b of the primary sensory cortex (S1) [99]. In this thesis SEPs are elicited by electrically stimulating the nose of anesthetized Göttingen minipigs and recorded using µECoG electrode arrays placed epicortically on the acutely prepared sensory cortex. The spatio-frequential characteristics of the SEPs are investigated and a cortical-depth resolved source analysis is performed. E. Object of this Thesis The overarching aim of this thesis was to push the limits of invasive and non-invasive EEG. Not only beyond the current state-of-the-art, but also beyond instated belief. This write-up will mainly focus on the simulation aspect of the thesis while the methodological and experimental aspects have been / are being / will be published separately [100]–[111]. Four independent research questions have been addressed here: It is a common misconception in neuroscience that intracranial recordings are exempt of bioelectric artifacts. To emphasize the presence of electromyographic (EMG) artifacts in

16 Introduction

ECoG recordings and to delineate this artefactual activity from brain activity, we investigating the spatio-frequential characteristics of chewing-related intracranial EMG signals in chapter V. That the brain solely communicates with the world via the five senses is a longstanding assumption in neuroscience. In the light of recent studies showing that neighboring neurons influence each other’s activity via volume conduction coupling of their membrane potential, we challenge this assumption and investigate the possibility that such coupling and thus communication not only exists at local, brain internal, scales between neurons but also at much larger, brain external, scales between chewing muscles and neuronal populations via volume conduction of chewing-related intracranial EMG electric field gradients. This work is covered in chapter VI. The major bulk of studies modeling volume conduction use extremely simplified models of the human head for forward and inverse modeling. This is based on the assumption that only gross anatomy is having a relevant influence on volume conduction. We challenge this assumption and investigating the possibility that millimeter-scale detail, such as blood vessels, may also influence non-invasive EEG volume conduction. This work is covered in chapter VII. Conventional epicortical recordings are performed with electrode arrays where contacts are ~4 mm wide and 1 cm apart. Recently however, a growing interest for higher spatial sampling can be observed. In chapters VIII & IX we therefore investigate the spatial, temporal, frequency and source characteristics of neuronal activity measured epicortically at millimeter and sub-millimeter scales during SEP stimulations of anesthetized Göttingen Minipigs.

17 Summary of the Results

II. SUMMARY OF THE RESULTS A. Spatial and Frequency-domain Characteristics of Intracranially- measured EMG In chapter V we show that EMG signals during chewing contaminates ongoing ECoG recordings. These artifacts spread over a larger area of the cortex and do not follow the contours set by the brain anatomy as brain activity would. Furthermore, chewing-related EMG has a much higher signal power in high frequencies than physiological brain activity. Lastly, chewing-related effects were about 10 times larger in the simultaneous non-invasive EEG comparted to ECoG. These three factors may be used to delineate EMG induced effects from brain activity. Our results additionally suggest that intracranially-measured EMG is attenuated by the ECoG silicone sheet, which is supported by recent findings by [112]–[114]. Moreover, our results indicate that chewing-related EMG might mainly propagates to ECoG recordings directly through the highly resistive skull. This would be in stark contrast to the standing assumption that intracranial EMG mainly propagates to the brain using low resistive ‘shortcuts’ like the burr holes or saw lines introduced by the ECoG surgery [115], [116]. These two preliminary observations are further investigated in chapter VI. B. Endogenous Brain Stimulation through EMG In chapter VI we show that the temporal pole is expected to be exposed to strong spatial electric gradients during chewing. During simulated vigorous chewing these gradients exceed the empirical brain stimulation threshold of 0.2 V/m. We thus postulate that vigorous chewing endogenously stimulates the temporal pole and might be partly responsible for the repeatedly observed effects of gum chewing on cognition. Furthermore, we confirm the first observation of chapter V that the ECoG silicone grid shields recordings from chewing-related EMG. Lastly, our simulations confirm the second observation of chapter V that chewing-related EMG mainly propagates to ECoG recordings directly through the skull. These results are a major challenge to the reports describing how intracranially-measured EMG mainly propagate to ECoG recordings via burr holes and saw line [115], [116]. C. Volume Conduction Modeling of Blood Vessels In chapter VII we show that millimeter-scale anatomical details like blood vessels do non- negligibly influence non-invasive EEG simulations. The size of the effect is strongly correlated with the local blood vessel density. Furthermore, we replicate previous findings that not modeling either the CSF or the dura lead to strongly distorted results. Lastly, we show the feasibility of using established methods for segmenting 7T MRI data, creating a sub- millimeter resolved VCHM and performing forward and inverse modeling using multi-million dipole and extended source models. D. Modeling µECoG Neuronal Sources (based on 2 papers) In chapter VIII we show that epicortically recorded SEP signals have higher spatial frequencies than current standard ECoG arrays can sample. We also show that additional information is progressively revealed when diminishing the electrode array scales. Furthermore, we show that blood vessels reduce the power of the signal measured at overlaying contacts, but only in the beta band. In chapter IX we show that it is possible to perform a frequency resolved source analysis with cortical-depth precision using the spatial profile of µECoG signals. Using this source analysis, we further show that the spatial reach of LFP, as measured with µECoG in anesthetized Göttingen minipigs during SEP stimulations, is in the order of 2 mm for superficial sources and 3.5 mm for deep sources.

18 Discussion

III. DISCUSSION This thesis’ main contribution to neuroscience was the development of advanced volume conduction models and their application to controversial and avant-garde topics. We have shown that, against popular belief, extracranial EMG artifacts do indeed influence intracranial recordings and that skull defects only play a marginal role in the propagation of extracranial EMG artifacts to intracranial recordings (cf. chapter V). In a further step, we show that the cortical electric fields generated by the EMG have the proper time, amplitude and frequency properties to influence ongoing neuronal activity and, for the first time, introduce the notion that it is possible to stimulate one’s own brain using endogenous EMG (cf. chapter VI). Similarly, we have shown that current computational structures and imaging techniques allow for much more detailed volume conduction modelling that used in clinical and scientific routine. Not only is including sub-millimeter detail in head models feasible but, as we have shown in the case of blood vessels (cf. chapter VII), it also provides important improvements of the modelling accuracy. Lastly we have shown that the spatial structure of somatosensory- evoked potentials (SEPs) is much more detailed than previously thought (cf. chapter VIII), that the spatial reach of LFP can be effectively modelled using dipoles and that the frequency- resolved sources of the SEPs can, unexpectedly simply, be localized with anatomically plausible depth precision using epicortical recordings and a sub-millimeter-resolved head model (cf. chapter IX). In the following we will discuss the perspectives which advanced volume conduction modelling make possible, especially with respect to real-world applications. A. Model-Guided Improvements of Intracranial Electrophysiological Measurements The detailed modeling of the human head is particularly interesting and challenging when atypical structures, like clinical electrodes, have to be included. In addition to being required for accurate simulation results [112]–[114], modeling the real extent and properties of the electrode support structures opens up new possibilities. For example, in chapters V and VI we have shown that both ECoG data and realistic simulations support the concept that the silicone sheet, in which the ECoG electrodes are embedded, attenuates the signal of extracranial electric sources. Thus, the silicone sheet actively contributes to the signal quality of ECoG. A first perspective from this finding would be to investigate if extending the edge of the silicone sheet beyond the current ~5 mm could further increase its shielding effect and consequently further increase the robustness of ECoG recordings against, e.g., EMG artifacts. First investigations could be performed in a detailed head model to determine the sheet- extension-to-signal-improvement relation. Before designing and in vivo testing the most promising sheet extension, the balance between signal improvement and potentially increased complication risks will have to be considered. Indeed, multiple studies have identified a number of parameters, including ECoG electrode grid parameters, significantly increasing the occurrence of clinical complications [6], [117]– [119]. Hamer and colleagues report, based on a sample of 198 implantations over 18 years, that following parameters significantly increase the occurrence of clinical complications: the number of implanted grids, implanting more than a total of 60 electrodes, the presence of additional burr holes, implantations lasting longer than 10 days, left hemisphere implantations and increased age [6]. The number of , patient IQ, anticonvulsants treatment or grid localization were not found to significantly increase the occurrence of clinical complications [6]. Wong and colleagues report, based on a sample of 71 implantations over 17 years, that following parameters significantly increase the occurrence of clinical complications: larger ECoG grids, increased total number of electrodes, increased density of electrodes, right central implantations and left central implantations [117]. Gender, age, epilepsy type, and

19 Discussion

implantation duration were not found to significantly increase the occurrence of clinical complications [117]. Önal and colleagues report, based on a sample of 35 implantations over 6 years, that following parameters significantly increase the occurrence of clinical complications: larger ECoG grids and increased total number of electrodes [118]. Parameters which were not found to significantly increase the occurrence of clinical complications were not reported. Wiggins and colleagues report, based on a sample of 38 implantations over 4.5 years, that following parameters significantly increase the occurrence of clinical complications: implanting more than a total of 100 electrodes, having more than 10 cables, having more than one cable outlet and implantations lasting longer than 14 days [119]. Gender, age, duration of antibiotics treatment and side of implantation were not found to significantly increase the occurrence of clinical complications [119]. Summarizing, following parameters were repeatedly found to significantly increase the occurrence of clinical complications: larger ECoG grids, increased total number of electrodes, longer implantation durations and left hemisphere implantation. It thus appears that larger grids significantly increase the occurrence of clinical complications [117], [118]. One would thus be tempted to conclude that increasing the extend of the ECoG grid’s silicone edges, consequently increasing the ECoG grid’s size, to improve the shielding properties of ECoG grids, will increase the occurrence of clinical complications. When looking closely at the definition of ECoG grid size it becomes apparent that it actually refers to the number of electrodes contained in the ECoG grid [117], [118]. Furthermore, Wong and colleagues argue that the increased stiffness and thus increase pressure on vessels is probably the main factor behind the significantly increased occurrence of clinical complications when implanting ‘larger’ ECoG grids [117]. The results discussed in the previous paragraph can therefore not be used to draw conclusion on the influence of broader ECoG grid silicone edges on clinical complications. We even argue that, given that one important property of silicone is its great mechanic flexibility, increasing the width of the ECoG grid’s silicone edges without additional metal components will not increase the ECoG grid’s stiffness. Such a procedure should actually reduce the overall stillness of the ECoG grid. Although the above assumptions remain to be tested, we postulate that increasing the width of ECoG grid silicone edges will, (i) not increase the occurrence of clinical complications, (ii) improve the shielding effect of ECoG grids and thus (iii) increase the signal quality and (iv) improve the quality of ECoG based diagnoses. In chapters V and VI we have, against expectations, shown that saw lines and burr holes play a minor role in extra- to intracranial volume conduction of chewing-related artifacts. It is clear that, had the skull defects and ECoG grid been placed in the direct vicinity of the artifact sources, their contribution would have been far greater. Thus, in respect to chewing-related artifact contamination, the skull trepanation and ECoG grid placement were unconsciously planned close to optimally by the neurosurgeons. An interesting perspective would be to consciously take into account the effect of saw lines and burr holes on extra- to intracranial volume conduction while planning trepanations. Such model-informed surgical planning could be very useful in the cases where the neurosurgeon has a set of clinically equivalent trepanation schemes. Detailed volume conduction could inform which of the available schemes would most probably lead to the most robust electrophysiological recordings. In a similar line of thought, model-informed selection and placement of electrodes could, for clinically equivalent configurations, be used to optimize the electrophysiological quality of the recordings. Lastly, the design of the electrodes themselves could be optimized to optimize the electrophysiological quality of the recordings. This last point is a key aspect of a grant proposal we have submitted to the Bundesministerium für Bildung und Forschung in April 2017.

20 Discussion

B. Chewing-Related Endogenous Brain Stimulation: Further Steps to Prove Hypothesis In chapter VI we have shown that chewing-related (ChR) electric fields (EF) potentially reach the temporal pole with amplitudes which have been shown to be sufficient to influence ongoing brain activity [120]–[122]. These ChR EF might in part explain the puzzling effects of gum chewing on cognition (seen next section). The results presented in chapter VI need to be refined before further claims can be made. Building on the simulated results a model-based selection of patients with implantation schemes best fitting the temporal pole hypothesis can now be performed. Best fitting implantation schemes would encompass EEG electrodes, stripe electrodes, depth electrodes and micro-wire electrodes. EEG electrodes above the m. temporalis are needed to assert the strength of the ChR EMG activity. Stripe electrodes in close proximity to the temporal pole are needed to assert the epicortical strength of the ChR EF. Depth electrodes in close proximity to the temporal pole and the hippocampus are needed to assert the intracortical strength of the ChR EF and their potential influence on memory. Finally, micro-wire electrodes are needed to assert the influence of ChR EF on multi- and single-cell activity and potentially directly prove the entrainment effect of ChR EF on ongoing brain activity. After asserting the influence of ChR EF on ongoing brain activity, a possible next step could be to assess the efficacy of this stimulation relative to other electrical stimulation modalities. Currently, three modalities of electrical stimulations are commonly used. Transcranial direct current stimulation (tDCS) uses a constant electrical current to induce a shift in brain excitability [123]. Transcranial alternating current stimulation (tACS) uses an alternating electrical current to entrain, enhance and shift brain activity at the frequency of stimulation [123]. Transcranial random noise stimulation (tRNS) uses a normally distributed random current to entrain brain activity [124]. In chapter VI we have discussed that ChR EF have (i) a frequency composition very similar to that of tRNS [124], (ii) a repetition rate very similar to the oscillation frequency of slow tACS used to potentiate memory [125] and (iii) simulated epicortical amplitudes similar to those of other electrical stimulation modalities. We can thus expect that ChR EF could achieve higher efficacies than the other modalities. To test this assumption, a double blinded experimental series on a large cohort of subjects evaluating the effect of each stimulation modality on cognitive performance would need to be undertaken. Key aspect of such a series would be that the ChR EF need to be delivered without subjects actually chewing, which would unblind the experiments and introduce confounding factors due to the act of chewing. Luckily, it is possible to program brain stimulators to produce arbitrary waveforms. Thus, a preprogrammed ChR stimulus, previously recorded or purposefully synthetized, could be used. In a similar line of though, a less extensive experimental series could be performed with real chewing in healthy subjects and individuals with different degrees of craniotomies. The hypothesis of such an experiment would be that chewing induced cognitive performance changes must be more pronounced in the craniotomy cohort, with a dependency on craniotomy size and location, if ChR EF indeed influence ongoing brain activity. As straight forward as this experiment sounds multiple pitfalls need to be considered. First, craniotomies are accompanied by skin incisions and potential chewing muscle damage [126]. In acute craniotomy patients, be it for neurosurgical treatment of medically intractable epilepsy or intracranial pressure relief, we can thus assume that pain and damaged muscles will reduce the amplitudes of ChR EF [126]. It would thus be preferable to compose the craniotomy cohort of individuals with chronic craniotomies, where skin and muscles can be assumed to have full recovered. Nonetheless, the remaining inter-individual variations in ChR EF amplitudes will need to be accounted for in the evaluation. Second, environmental factors, like chewing previous to the experiments, need to be closely monitored and if possible

21 Discussion

controlled for. While food intake can be restricted before the experiments, nervous chewing and gnashing will be much more difficult to control. The effect of such unconscious muscle activity on cognitive performance, specifically frowning during concentration, is currently being investigated by Moritz Grosse-Wentrup at the Max Planck Institute for Intelligent Systems in Tübingen. First results indicate higher cognitive scores in the frowner cohort [personal communications]. The fact that others have also taken up the path of EMG mediated endogenous brain stimulation is a clear indication of its scientific and social potential. C. Chewing-Related Endogenous Brain Stimulation: Beneficial or Detrimental? Thoughts on Applications to Evolution and Food Intake In chapter VI we have discussed that ChR endogenous brain stimulation could, in part, be responsible for the puzzling effect of gum chewing on cognition. As these effects are not clear cut it is difficult to make a prognosis as to whether ChR endogenous brain stimulation could have a beneficial or a detrimental influence on cognition. Indeed, the effect of gum chewing on cognition is not clear cut either. Wilkinson and colleagues, which were the first to provide empirical evidence, show that gum chewing prior to batteries of cognitive tests significantly improves performance compared to non-chewing controls [127]. In the year following the seminal work, studies were published showing improved [128]–[134] and reduced [129], [130], [133], [135]–[138] cognitive performance. Quite a few of these studies could further not reproduce some of previously described effects, raising concerns on the robustness of chewing-induced changes in cognitive performance [135], [136], [139]–[141]. Later results showing a differential effect of gum chewing related to the time of chewing relative to cognitive testing seem to somewhat reconcile the disparate findings [142], [143]. Robust findings that gum chewing improves alertness [132], [140], [144]–[147] further support a link between chewing and brain functions. For more details on the effects of gum chewing on cognitions please refer to reviews [148]–[151]. The PhD thesis of Andrew P. Allen offers an even more exhaustive overview of the field [152]. In the following we argue how ChR EF could influence cognition in both beneficial and detrimental ways. With 0.2 V/m, the ChR EF simulated in chapter VI were also at the lower limit of the empirically defined efficacy range of brain stimulation [120]–[122]. On the one hand, although the simulation parameters were chosen very conservatively, it is still possible that ChR EF reach the brain slightly below stimulation threshold. Such a scenario could, under the assumption that ChR endogenous brain stimulation is detrimental for cognitive performance, reflect an evolutionary safety measure where ChR muscle- and skull-parameters jointly evolved to ensure that brain activity is not influenced by ChR EF. Consequently, one could envision that the discovery of fire, which is hypothesized to have triggered a rapid cognitive evolution of human kind by enabling us to cook food, thus making consumption and digestion more efficient [153], could also have released the brain from detrimental daylong ChR stimulation by powerful muscles. This release could have further accelerated human kind’s cognitive evolution. On the other hand, it could be that ChR EF reach the brain slightly above stimulation threshold. Such a scenario could, under the assumption that ChR endogenous brain stimulation is beneficial for cognitive performance, reflect an evolutionary process where ChR muscle- and skull-parameters jointly evolved to ensure that brain activity is influenced by ChR EF. Indeed, cooking food not only reduced the need for strong chewing muscles. Cooking food also reduced the need for prominent skull features previously needed to withstand the high pressure generated by a strong chewing musculature [154], [155]. Thus, it could be that the discovery of fire and the associated reduction of skull thickness first made it possible for human kind to endogenously stimulate their brains with ChR EF. Consequently, the beneficial stimulation could have further accelerated human kind’s cognitive evolution.

22 Discussion

Following the line of thought that ChR endogenous brain stimulation is beneficial for cognitive performance, questions regarding soft diets arise. It is a timely conception that soft food, e.g. junk food, are bad and hard food, e.g. vegetables and fruits, are good for our health. Such concepts could also in part arise because of the reduced ChR brain stimulation benefits of soft food. One way to segregate the non-healthy factor from ChR factor would be to asses if prolonged absence of chewing leads to reduced cognitive performance. Most non-chewing populations are usually created by medical necessities where non-healthy factor dominates. Fortunately, the smoothie diet, a trendy soft food diet, where adherents solely drink food, could offer a good testing environment where non-healthy factors can be assumed to be the same as in the normally chewing control group. Including a strongly chewing group in such a study should be fairly easy as raw diets are broadly available. The comparison of the cognitive scores of both groups would be an interesting research question. Furthermore, quantifying the amount of unconscious chewing muscle activity, e.g. nervous chewing and gnashing, in both groups could possibly give insights into compensatory mechanisms to alleviate the lack of chewing. A possible pitfall for such a study could be the potentially reduced/increased digestion load of the non-chewing and strongly chewing groups, respectively. As these effects would apply inversely to cognitive performance, disregarding them in the event of positive results would be acceptable. It should be noted however that both negative [156] and positive [157] effects of long-term soft diets on rat learning abilities have been reported. Effects of long-term chewing behavior are also increasingly being reported in elderly populations [158]–[161], see [148], [162]–[164] for recent reviews. It appears that the loss of masticatory function increases the risk of dementia in humans and impairs spatial memory in rodents. These effects have been attributed to changes in Hippocampus morphology and function. Multiple possible mechanisms coupling chewing to the Hippocampus have been proposed but evidence for any of these is still missing, making ChR endogenous brain stimulation once again a possible candidate. I would like to briefly mention one interesting possible indirect mechanism. Experimental evidence shows that chewing/biting in rodents is a stress-coping strategy which effectively suppresses the detrimental effects of stress [165], [166], see [148], [167] for recent reviews. Combined with the effect of stress-induced bruxism [148], the effect of relaxation techniques on cognition [168]–[170] and the controversial usage of mouthguards to improve athletic performance [171], [172], relaxation of body and mind might have an important contribution to ChR findings. Less provocative scenarios are just as likely. For example, it could well be that ChR EF do modulate brain activity but do not have any effect on cognition. Alternatively, it could be that the relation between chewing muscle size and skull thickness stayed constant throughout the evolution of mankind to prevent detrimental influences or to allow beneficial influences. Regardless of whether ChR EF influence cognition or not, the wealth of literature showing that chewing is a crucial factor for mental health clearly indicates that chewing itself has an effect on cognition. D. Blood Vessels as Windows to the Brain – a Recording and Stimulation Perspective In chapter VII we have shown that blood vessels play an important role in volume conduction of neuronal activity to the surface of the head. Extrapolating on these results and on the results of my Diploma thesis [173], blood vessels open an interesting perspective for very- high-density EEG recordings, thereafter called µEEG. Indeed, we have shown the point spread function (PSF) of dipoles simulated in close proximity of emissary vessels to be narrower than expected for a closed skull [173]. In chapter VII we show that not only emissary vessels, but also diploe blood vessels reduce the blurring effect of the skull.

23 Discussion

Combined with the results from Dannhauer et al. that the whole diploe can be expected to have a similar effect [42], we think it likely that µEEG targeted at these locations could open a new non-invasive window to the brain. Indeed, the major shortcoming of non-invasive EEG is its low spatial resolution, in part due to the large distance between the measurement electrodes and the signal sources, but mostly due to the very low electric conductivity of the skull [4]. This low conductivity, combined with the high conductivity of brain tissue and CSF, introduces a spatial frequency low-pass filter [4]. The effect of this low-pass filter can be expected to be at its weakest in areas of high diploe density because of its higher conductivity. Furthermore, the effect of this low-pass filter can be expected to be non-existent above emissary foramina because there no skull would be located between electrodes and signal sources. But, as we have shown in chapter VII, the effect of the blood-vessel endothelium could potentially drastically reduce these effects. We have taken first steps to investigate this window in real human data by developing a µEEG grid with 8*8 electrodes 1 mm in diameter and 4 mm center-to-center intervals. A further promising perspective using blood vessels to ease the access to brain activity, is stent recordings. Stents are self-expanding metal meshes which are routinely used to mechanically dilate blood vessels which have become too narrow and are at risk of occluding. As the occlusion of a brain blood vessel has dramatic consequences (stroke), many stents are placed in brain blood vessels. Once implanted stents remain in place for the rest of the patients’ lives. Being able to place electrodes in stents and wirelessly recoding from them thus create an unparalleled opportunity for collecting invasive data using a routine procedure. First steps were already take in the early 1970s using simple wires [174], with first wired stent recordings in 2016 [175]. ‘Stentrodes’ (stent-electrodes) are nowadays being integrated with miniaturized wireless and energy harvesting technologies [personal communication Byron M. Yu]. We refer the interested reader to [176] deeper insights. We are currently collaborating with the University of Gent to identify the stent targets with the most potential using the head model presented in chapter VII. Blood vessels are not only interesting for improving non-invasive recordings of brain activity but also for improving non-invasive electrical brain stimulation. Indeed, non-invasive electrical brain stimulation underlies two major drawbacks. The first major drawback is the reciprocal effect of the skull blurring described in the previous paragraph. The skull makes it difficult to inject sufficient amounts of electricity into the brain because most of the electricity will take the electrically shorter route through the skin. Simply increasing the amount of injected electricity is not feasible as this would burn the skin. Using emissary vessels and the skull diploe to maximize the amount of electricity reaching the brain is an interesting perspective. Moreover, as Sven Wagner and colleagues have shown, it is possible to use the CSF to maximize the small amount of electricity reaching deeper cortical rejoins like the human auditory cortex [177]. It is therefore possible to envision doing the same using blood vessel to reach even deeper targets. The second major drawback of non-invasive electrical brain stimulation resides in the physical incompatibility of devising a stimulation which is maximally focal, i.e. which will stimulate only our target, and which has maximal intensity, i.e. which will be strong enough to stimulate our target. Based on this incompatibility, it is also close to impossible to stimulate deeper brain regions without using cortical brain regions as proxies. Stent-based electrical brain stimulation potentially alleviates these drawbacks. Using electrodes placed on stents to stimulate the brain can be expected to be as focal as deep brain stimulation, given their similar spatial proximity to the target. Blood vessels are thus a promising proxy for future neuroscientific research and applications. Nonetheless, the effect of electrical stimulation on the contractile blood vessel endothelium will need to be investigated to ensure that blood circulation is not hindered by the stimulation.

24 Discussion

E. Depth Localization of Neuronal Activity To the best of our knowledge, the work presented in chapter IX is the first attempt at depth localization of neuronal activity based on epicortical recordings. Traditionally, depth localization of neuronal activity is performed using current source density maps created with intracortical laminar electrodes. These electrodes have multiple contacts distributed along the electrode shank and can nowadays record from more than 1000 cites [178], [179]. The contacts thus record from different points across the cortical ribbon and allow a simple and precise depth localization of current sources and sinks. Unfortunately, such electrodes can only be used to study the neuronal activity in the direct vicinity of the shaft [180]. Sampling a broader cortical area would require the use of a multitude of such electrodes [181], which is technically challenging and strongly disrupts tissue integrity in the area to be studied. Using epicortical recordings to study the depth localization of neuronal activity has the advantage of being an electrophysiologically simple technique, which preserves the integrity of the cortical tissue. It does however appear that depth localization of neuronal activity using a quasi-planar array of electrodes located outside of the cortex would require complex mathematical procedures, namely inverse source localization. Inverse source localization has been developed in the context of non-invasive EEG to improve its spatial resolution [4]. Recently, a growing body of literature has applied inverse source localization to invasive EEG. These studies show that invasive EEG, both ECoG and SEEG, can indeed be spatially localized using source localization. Accurate localization is possible in the direct vicinity of the electrodes [182], [183] when the activity to localize is constituted of few synchronous sources [184] of small size [185]. Interestingly, accurate modeling of the SEEG electrodes and shaft is not required for accurate localizations [186], while it is for ECoG [113]. Based on the available literature, it appears that the major strength of invasive EEG in source localization is its high accuracy for small focal source directly next to the electrodes. This does sound quite unspectacular in the context of whole-brain localization but raises interesting possibilities for cortical depth localization based on epicortical recordings, especially in the case of µECoG. In chapter VIII we have presented the spatio-frequential characteristics of SEPs recorded using µECoG. In chapter IX we have made a first step towards localizing this activity with a laminar depth resolution. Despite having used a much simpler method than inverse source localization, a physiologically plausible distribution of source size, frequency and depth could be generated. Neuronal activity >= 80 Hz was consistently localized at depths <= 1 mm while neuronal activity <= 10 Hz was consistently localized at depths >= 1.4 mm. For an overview of the literature supporting these findings please see Table S1 in chapter IX. In the range between 10 Hz and 80 Hz results were less clear cut. Both deep and shallow localizations were made. Similarly, neuronal activity neuronal activity >= 100 Hz tended to be best represented by spatially extended sources while neuronal activity < 100 Hz tended to be best represented by single dipolar sources. Source size results were less clear cut than source depth results. This discrepancy could potentially indicate that the size of a synchronous neuronal population might not necessary dictate it’s frequency of synchronization while the depth location of a synchronous neuronal population might indeed impose a stronger constraint on synchronization frequency. Although these preliminary results appear to be quite in line with previous findings, caution has to be applied until a definitive validation using concurrent laminar recordings can be provided. Caution also applies to the results because of the very experimental nature of the methods used. Despite having taken extensive care, the anatomy of the VCHM is only an approximation due to the limited nature of the available MRI data. The MRI data, the formalin preserved Göttingen minipig brain from which the data was acquired, was limited

25 Discussion

because only the brain was present. We assumed that the missing head tissues would only minimally influence the µECoG forward solutions, as is being assumed in iEEG source localization (cf. second § of section), but sill manually implemented a CSF and dura compartment. Furthermore, the formalin preservation made gray and white matter close to undistinguishable. We thus approximated the gray matter compartment based on the curvature of the pial surface and assumed a constant thickness of the cortical ribbon based on manual inspections of the data. Besides having approximated the anatomy, the electrode contacts are also approximated. We took great care to model the position and inter-contact distances correctly but had to model the electrodes as point electrodes (one model node) to make the forward simulations tractable. It can be assumed that some spatial blurring and thus smoothing of the PSF would take place when using a realistic electrode model. We expect our approximation to have slightly underestimated the PSF’s HWHM, leading to results that should be slightly biased towards deeper and larger sources. As we have not observed this pattern in our results, larger sources tended to be superficial while still being small (radius max 1.5 mm) and deepest sources were consistently single dipole sources, we think that approximating the electrodes as single nodes had minimal effects. Lastly, we have extracted frequency-resolved topographies from stationary simulations by combining these with the data to be described by the simulations. This procedure was necessary to generate the relative spectras needed to compare the measured data with. In itself this procedure is a circularity and it is unclear how it affects the results. We have tried to limit this effect by combining 90 % of the data with the simulations and using these to analyze the remaining 10 % in a 10-fold cross-validation. Nevertheless, reproducing the results using absolute spectra and the raw simulation outputs would alleviate possible doubts and validate our procedure. As neither the procedure, nor any of the other parameters, were optimized to produce the reported, physiologically plausible, results, we do think it is safe to assume their validity. We believe that addressing the limitations of our model will not change the results but make them clearer, with an ever better overlap with electrophysiological literature. For future work, following questions arise which will need to be addressed: i) Can depth localization be performed using standard ECoG? ii) Can depth localization be performed using even simpler methods? iii) Can depth localization be performed using inverse source localization? iv) Are the localizations consistent across all methods? v) Can laminar depth resolution be achieved? Addressing these open questions is crucial for the broad adoption of depth localization based on epicortical recordings. If this method confirms its robustness and reliability, depth localization based on epicortical recordings would open up new perspectives on functional interpretation of LFP, especially for human research where the use of laminar electrodes is not justifiable for medical purposes.

26 References

IV. REFERENCES [1] S. Waldert, T. Pistohl, C. Braun, T. Ball, A. Aertsen, and C. Mehring, “A review on directional information in neural signals for brain-machine interfaces,” J. Physiol.-Paris, vol. 103, no. 3–5, pp. 244– 254, May 2009. [2] H. Berger, “Über das Elektrenkephalogramm des Menschen. Erste Mitteilung,” Arch. Für Psychiatr. Nervenkrankh., vol. 87, no. 1, pp. 527–570, Dec. 1929. [3] D. L. Schomer and F. L. da Silva, Eds., Niedermeyer’s Electroencephalography: Basic Principles, Clinical Applications, and Related Fields, 6. ed. Lippincott Williams & Wilkins, 2010. [4] P. L. Nunez and R. Srinivasan, Electric Fields of the Brain: The Neurophysics of EEG. Oxford University Press, 2006. [5] R. Cooper, A. . Winter, H. . Crow, and W. G. Walter, “Comparison of subcortical, cortical and scalp activity using chronically indwelling electrodes in man,” Electroencephalogr. Clin. Neurophysiol., vol. 18, no. 3, pp. 217–228, Feb. 1965. [6] H. M. Hamer et al., “Complications of invasive video-EEG monitoring with subdural grid electrodes,” Neurology, vol. 58, no. 1, pp. 97–103, Jan. 2002. [7] J. Derix, O. Iljina, A. Schulze-Bonhage, A. Aertsen, and T. Ball, “‘Doctor’ or ‘darling’? Decoding the communication partner from ECoG of the anterior temporal lobe during non-experimental, real-life social interaction,” Front. Hum. Neurosci., vol. 6, p. 251, 2012. [8] I. I. Rabi, J. R. Zacharias, S. Millman, and P. Kusch, “A New Method of Measuring Nuclear Magnetic Moment,” Phys. Rev., vol. 53, no. 4, p. 318, 1938. [9] F. Bloch, W. W. Hansen, and M. Packard, “Nuclear Induction,” Phys. Rev., vol. 69, no. 3–4, p. 127, 1946. [10] E. M. Purcell, H. C. Torrey, and R. V. Pound, “Resonance Absorption by Nuclear Magnetic Moments in a Solid,” Phys. Rev., vol. 69, no. 1–2, pp. 37–38, 1946. [11] E. L. Hahn, “Spin Echoes,” Phys. Rev., vol. 80, no. 4, pp. 580–594, 1950. [12] H. Y. Carr, “Free Precession Techniques in Nuclear Magnetic Resonance.,” 1953. [13] P. C. Lauterbur, “Image Formation by Induced Local Interactions: Examples Employing Nuclear Magnetic Resonance,” Nature, vol. 242, no. 5394, pp. 190–191, Mar. 1973. [14] P. C. Lauterbur, “Magnetic resonance zeugmatography,” Pure Appl. Chem., vol. 40, no. 1–2, Jan. 1974. [15] P. K. Grannell and P. Mansfield, “Microscopy in vivo by nuclear magnetic resonance,” Phys. Med. Biol., vol. 20, no. 3, p. 477, 1975. [16] R. Damadian, M. Goldsmith, and L. Minkoff, “NMR in cancer: XVI. FONAR image of the live human body.,” Physiol. Chem. Phys., vol. 9, no. 1, p. 97, 1977. [17] W. S. Hinshaw, P. A. Bottomley, and G. N. Holland, “Radiographic thin-section image of the human wrist by nuclear magnetic resonance,” Nature, vol. 270, no. 5639, pp. 722–723, Dec. 1977. [18] P. Mansfield, “Multi-planar image formation using NMR spin echoes,” J. Phys. C Solid State Phys., vol. 10, no. 3, p. L55, 1977. [19] J. Larmor, “LXIII. On the theory of the magnetic influence on spectra; and on the radiation from moving ions,” Philos. Mag. Ser. 5, vol. 44, no. 271, pp. 503–512, Dec. 1897. [20] R. D. Pascual-Marqui, “Review of methods for solving the EEG inverse problem,” Int. J. Bioelectromagn., vol. 1, no. 1, pp. 75–86, 1999. [21] C. M. Michel, M. M. Murray, G. Lantz, S. Gonzalez, L. Spinelli, and R. Grave de Peralta, “EEG source imaging,” Clin. Neurophysiol., vol. 115, no. 10, pp. 2195–2222, Oct. 2004. [22] R. Grech et al., “Review on solving the inverse problem in EEG source analysis,” J. NeuroEngineering Rehabil., vol. 5, no. 1, p. 25, Nov. 2008. [23] C. M. Michel and M. M. Murray, “Towards the utilization of EEG as a brain imaging tool,” NeuroImage, vol. 61, no. 2, pp. 371–385, Jun. 2012. [24] M. R. Schneider, “A Multistage Process for Computing Virtual Dipolar Sources of EEG Discharges from Surface Information,” IEEE Trans. Biomed. Eng., vol. BME-19, no. 1, pp. 1–12, 1972. [25] T. J. Carter, M. Sermesant, D. M. Cash, D. C. Barratt, C. Tanner, and D. J. Hawkes, “Application of soft tissue modelling to image-guided surgery,” Med. Eng. Phys., vol. 27, no. 10, pp. 893–909, Dec. 2005. [26] K. Miller et al., “Modelling brain deformations for computer-integrated neurosurgery,” Int. J. Numer. Methods Biomed. Eng., vol. 26, no. 1, pp. 117–138, 2010. [27] L. Voo, S. Kumaresan, F. A. Pintar, N. Yoganandan, and A. J. Sances, “Finite-element models of the human head,” Med. Biol. Eng. Comput., vol. 34, no. 5, pp. 375–381, Sep. 1996. [28] J. (James) Yang, J. Dai, and Z. Zhuang, “Human Head Modeling and Personal Head Protective Equipment: A Literature Review,” in Digital Human Modeling, V. G. Duffy, Ed. Springer Berlin Heidelberg, 2009, pp. 661–670.

27 References

[29] M. B. Panzer, B. S. Myers, B. P. Capehart, and C. R. Bass, “Development of a Finite Element Model for Blast Brain Injury and the Effects of CSF Cavitation,” Ann. Biomed. Eng., vol. 40, no. 7, pp. 1530–1544, Feb. 2012. [30] A. Opitz, M. Windhoff, R. M. Heidemann, R. Turner, and A. Thielscher, “How the brain tissue shapes the electric field induced by transcranial magnetic stimulation,” NeuroImage, vol. 58, no. 3, pp. 849–859, Oct. 2011. [31] A. Datta, J. P. Dmochowski, B. Guleyupoglu, M. Bikson, and F. Fregni, “Cranial electrotherapy stimulation and transcranial pulsed current stimulation: A computer based high-resolution modeling study,” NeuroImage, vol. 65, pp. 280–287, Jan. 2013. [32] M. Fernández-Corazza, L. Beltrachini, N. von Ellenrieder, and C. H. Muravchik, “Analysis of parametric estimation of head tissue conductivities using Electrical Impedance Tomography,” Biomed. Signal Process. Control, vol. 8, no. 6, pp. 830–837, Nov. 2013. [33] R. Bayford and A. Tizzard, “Bioimpedance imaging: an overview of potential clinical applications,” Analyst, vol. 137, no. 20, pp. 4635–4643, Sep. 2012. [34] M. Vonach et al., “A method for rapid production of subject specific finite element meshes for electrical impedance tomography of the human head,” Physiol. Meas., vol. 33, no. 5, p. 801, May 2012. [35] C. Ramon, P. H. Schimpf, and J. Haueisen, “Influence of head models on EEG simulations and inverse source localizations,” Biomed. Eng. OnLine, vol. 5, p. 10, Feb. 2006. [36] J. Haueisen, C. Ramon, M. Eiselt, H. Brauer, and H. Nowak, “Influence of tissue resistivities on neuromagnetic fields and electric potentials studied with a finite element model of the head,” IEEE Trans. Biomed. Eng., vol. 44, no. 8, pp. 727–735, Aug. 1997. [37] C. Ramon et al., “Similarities Between Simulated Spatial Spectra of Scalp EEG, MEG and Structural MRI,” Brain Topogr., vol. 22, no. 3, pp. 191–196, Nov. 2009. [38] M. W. Slutzky, L. R. Jordan, T. Krieg, M. Chen, D. J. Mogul, and L. E. Miller, “Optimal Spacing of Surface Electrode Arrays for Brain Machine Interface Applications,” J. Neural Eng., vol. 7, no. 2, p. 26004, Apr. 2010. [39] R. Srinivasan, D. Tucker, and M. Murias, “Estimating the spatial Nyquist of the human EEG,” Behav. Res. Methods, vol. 30, no. 1, pp. 8–19, 1998. [40] G. Huiskamp, M. Vroeijenstijn, R. van Dijk, G. Wieneke, and A. C. van Huffelen, “The need for correct realistic geometry in the inverse EEG problem,” IEEE Trans. Biomed. Eng., vol. 46, no. 11, pp. 1281– 1287, Nov. 1999. [41] F. Vatta, P. Bruno, S. Mininel, and P. Inchingolo, “EEG Simulation Accuracy: Reference Choice and Head Models Extension,” Int. J. Bioelectromagn., vol. 7, no. 1, pp. 154–157, 2005. [42] M. Dannhauer, B. Lanfer, C. H. Wolters, and T. R. Knösche, “Modeling of the human skull in EEG source analysis,” Hum. Brain Mapp., vol. 32, no. 9, pp. 1383–1399, Sep. 2011. [43] B. Lanfer, M. Scherg, M. Dannhauer, T. R. Knösche, M. Burger, and C. H. Wolters, “Influences of skull segmentation inaccuracies on EEG source analysis,” NeuroImage, vol. 62, no. 1, pp. 418–431, Aug. 2012. [44] K. Wendel, J. Väisänen, A. Kybartaite, J. Hyttinen, and J. Malmivuo, “The significance of relative conductivity on thin layers in EEG sensitivity distributions,” Biomed. Tech. Eng., vol. 55, no. 3, pp. 123– 131, Jan. 2010. [45] A. Kybartaite, “Computational representation of a realistic head and brain volume conductor model: electroencephalography simulation and visualization study,” Int. J. Numer. Methods Biomed. Eng., vol. 28, no. 11, pp. 1144–1155, May 2012. [46] M. A. B. Brazier, “The electrical fields at the surface of the head during sleep,” Electroencephalogr. Clin. Neurophysiol., vol. 1, no. 1–4, pp. 195–204, 1949. [47] C. D. Geisler and G. L. Gerstein, “The surface EEG in relation to its sources,” Electroencephalogr. Clin. Neurophysiol., vol. 13, no. 6, pp. 927–934, Dec. 1961. [48] E. Frank, “Electric Potential Produced by Two Point Current Sources in a Homogeneous Conducting Sphere,” J. Appl. Phys., vol. 23, no. 11, pp. 1225–1228, Nov. 1952. [49] F. N. Wilson and R. H. Bayley, “The Electric Field of an Eccentric Dipole in a Homogeneous Spherical Conducting Medium,” Circulation, vol. 1, no. 1, pp. 84–92, Jan. 1950. [50] R. S. Hosek, A. Sances, R. W. Jodat, and S. J. Larson, “The Contributions of Intracerebral Currents to the EEG and Evoked Potentials,” IEEE Trans. Biomed. Eng., vol. BME-25, no. 5, pp. 405–413, Sep. 1978. [51] J. W. H. Meijs and M. J. Peters, “The EEG and MEG, Using a Model of Eccentric Spheres to Describe the Head,” IEEE Trans. Biomed. Eng., vol. BME-34, no. 12, pp. 913–920, Dec. 1987. [52] J. G. Witwer, G. J. Trezek, and D. L. Jewett, “The Effect of Media Inhomogeneities Upon Intracranial Electrical Fields,” IEEE Trans. Biomed. Eng., vol. BME-19, no. 5, pp. 352–362, Sep. 1972. [53] J. W. . Meijs, F. G. . Bosch, M. . Peters, and F. . Lopes da Silva, “On the magnetic field distribution generated by a dipolar current source situated in a realistically shaped compartment model of the head,” Electroencephalogr. Clin. Neurophysiol., vol. 66, no. 3, pp. 286–298, Mar. 1987. [54] M. Fuchs, J. Kastner, M. Wagner, S. Hawes, and J. S. Ebersole, “A standardized boundary element method volume conductor model,” Clin. Neurophysiol., vol. 113, no. 5, pp. 702–712, May 2002.

28 References

[55] N. G. Sepulveda, C. F. Walker, and R. G. Heath, “Finite element analysis of current pathways with implanted electrodes,” J. Biomed. Eng., vol. 5, no. 1, pp. 41–48, Jan. 1983. [56] Y. Yan, P. Nunez, and R. Hart, “Finite-element model of the human head: scalp potentials due to dipole sources,” Med. Biol. Eng. Comput., vol. 29, no. 5, pp. 475–481, 1991. [57] S. Abboud, Y. Eshel, S. Levy, and M. Rosenfeld, “Numerical Calculation of the Potential Distribution Due to Dipole Sources in a Spherical Model of the Head,” Comput. Biomed. Res., vol. 27, no. 6, pp. 441– 455, Dec. 1994. [58] C. Ramon, P. Schimpf, J. Haueisen, M. Holmes, and A. Ishimaru, “Role of Soft Bone, CSF and Gray Matter in EEG Simulations,” Brain Topogr., vol. 16, no. 4, pp. 245–248, Jun. 2004. [59] N. Chauveau, X. Franceries, B. Doyon, B. Rigaud, J. P. Morucci, and P. Celsis, “Effects of skull thickness, anisotropy, and inhomogeneity on forward EEG/ERP computations using a spherical three- dimensional resistor mesh model,” Hum. Brain Mapp., vol. 21, no. 2, pp. 86–97, 2004. [60] C. G. Bénar and J. Gotman, “Modeling of post-surgical brain and skull defects in the EEG inverse problem with the boundary element method,” Clin. Neurophysiol., vol. 113, no. 1, pp. 48–56, Jan. 2002. [61] B. C. Heasman, A. Valentín, G. Alarcón, J. J. García Seoane, C. D. Binnie, and C. N. Guy, “A hole in the skull distorts substantially the distribution of extracranial electrical fields in an in vitro model,” J. Clin. Neurophysiol., vol. 19, no. 2, p. 163, 2002. [62] A. Anwander, C. H. Wolters, M. Dumpelmann, and T. Knösche, “Influence of realistic skull and white matter anisotropy on the inverse problem in EEG/MEG-source localization,” in Proceedings of The 13th International Conference on Biomagnetism, 2002, pp. 679–681. [63] J. P. Ary, S. A. Klein, and D. H. Fender, “Location of sources of evoked scalp potentials: corrections for skull and scalp thicknesses,” Biomed. Eng. IEEE Trans. On, no. 6, pp. 447–452, 1981. [64] B. N. Cuffin, “Effects of local variations in skull and scalp thickness on EEG’s and MEG’s,” IEEE Trans. Biomed. Eng., vol. 40, no. 1, pp. 42–48, Jan. 1993. [65] S. P. van den Broek, F. Reinders, M. Donderwinkel, and M. J. Peters, “Volume conduction effects in EEG and MEG,” Electroencephalogr. Clin. Neurophysiol., vol. 106, no. 6, pp. 522–534, Jun. 1998. [66] J. Haueisen et al., “The Influence of Brain Tissue Anisotropy on Human EEG and MEG,” NeuroImage, vol. 15, no. 1, pp. 159–166, Jan. 2002. [67] D. Gullmar, J. Haueisen, and J. R. Reichenbach, “Influence of anisotropic electrical conductivity in white matter tissue on the EEG/MEG forward and inverse solution. A high-resolution whole head simulation study,” NeuroImage, vol. 51, no. 1, pp. 145–163, May 2010. [68] N. B. Bangera et al., “Experimental validation of the influence of white matter anisotropy on the intracranial EEG forward solution,” J. Comput. Neurosci., vol. 29, no. 3, pp. 371–387, Dec. 2010. [69] C. H. Wolters, A. Anwander, X. Tricoche, D. Weinstein, M. A. Koch, and R. S. MacLeod, “Influence of tissue conductivity anisotropy on EEG/MEG field and return current computation in a realistic head model: A simulation and visualization study using high-resolution finite element modeling,” NeuroImage, vol. 30, no. 3, pp. 813–826, Apr. 2006. [70] K. Wendel, N. G. Narra, M. Hannula, P. Kauppinen, and J. Malmivuo, “The Influence of CSF on EEG Sensitivity Distributions of Multilayered Head Models,” IEEE Trans. Biomed. Eng., vol. 55, no. 4, pp. 1454–1456, Apr. 2008. [71] B. Lanfer, I. Paul-Jordanov, M. Scherg, and C. H. Wolters, “Influence of interior cerebrospinal fluid compartments on EEG source analysis,” Biomed. Tech. (Berl), Aug. 2012. [72] P. Bruno, F. Vatta, S. Mininel, and P. Inchingolo, “Head model extension for the study of bioelectric phenomena,” Biomed. Sci. Instrum., vol. 39, pp. 59–64, 2003. [73] A. Datta, J. M. Baker, M. Bikson, and J. Fridriksson, “Individualized model predicts brain current flow during transcranial direct-current stimulation treatment in responsive stroke patient,” Brain Stimulat., vol. 4, no. 3, pp. 169–174, Jul. 2011. [74] M. Windhoff, A. Opitz, and A. Thielscher, “Electric field calculations in brain stimulation based on finite elements: An optimized processing pipeline for the generation and usage of accurate individual head models,” Hum. Brain Mapp., vol. 34, no. 4, pp. 923–935, 2013. [75] F. Lusebrink, A. Wollrab, and O. Speck, “Cortical thickness determination of the human brain using high resolution 3T and 7T MRI data,” NeuroImage, vol. 70, pp. 122–131, Apr. 2013. [76] J. Haueisen, C. Ramon, P. Czapski, and M. Eiselt, “On the influence of volume currents and extended sources on neuromagnetic fields: a simulation study,” Ann. Biomed. Eng., vol. 23, no. 6, pp. 728–739, Dec. 1995. [77] M. Rullmann, A. Anwander, M. Dannhauer, S. K. Warfield, F. H. Duffy, and C. H. Wolters, “EEG source analysis of epileptiform activity using a 1mm anisotropic hexahedra finite element head model,” NeuroImage, vol. 44, no. 2, pp. 399–410, Jan. 2009. [78] A. Datta, J. M. Baker, M. Bikson, and J. Fridriksson, “Individualized model predicts brain current flow during transcranial direct-current stimulation treatment in responsive stroke patient,” Brain Stimulat., vol. 4, no. 3, pp. 169–174, Jul. 2011.

29 References

[79] S. Pursiainen, F. Lucka, and C. H. Wolters, “Complete electrode model in EEG: relationship and differences to the point electrode model,” Phys. Med. Biol., vol. 57, no. 4, pp. 999–1017, Feb. 2012. [80] T. Neuling, S. Wagner, C. H. Wolters, T. Zaehle, and C. S. Herrmann, “Finite-Element Model Predicts Current Density Distribution for Clinical Applications of tDCS and tACS,” Front. Psychiatry, vol. 3, Sep. 2012. [81] F. Lucka, S. Pursiainen, M. Burger, and C. H. Wolters, “Hierarchical Bayesian inference for the EEG inverse problem using realistic FE head models: Depth localization and source separation for focal primary currents,” NeuroImage, vol. 61, no. 4, pp. 1364–1382, Jul. 2012. [82] L. D. J. Fiederer et al., “The role of blood vessels in high-resolution volume conductor head modeling of EEG,” NeuroImage, vol. 128, pp. 193–208, Mar. 2016. [83] H. Buchner, Praxisbuch Evozierte Potenziale: Grundlagen, Befundung, Beurteilung und differenzialdiagnostische Abgrenzung, 1st ed. Stuttgart: Georg Thieme Verlag, 2014. [84] A. J. Derbyshire, B. Rempel, A. Forbes, and E. F. Lambert, “The effects of anesthetics on action potentials in the cerebral cortex of the cat,” Am. J. Physiol., vol. 116, no. 3, pp. 577–596, 1936. [85] S. H. Bartley and P. Heinbecker, “The response of the sensorimotor cortex to stimulation of a peripheral nerve,” Am. J. Physiol., vol. 121, no. 1, pp. 21–31, 1937. [86] A. Forbes and B. R. Morison, “Cortical Response to Sensory Stimulation Under Deep Barbiturate Narcosis,” J. Neurophysiol., vol. 2, no. 2, pp. 112–128, Mar. 1939. [87] E. D. Adrian, “Afferent discharges to the cerebral cortex from peripheral sense organs,” J. Physiol., vol. 100, no. 2, pp. 159–191, Sep. 1941. [88] G. D. Dawson, “Cerebral Responses to Electrical Stimulation of Peripheral Nerve in Man,” J. Neurol. Neurosurg. Psychiatry, vol. 10, no. 3, pp. 134–140, Aug. 1947. [89] G. D. Dawson, “Investigations on a Patient Subject to Myoclonic Seizures After Sensory Stimulation,” J. Neurol. Neurosurg. Psychiatry, vol. 10, no. 4, pp. 141–162, Nov. 1947. [90] L.-E. Larsson, “Electroencephalographic responses to peripheral nerve stimulation in man,” Electroencephalogr. Clin. Neurophysiol., vol. 5, no. 3, pp. 377–384, Aug. 1953. [91] T. Allison, “Recovery functions of somatosensory evoked responses in man,” Electroencephalogr. Clin. Neurophysiol., vol. 14, no. 3, pp. 331–343, Jun. 1962. [92] T. Allison, W. R. Goff, P. D. Williamson, and J. C. Van Gilder, “On the neural origin of early components of the human somatosensory ,” in Clinical Uses of Cerebral, Brainstem and Spinal Somatosensory Evoked Potentials., vol. 7, J. E. Desmedt, Ed. Basel: Karger, 1980, pp. 51–68. [93] R. Broughton, in Average Evoked Potentials: Methods, Results, and Evaluations, E. Donchin and D. B. Undsley, Eds. Washington: National Aeronautics and Space Administration, 1969, pp. 79–84. [94] D. R. Giblin, “Somatosensory Evoked Potentials in Healthy Subjects and in Patients with Lesions of the Nervous System*†,” Ann. N. Y. Acad. Sci., vol. 112, no. 1, pp. 93–142, May 1964. [95] J. E. Desmedt and G. Cheron, “Non-cephalic reference recording of early somatosensory potentials to finger stimulation in adult or aging normal: differentiation of widespread N18 and contralateral N20 from the prerolandic p22 and N30 components,” Electroencephalogr. Clin. Neurophysiol., vol. 52, no. 6, pp. 553–570, Dec. 1981. [96] T. Allison, G. McCarthy, C. C. Wood, and S. J. Jones, “Potentials evoked in human and monkey cerebral cortex by stimulation of the median nerve. A review of scalp and intracranial recordings,” Brain J. Neurol., vol. 114 ( Pt 6), no. 6, pp. 2465–2503, Dec. 1991. [97] S. J. Dreyer, D. Dumitru, and J. C. King, “Anodal block V anodal stimulation. Fact or fiction,” Am. J. Phys. Med. Rehabil. Assoc. Acad. Physiatr., vol. 72, no. 1, pp. 10–18, Feb. 1993. [98] A. S. Wee, A. A. Leis, A. R. Kuhn, and R. W. Gilbert, “Anodal block: can this occur during routine nerve conduction studies?,” Electromyogr. Clin. Neurophysiol., vol. 40, no. 7, pp. 387–391, Nov. 2000. [99] J. Tiihonen, R. Hari, and M. Hämäläinen, “Early deflections of cerebral magnetic responses to median nerve stimulation,” Electroencephalogr. Clin. Neurophysiol. Potentials Sect., vol. 74, no. 4, pp. 290–296, Jul. 1989. [100] L. D. J. Fiederer et al., “An Augmented Reality Approach for BCI Control of Intelligent Autonomous Robots,” in Biomed. Eng. (Berlin), 2014, vol. 59, pp. S1097–S1097. [101] T. Lampe, L. D. J. Fiederer, M. Voelker, A. Knorr, M. Riedmiller, and T. Ball, “A Brain-Computer Interface for High-Level Remote Control of an Autonomous, Reinforcement-Learning-Based Robotic System for Reaching and Grasping,” in Proceedings of the 19th International Conference on Intelligent User Interfaces, New York, NY, USA, 2014, pp. 83–88. [102] S. Schröer et al., “An autonomous robotic assistant for drinking,” in Robotics and Automation (ICRA), 2015 IEEE International Conference on, 2015, pp. 6482–6487. [103] R. T. Schirrmeister et al., “ with convolutional neural networks for EEG decoding and visualization,” Hum. Brain Mapp., vol. 38, no. 11, pp. 5391–5420, Nov. 2017. [104] F. Burget et al., “Acting Thoughts: Towards a Mobile Robotic Service Assistant for Users with Limited Communication Skills,” ArXiv170706633 Cs, Jul. 2017.

30 References

[105] R. T. Schirrmeister et al., “DESIGNING AND UNDERSTANDING CONVOLUTIONAL NETWORKS FOR DECODING EXECUTED MOVEMENTS FROM EEG,” in The First Biannual Neuroadaptive Technology Conference, 2017, pp. 143–144. [106] M. Voelker, S. Berberich, E. Andreev, L. D. Fiederer, W. Burgard, and T. Ball, “BETWEEN-SUBJECT TRANSFER LEARNING FOR CLASSIFICATION OF ERROR-RELATED SIGNALS IN HIGH- DENSITY EEG,” in The First Biannual Neuroadaptive Technology Conference, 2017, vol. 81, pp. 47–50. [107] M. Völker, R. T. Schirrmeister, L. D. J. Fiederer, W. Burgard, and T. Ball, “Deep Transfer Learning for Error Decoding from Non-Invasive EEG,” ArXiv171009139 Cs Q-Bio, Oct. 2017. [108] L. D. J. Fiederer, M. Völker, M. Glasstetter, and T. Ball, “Non-invasive high-gamma mapping in an optimized EEG lab,” Clin. Neurophysiol., vol. 128, no. 10, p. e313, Oct. 2017. [109] M. Voelker et al., “Global and Local Dynamics of High-Gamma Activity Underlying Error Processing in the Human Brain Revealed by Noninvasive and Intracranial EEG,” bioRxiv, p. 166280, Aug. 2017. [110] L. L. Kahle, L. D. J. Fiederer, S. Contzen, M. Völker, and T. Ball, “Spatio-temporal characteristics of somatosensory-evoked 50-70 Hz high-gamma activity of the EEG,” submitted. [111] D. Kuhner et al., “Deep Learning Based BCI Control of a Robotic Service Assistant Using Intelligent Goal Formulation,” bioRxiv, p. 282848, Mar. 2018. [112] N. von Ellenrieder, L. Beltrachini, C. H. Muravchik, and J. Gotman, “Extent of cortical generators visible on the scalp: Effect of a subdural grid,” NeuroImage, vol. 101, pp. 787–795, Nov. 2014. [113] B. Lanfer, C. Röer, M. Scherg, S. Rampp, C. Kellinghaus, and C. H. Wolters, “Influence of a Silastic ECoG Grid on EEG/ECoG Based Source Analysis,” Brain Topogr., vol. 26, no. 2, pp. 212–228, Sep. 2012. [114] Y. Zhang, L. Ding, W. van Drongelen, K. Hecox, D. M. Frim, and B. He, “A cortical potential imaging study from simultaneous extra- and intracranial electrical recordings by means of the finite element method,” NeuroImage, vol. 31, no. 4, pp. 1513–1524, Jul. 2006. [115] J. X. Tao, A. Ray, S. Hawes‐Ebersole, and J. S. Ebersole, “Intracranial EEG Substrates of Scalp EEG Interictal Spikes,” Epilepsia, vol. 46, no. 5, pp. 669–676, May 2005. [116] J. X. Tao, M. Baldwin, S. Hawes-Ebersole, and J. S. Ebersole, “Cortical Substrates of Scalp EEG Epileptiform Discharges,” J. Clin. Neurophysiol., vol. 24, no. 2, pp. 96–100, Apr. 2007. [117] C. H. Wong et al., “Risk factors for complications during intracranial electrode recording in presurgical evaluation of drug resistant partial epilepsy,” Acta Neurochir. (Wien), vol. 151, no. 1, p. 37, Jan. 2009. [118] Ç. Önal et al., “Complications of invasive subdural grid monitoring in children with epilepsy,” J. Neurosurg., vol. 98, no. 5, pp. 1017–1026, May 2003. [119] G. C. Wiggins, K. Elisevich, and B. J. Smith, “Morbidity and infection in combined subdural grid and strip electrode investigation for intractable epilepsy,” Epilepsy Res., vol. 37, no. 1, pp. 73–80, Oct. 1999. [120] D. Reato, A. Rahman, M. Bikson, and L. C. Parra, “Low-Intensity Electrical Stimulation Affects Network Dynamics by Modulating Population Rate and Spike Timing,” J. Neurosci., vol. 30, no. 45, pp. 15067– 15079, Nov. 2010. [121] F. Fröhlich and D. A. McCormick, “Endogenous Electric Fields May Guide Neocortical Network Activity,” Neuron, vol. 67, no. 1, pp. 129–143, Jul. 2010. [122] C. A. Anastassiou, R. Perin, H. Markram, and C. Koch, “Ephaptic coupling of cortical neurons,” Nat. Neurosci., vol. 14, no. 2, pp. 217–223, Feb. 2011. [123] C. Miniussi, W. Paulus, and P. M. Rossini, Eds., Transcranial Brain Stimulation. Boca Raton, FL: CRC Press, 2012. [124] D. Terney, L. Chaieb, V. Moliadze, A. Antal, and W. Paulus, “Increasing Human Brain Excitability by Transcranial High-Frequency Random Noise Stimulation,” J. Neurosci., vol. 28, no. 52, pp. 14147–14155, Dec. 2008. [125] L. Marshall, H. Helgadóttir, M. Mölle, and J. Born, “Boosting slow oscillations during sleep potentiates memory,” Nature, vol. 444, no. 7119, pp. 610–613, Nov. 2006. [126] L. D. J. Fiederer et al., “Electrical Stimulation of the Human Cerebral Cortex by Extracranial Muscle Activity: Effect Quantification With Intracranial EEG and FEM Simulations,” IEEE Trans. Biomed. Eng., vol. 63, no. 12, pp. 2552–2563, Dec. 2016. [127] L. Wilkinson, A. Scholey, and K. Wesnes, “Chewing gum selectively improves aspects of memory in healthy volunteers,” Appetite, vol. 38, no. 3, pp. 235–236, Jun. 2002. [128] R. Stephens and R. J. Tunney, “Role of glucose in chewing gum-related facilitation of cognitive function,” Appetite, vol. 43, no. 2, pp. 211–213, Oct. 2004. [129] O. Tucha, L. Mecklinger, K. Maier, M. Hammerl, and K. W. Lange, “Chewing gum differentially affects aspects of attention in healthy subjects,” Appetite, vol. 42, no. 3, pp. 327–329, Jun. 2004. [130] J. R. Baker, J. B. Bezance, E. Zellaby, and J. P. Aggleton, “Chewing gum can produce context-dependent effects upon memory,” Appetite, vol. 43, no. 2, pp. 207–210, Oct. 2004. [131] Y. Hirano et al., “Effects of chewing in working memory processing,” Neurosci. Lett., vol. 436, no. 2, pp. 189–192, May 2008.

31 References

[132] A. Smith, “Effects of chewing gum on mood, learning, memory and performance of an intelligence test,” Nutr. Neurosci., vol. 12, no. 2, pp. 81–88, Apr. 2009. [133] D. H. Rost, L. Wirthwein, K. Frey, and E. Becker, “Steigert Kaugummikauen das kognitive Leistungsvermögen?,” Z. Für Pädagog. Psychol., vol. 24, no. 1, pp. 39–49, Jan. 2010. [134] C. A. Johnston, C. Tyler, S. A. Stansberry, J. P. Moreno, and J. P. Foreyt, “Brief report: Gum chewing affects standardized math scores in adolescents,” J. Adolesc., vol. 35, no. 2, pp. 455–459, Apr. 2012. [135] A. J. Johnson and C. Miles, “Evidence against memorial facilitation and context-dependent memory effects through the chewing of gum,” Appetite, vol. 48, no. 3, pp. 394–396, May 2007. [136] C. Miles and A. J. Johnson, “Chewing gum and context-dependent memory effects: A re-examination,” Appetite, vol. 48, no. 2, pp. 154–158, Mar. 2007. [137] L. Tucha et al., “Detrimental effects of gum chewing on vigilance in children with attention deficit hyperactivity disorder,” Appetite, vol. 55, no. 3, pp. 679–684, Dec. 2010. [138] M. D. Kozlov, R. W. Hughes, and D. M. Jones, “Gummed-up memory: Chewing gum impairs short-term recall,” Q. J. Exp. Psychol., vol. 65, no. 3, pp. 501–513, Mar. 2012. [139] A. A. Overman, J. Sun, A. C. Golding, and D. Prevost, “Chewing gum does not induce context-dependent memory when flavor is held constant,” Appetite, vol. 53, no. 2, pp. 253–255, Oct. 2009. [140] A. Smith, “Effects of chewing gum on cognitive function, mood and physiology in stressed and non- stressed volunteers,” Nutr. Neurosci., vol. 13, no. 1, pp. 7–16, Feb. 2010. [141] M. Wada, Y. Hoshi, Y. Iguchi, and I. Kida, “Near-infrared spectroscopic study on the effects of chewing on short-term memory,” Appetite, vol. 57, no. 3, pp. 749–752, Dec. 2011. [142] L. Tucha and W. Simpson, “The role of time on task performance in modifying the effects of gum chewing on attention,” Appetite, vol. 56, no. 2, pp. 299–301, Apr. 2011. [143] S. V. Onyper, T. L. Carr, J. S. Farrar, and B. R. Floyd, “Cognitive advantages of chewing gum. Now you see them, now you don’t,” Appetite, vol. 57, no. 2, pp. 321–328, Oct. 2011. [144] A. P. Allen and A. P. Smith, “Effects of chewing gum and time-on-task on alertness and attention,” Nutr. Neurosci., vol. 15, no. 4, pp. 176–185, Jul. 2012. [145] A. P. Allen and A. P. Smith, “Demand characteristics, pre-test attitudes and time-on-task trends in the effects of chewing gum on attention and reported mood in healthy volunteers,” Appetite, vol. 59, no. 2, pp. 349–356, Oct. 2012. [146] A. J. Johnson et al., “The effect of chewing gum on physiological and self-rated measures of alertness and daytime sleepiness,” Physiol. Behav., vol. 105, no. 3, pp. 815–820, Feb. 2012. [147] Y. Hirano et al., “Effects of chewing on cognitive processing speed,” Brain Cogn., vol. 81, no. 3, pp. 376– 381, Apr. 2013. [148] R. A. F. Weijenberg and F. Lobbezoo, “Chew the Pain Away: Oral Habits to Cope with Pain and Stress and to Stimulate Cognition,” BioMed Research International, 2015. [Online]. Available: https://www.hindawi.com/journals/bmri/2015/149431/. [Accessed: 27-Aug-2017]. [149] L. Tucha and J. Koerts, “Gum Chewing and Cognition: An Overview,” Neurosci. Med., vol. 03, no. 03, p. 243, Sep. 2012. [150] A. P. Allen and A. P. Smith, “A review of the evidence that chewing gum affects stress, alertness and cognition,” J. Behav. Neurosci. Res., vol. 9, no. 1, pp. 7–23, 2011. [151] R. A. F. Weijenberg, E. J. A. Scherder, and F. Lobbezoo, “Mastication for the mind—The relationship between mastication and cognition in ageing and dementia,” Neurosci. Biobehav. Rev., vol. 35, no. 3, pp. 483–497, Jan. 2011. [152] A. P. Allen, “Chewing gum’s effects on alertness, performance and stress,” phd, Cardiff University, 2013. [153] R. Wrangham, Catching Fire: How Cooking Made Us Human, 1st ed. New York: Basic Books, 2009. [154] W. J. Moore, “Masticatory function and skull growth,” Proc. Zool. Soc. Lond., vol. 146, no. 2, pp. 123– 131, Oct. 1965. [155] D. C. Katz, M. N. Grote, and T. D. Weaver, “Changes in human skull morphology across the agricultural transition are consistent with softer diets in preindustrial farming groups,” Proc. Natl. Acad. Sci., vol. 114, no. 34, pp. 9050–9055, Aug. 2017. [156] S. Kushida et al., “Soft-diet feeding decreases dopamine release and impairs aversion learning in Alzheimer model rats,” Neurosci. Lett., vol. 439, no. 2, pp. 208–211, Jul. 2008. [157] Y. Endo, T. Mizuno, K. Fujita, T. Funabashi, and F. Kimura, “Soft-diet feeding during development enhances later learning abilities in female rats,” Physiol. Behav., vol. 56, no. 4, pp. 629–633, Oct. 1994. [158] J. Li, H. Xu, W. Pan, and B. Wu, “Association between tooth loss and cognitive decline: A 13-year longitudinal study of Chinese older adults,” PLOS ONE, vol. 12, no. 2, p. e0171404, Feb. 2017. [159] Y. Fukushima-Nakayama et al., “Reduced Mastication Impairs Memory Function,” J. Dent. Res., vol. 96, no. 9, pp. 1058–1066, Aug. 2017. [160] C.-S. Lin, S.-Y. Wu, C.-Y. Wu, and H.-W. Ko, “Gray Matter Volume and Resting-State Functional Connectivity of the Motor Cortex-Cerebellum Network Reflect the Individual Variation in Masticatory Performance in Healthy Elderly People,” Front. Aging Neurosci., vol. 7, 2016.

32 References

[161] F. Elsig et al., “Tooth loss, chewing efficiency and cognitive impairment in geriatric patients,” Gerodontology, vol. 32, no. 2, pp. 149–156, Jun. 2015. [162] H. Chen, M. Iinuma, M. Onozuka, and K.-Y. Kubo, “Chewing Maintains Hippocampus-Dependent Cognitive Function,” Int. J. Med. Sci., vol. 12, no. 6, pp. 502–509, 2015. [163] F. B. Teixeira et al., “Masticatory Deficiency as a Risk Factor for Cognitive Dysfunction,” Int. J. Med. Sci., vol. 11, no. 2, pp. 209–214, 2014. [164] K. Sakamoto, H. Nakata, M. Yumoto, and R. Kakigi, “Effect of Mastication on Human Brain Activity,” Anti-Aging Med., vol. 7, no. 13, pp. 153–160, 2010. [165] H. L. Hollingworth, “Chewing as a Technique of Relaxation,” Science, vol. 90, no. 2339, pp. 385–387, Oct. 1939. [166] K. Kubo et al., “Chewing under restraint stress inhibits the stress-induced suppression of cell birth in the dentate gyrus of aged SAMP8 mice,” Neurosci. Lett., vol. 466, no. 3, pp. 109–113, Dec. 2009. [167] K. Kubo, M. Iinuma, H. Chen, K. Kubo, M. Iinuma, and H. Chen, “Mastication as a Stress-Coping Behavior,” BioMed Res. Int. BioMed Res. Int., vol. 2015, 2015, p. e876409, May 2015. [168] H. Lavretsky et al., “A pilot study of yogic meditation for family dementia caregivers with depressive symptoms: effects on mental health, cognition, and telomerase activity,” Int. J. Geriatr. Psychiatry, vol. 28, no. 1, pp. 57–65, Jan. 2013. [169] K.-T. So and D. W. Orme-Johnson, “Three randomized experiments on the longitudinal effects of the Transcendental Meditation technique on cognition,” Intelligence, vol. 29, no. 5, pp. 419–440, Sep. 2001. [170] “Relaxation and memory training in 39 elderly patients,” Am. J. Psychiatry, vol. 141, no. 6, pp. 778–781, Jun. 1984. [171] C. Dunn-Lewis et al., “The Effects of a Customized Over-the-Counter Mouth Guard on Neuromuscular Force and Power Production in Trained Men and Women:,” J. Strength Cond. Res., vol. 26, no. 4, pp. 1085–1093, Apr. 2012. [172] S. N. Drum, A. M. Swisher, C. A. Buchanan, and L. Donath, “Effects of a Custom Bite-Aligning Mouthguard on Performance in College Football Players:,” J. Strength Cond. Res., vol. 30, no. 5, pp. 1409–1415, May 2016. [173] L. D. J. Fiederer, “Multiscale Electrophysiological Neuroimaging: High-Resolution Forward Modeling,” Diploma Thesis, Albert-Ludwigs-Universität Freiburg, Freiburg, 2012. [174] R. D. Penn, S. K. Hilal, W. J. Michelsen, E. S. Goldensohn, and J. Driller, “Intravascular intracranial EEG recording,” J. Neurosurg., vol. 38, no. 2, pp. 239–243, Feb. 1973. [175] T. J. Oxley et al., “Minimally invasive endovascular stent-electrode array for high-fidelity, chronic recordings of cortical neural activity,” Nat. Biotechnol., vol. 34, no. 3, pp. 320–327, Mar. 2016. [176] R. K. Sefcik, N. L. Opie, S. E. John, C. P. Kellner, J. Mocco, and T. J. Oxley, “The evolution of endovascular electroencephalography: historical perspective and future applications,” Neurosurg. Focus, vol. 40, no. 5, p. E7, May 2016. [177] S. Wagner et al., “Investigation of tDCS volume conduction effects in a highly realistic head model,” J. Neural Eng., vol. 11, no. 1, p. 016002, Feb. 2014. [178] B. C. Raducanu et al., “Time multiplexed active neural probe with 678 parallel recording sites,” in 2016 46th European Solid-State Device Research Conference (ESSDERC), 2016, pp. 385–388. [179] A. S. Herbawi, L. Kiessner, O. Paul, and P. Ruther, “High-density CMOS neural probe implementing a hierarchical addressing scheme for 1600 recording sites and 32 output channels,” in 2017 19th International Conference on Solid-State Sensors, Actuators and Microsystems (TRANSDUCERS), 2017, pp. 20–23. [180] H. Lindén et al., “Modeling the Spatial Reach of the LFP,” Neuron, vol. 72, no. 5, pp. 859–872, Dec. 2011. [181] G. Buzsáki, “Large-scale recording of neuronal ensembles,” Nat. Neurosci., vol. 7, no. 5, p. nn1233, Apr. 2004. [182] V. Caune et al., “Evaluating dipolar source localization feasibility from intracerebral SEEG recordings,” NeuroImage, vol. 98, pp. 118–133, Sep. 2014. [183] M. Dümpelmann, T. Ball, and A. Schulze-Bonhage, “sLORETA allows reliable distributed source reconstruction based on subdural strip and grid recordings,” Hum. Brain Mapp., vol. 33, no. 5, pp. 1172– 1188, May 2012. [184] S. Le Cam et al., “SEEG dipole source localization based on an empirical Bayesian approach taking into account forward model uncertainties,” NeuroImage, vol. 153, pp. 1–15, Jun. 2017. [185] D. Cosandier-Rimélé, G. Ramantani, J. Zentner, A. Schulze-Bonhage, and M. Dumpelmann, “A realistic multimodal modeling approach for the evaluation of distributed source analysis: application to sLORETA,” J. Neural Eng., vol. 14, no. 5, p. 056008, 2017. [186] N. von Ellenrieder, L. Beltrachini, and C. H. Muravchik, “Electrode and brain modeling in stereo-EEG,” Clin. Neurophysiol., vol. 123, no. 9, pp. 1745–1754, Sep. 2012.

33 References

34 To be submitted to the Journal of

V. SPATIAL AND FREQUENCY-DOMAIN CHARACTERISTICS OF INTRACRANIALLY-MEASURED EMG

J. Lahr a-c, L.D.J. Fiederer c-g, O. Iljina c,d, A. Aertsen e,f, A. Schulze-Bonhage c,f,g, T. Ball c,f,g a Department of Psychiatry and Psychotherapy, Medical Center – University of Freiburg, Freiburg, GER b Freiburg Brain Imaging, Medical Center – University of Freiburg, Freiburg, GER c Translational Neurotechcology Lab, Epilepsy Center, Medical Center – University of Freiburg, Freiburg, GER d GRK 1624 “Frequency Effects in Language”, Freiburg, GER e Neurobiology and Biophysics, Faculty of Biology, University of Freiburg, GER f Bernstein Center Freiburg, GER g BrainLinks-BrainTools Cluster of Excellence, University of Freiburg, GER

Corresponding author: Lukas Fiederer, Engesserstr. 4, 5th floor, EEG lab AG Ball, 79108 Freiburg, Germany. [email protected], Phone: +4976127093283

35 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

ABSTRACT Objective: Electrocorticography (ECoG) plays an increasingly important role in neuroscientific research and might be used as a control signal in brain-machine interfaces (BMI). While it is clear that electromyographic (EMG) activity of extracranial origin reaches intracranial recordings, the exact spatial and frequency-domain characteristics of such effects are not known. Methods: We examined natural chewing as a source of EMG activity. Chewing-related effects (ChREs) and, for comparison, physiological brain activity during several tasks in epilepsy patients under invasive pre-neurosurgical EEG monitoring were compared with respect to their spatial and spectral characteristics. Results: We found differences in the spatial distribution and spectral features between ChRE and physiological brain activity. ChREs had a higher maximal relative power and a higher peak frequency. Further, we found indications that the silicone grid, on which the ECoG electrodes are mounted, electrically shields the recordings. Conclusion: Chewing-related EMG reaches the brain with amplitudes significantly stronger than those of typical examples of physiological brain activity. The silicone grid attenuates extracranial-to-ECoG signal propagation. Significance: The present work is the first systematic evaluation of the exact spatial and frequency-domain characteristics of EMG effects in ECoG based on a large sample of patients, and describes how they differ from those of brain activity. The silicone component of ECoG grids may be an important feature to ensure high signal quality in the presence of strong EMG activity in a range of BMI applications. Detailed knowledge of EMG properties may help in designing both electrodes and EMG-reducing algorithms to optimize ECoG signal quality.

Key Words: Brain-machine interfaces, brain-computer interfaces, electrocorticography, electromyography, spectral analysis, electrode design. Highlights: First extensive description of chewing EMG artifacts in ECoG recordings Chewing EMG reaches the brain with stronger amplitudes than typical brain activity EMG artifacts and brain activity have distinct characteristics in ECoG recordings Silicone ECoG grids attenuate extracranial-to-ECoG signal propagation Shielding may be an important feature to ensure signal quality in BMI applications

36 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

1. INTRODUCTION Intracranially-measured electrocorticography (ECoG) is increasingly being used in neuroscientific research because of its considerably superior signal quality over non-invasive electroencephalography (EEG) (Ball et al., 2009c; Schalk, 2010). In addition, brain-machine interfaces (BMIs) for paralyzed patients have recently emerged as a new field for clinical application of ECoG (Ball et al., 2009c; Lahr et al., 2015; Leuthardt et al., 2004, 2006; Pistohl et al., 2008; Schalk et al., 2007). Whereas artifact contamination has been extensively investigated in EEG (Dworetzky et al., 2010; Goncharova et al., 2003; Tong and Thankor, 2009), only few studies on artifacts in ECoG have appeared to date, such as those related to blinking (Ball et al., 2009a), saccades (Jerbi et al., 2009a; Kovach et al., 2011a), heart cycle (Kern et al., 2013) and electromyographic (EMG) activity (Fiederer et al., 2016; Liu et al., 2004a; Otsubo et al., 2008a). Together, these previous studies demonstrate that such artifacts have an impact on ECoG recordings. The previously-reported amplitudes of ECoG contamination by artifacts, however, were often much smaller than those of neural activity recorded using ECoG ((e.g., Ball et al., 2009a)), and they maximally reached those of the neural responses ((e.g., Kovach et al., 2011a)). Hence, although it is well established that extracranial signals can enter the skull and affect ECoG signals, as also show in our preliminary work (Fiederer et al., 2016), the exact conduction pathways by which this extra-to-intracranial signal propagation takes place have remained unknown. Moreover, whether the amplitudes of contaminating signals might even become substantially larger than those of neural ECoG signals is still unclear. The present study was designed to address these hitherto unresolved issues. Prior to examining the possible pathways of artifact conduction, it is useful to consider the exact surgical procedure by which subdural grid implantation is performed (see also (see also Greenberg, 2010)): after the skin incision and a partial incision of the temporal muscle, the scalp flap is retracted to expose the skull. Then, typically between 2 and 4 burr holes 9 to 15 mm in diameter are made using a skull drill. The burr holes are connected by saw lines with a craniotome before the bone flap is removed, resulting in a saw line of up to a few mm in width. After electrode placement, the bone flap is returned to its previous location and sutured in place. While a single burr hole would be sufficient for craniotomy and electrode implantation, additional holes are usually drilled for cable connections. Due to this surgical procedure, there are two different ways by which extracranial signals may enter the intracranial cavity: first, through craniotomy defects (burr holes, saw lines), and second, through the intact skull tissue. In the first case, intracranial artifacts would be expected to occur in a relatively focalized manner, underneath the craniotomy defects, constituting a ‘reverse breach effect’ (see the personal communication of J. Gotman cited in (Otsubo et al., 2008b)) that may be difficult to distinguish from the typically likewise focalized ECoG responses of neural origin. In the second scenario, intracranial contamination from extracranial sources would be expected to show a spatially widespread distribution, independent of the individual configuration of the craniotomy defects, due to the spatial filtering properties of the skull (Neuling et al., 2012; Paul L Nunez and Srinivasan, 2006) – similarly to the well-known spatial blurring of EEG taking place in the opposite direction (Dannhauer et al., 2011; Lanfer et al., 2012b). Two questions arise from the considerations above: (i) Are the intracranial manifestations of EMG spatially focalized or diffuse? (ii) Does their overall topography depend on the configuration of the craniotomy defects in the individual patient, as would be expected if volume conduction through these defects constituted the major pathway along which EMG enters the skull? (iii) How does iEMG differ from event realted brain activity?

37 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

To address these questions we investigated intracranial manifestations of chewing-related effects (ChRE) in patients with subdural grids and different configurations of craniotomy defects. We focused on ChREs as they constitute an important artifact to be accounted for in the development of clinical BMI applications for particularly in tetraplegia patients, because here self-feeding is a vitally important application that needs to be covered by future BMI technology (Velliste et al., 2008). 2. METHODS 2.1 Intracranial EEG 2.1.1 Patients Twelve patients participated in the study after having provided written informed consent. All patients had subdurally implanted ECoG grids as part of their evaluation for neurosurgical treatment of medically-intractable epilepsy (Table I). Depending on the individual clinical requirements, ECoG grids were implanted for a period of 5 to 10 days. The electrode contacts were mounted on a flexible silicone substrate (Ad-Tech, Racine, WI) at a 10-mm center-to- center inter-electrode distance and made of stainless-steel or platinum discs 4 mm in diameter. Linearly-arranged strip electrodes or penetrating depth electrodes in the hippocampus (1-mm diameter, 10 contacts with a 5-mm contact-to-contact distance) were also implanted in some patients, but the effects in these electrodes were not the object of the present study. Pre- neurosurgical diagnostics were the sole requirement for the type and placement of all electrodes. The study was approved by the University Clinic’s Ethics Committee.

Table 1: Patient Overview

grid pat. age sex diagnosis / lesion onset zone data localization

temporal-lobe epilepsy r.: fronto- S1 34 f (r.), frontal Ch parietal FCD

frontal-lobe epilepsy (r.), S2 50 f r.: frontal frontal Ch FCD

frontal-lobe epilepsy (r.), S3 21 f r.: frontal frontal Ch FCD r. frontal

fronto-central epilepsy S4 48 f (l.), l.: frontal frontal Ch,Sp FCD

epilepsy, status post trauma, l.:fronto- S5 54 m frontal Ch,Sp substance defect, l. temporal frontal

frontal-lobe epilepsy (l.), l.: fronto- S6 40 m l. frontal Sp FCD parietal

temporal-lobe epilepsy S7 31 m (l.) , l.: temporal l. temporo-mesial Mu FCD temp.-med.-ant.

38 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

temporal-lobe epilepsy S8 18 m (l.), l.: temporal l. temporal Mu FCD l. temporal

temporal-lobe epilepsy l. intrahippocampal, S9 29 m (l.), l.: temporal Mu temporo-polar FCD l. temporo-polar

frontal-lobe epilepsy (l.), S10 27 m l.: frontal l. frontal Fi FCD l. frontal

parieto-occipital-lobe l.: temporo- S11 57 m epilepsy, l. parietal Fi parietal hippocampal sclerosis

fronto-temporal epilepsy l.: fronto- S12 45 f (l.), fronto-basal-mesial Fi parietal FCD l. frontal

pat. : patient; data.: data sets analyzed (Ch: chewing; Mu: Music perception; Fi: Finger movement; Sp: Speech production); m: male, f: female; FCD: focal cortical dysplasia; r.: right; l.: left, temp.: temporal, med.: medial, ant.: anterior

2.1.2. Data acquisition ECoG was recorded at a sampling rate of 1024 Hz using a clinical AC EEG-system (IT-Med, Usingen, Germany) in all patients except S11, in whom the sampling rate was 256 Hz. Digital video, synchronized with neural data, was recorded at 25 frames per second at VGA resolution. EEG (10-20-system (Klem et al., 1999)) was recorded simultaneously in all patients in whom ChRE were analyzed. We excluded channels with technical recording problems (e.g., broken wires) from further analyses. 2.2 Chewing-related Effects 2.2.1 Trial selection for ChRE analysis The trial selection for ChRE analysis has been described elsewere (Fiederer et al., 2016). In brief, the onset and offset of chewing-related EMG bursts were marked for each chewing event (c.f. Fig. 1a for an example), and their arithmetic mean was defined as the 0-s time point for each trial. A total of 1652 trials was acquired (S1: 551 trials; S2: 438 trials; S3: 264 trials; S4: 252 trials; S5: 147 trials).

39 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

Fig. 1. Examples of ChRE in simultaneously-recorded EEG and ECoG. (a) Ongoing EEG from channels C4, T4 and F8 of S1 together with the data from three ECoG channels (F6, G7 and H8) simultaneously recorded in the same patient. The time epoch of a chewing event, as marked for the analysis, is indicated by a blue box. The EEG traces reveal distinct EMG bursts, and close inspection of the ECoG channel H8 also reveals chewing- related high-frequency bursts, albeit of much lower amplitude than in EEG. The three lower traces show the high-pass-filtered ECoG signal from the same channels, enhancing the visibility of chewing-related high- frequency bursts. (b) Time-resolved chewing-related relative spectral power changes in the EEG channel T4 from (a) involving a broad frequency range. Median time points of the preceding and following chewing event are indicated above the time-frequency plot (error bars: inter-quartile range). Color encodes the logarithmic power change relative to the baseline (see Methods for further details). (c) Same as (b) but for the ECoG channel H8 from (a), which was on the right anterior Sylvian fissure. Figure adapted from (Fiederer et al., 2016) 2.2.2 Data analysis The EEG and ECoG data were separately re-referenced to a common average reference (CAR), as it is common in EEG and ECoG studies (Ball et al., 2009b; Canolty et al., 2007; Crone et al., 2006; Towle et al., 2007). Trials were cut from the continuous data from -2 s to 2 s with respect to the 0-s time point in the chewing event. Sliding-window Fast Fourier Transformations (FFTs) were performed with a window length of 250 ms and a step width of 24.41 ms for the 1024 Hz data, and with a window size of 250 ms and a step width of 23.44 ms for the 256 Hz recordings. For the chewing-related data, a baseline period (200 ms) centered between consecutive chewing events was defined. For the control data representing neural activity used for comparison of the ChRE with the typical profiles of neural activity (see Table I), a baseline period was selected at the beginning of the corresponding data epochs. Statistical analysis was performed using a two-tailed sign test. False discovery rate (FDR) correction suitable for correlated p-values (neighboring time and frequency bins) (Benjamini and Hochberg, 1995) was performed with a q-level of 0.001 to control for multiple testing. 2.2.3 Relation of electrodes to craniotomy defects and anatomical landmarks To determine the relation of the implanted electrodes to craniotomy defects (burr holes and saw lines) and to major anatomical landmarks (lateral and central sulci), we used computer tomography (CT)-acquired images (Fig. 2), X-ray data (Fig. 5), complemented by data from magnetic resonance imaging (MRI) (Figs. 2, 3, 7), all acquired after electrode implantation in the individual patients.

40 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

Fig. 2. CT imaging of craniotomy defects. Axial CT images taken after subdural electrode implantation in S4 are shown. Arrowheads indicate the burr holes (red), saw lines (blue), and individual electrode contacts (green) (electrodes marked only in (c)). Examples of subdural air are indicated by yellow asterisks. In (b) and (c), the low signal intensity of the saw lines indicates that here the saw lines were filled with air (in contrast to their higher signal intensity as in (d) when filled with liquid).

2.3 Neural-Activity-Related Effects 2.3.1. Electrical cortical stimulation Electrical cortical stimulation was performed to identify eloquent cortex using a nerve stimulator (NS 60, Inomed, Emmendingen, Germany) in a constant-current mode. Trains of 7- s duration consisting of 20 ms pulses of alternating-polarity square waves of 200 µs each were applied. Stimulation intensity was increased in steps of 1-2 mA up to 15 mA or until the observation of a sensory, speech-related, or motor effect, whichever occurred first. See (Ruescher et al., 2013) for more details. 2.3.2 Comparison with neural activity Data obtained from several different other conditions were used for comparison of ChRE (S1- 5) with activity of neural origin: S7, S8 and S9 participated in a music perception task where the patients were presented with six-chord sequences (Sammler et al., 2013, 2009); S10, S11 and S12 performed index-finger flexions (Ball et al., 2004) and in S4, S5 and S6, speaking-

41 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

related activity during non-experimental, real-life communication was analyzed. This sample with different conditions was used to extract general spectral properties of the ECoG responses. Because of the different task conditions, though, a single time window of fixed length cannot be expected to match the timing of all neural responses across these different conditions. Therefore, for computing spectral responses related to both chewing and the neural data in the control conditions, we selected time windows fitting to the duration of the responses by computing the median over all time windows of durations in the interval of [25, 500] ms and with all possible starting points in the 4 s data epochs. The number of frequency bins with significant power increases was determined for each case (FDR-corrected for all tests in the optimization procedure, see above) and the parameter combination (window length and the starting point) which elicited the highest number of significant frequency bins was selected for further analysis. To characterize the spectral profiles, we calculated the maximal power (MP) and frequency of the maximal power (FMP) for each channel as illustrated for data from one patient and electrode (electrode D6 from patient S4) in Fig. 4c. For the MP, we compared the maximal values per patient (i.e., spread in Fig. 4d are across patients) and for the FMP, we compared all channels with significant responses in the ChRE and control conditions (i.e., spread in Fig. 4e are across channels). Results with FMP across patients (not shown) were very similar to that in MP (Fig. 4d) and supported the same conclusions. To determine the spatial distribution of ECoG effects, relative power changes in a time window of 250 ms centered on the marked chewing events and in a frequency range of 32- 400 Hz were averaged, and effects were interpolated across the topography of the electrode grid. Correlation coefficients (Spearman's ρ) were calculated using the relative power and the smallest distance from electrodes to the border of the grid, for both ChRE and the responses in the control data (Figs. 4a,b, respectively). 3. RESULTS 3.1 Chewing-related effects (ChREs) are clearly present in the ECoG As already shown in our preliminary work (Fiederer et al., 2016), ChREs are clearly present in the ECoG. From the examples of simultaneously recorded ChREs in EEG and ECoG shown in Fig. 1 it is apparent that chewing-related bursts of high-frequency activity were clearly visible in the ongoing EEG recordings in all patients (here shown for S1 in Fig. 1a, top three traces). Similar high-amplitude ChREs were never observed in the simultaneously- recorded ECoG (Fig. 1a, middle three traces). However, after high-pass filtering (cutoff=100Hz), high-frequency chewing-related bursts with peak-to-peak amplitudes of approx. 40 µV became visible in the ongoing ECoG (Fig. 1a, three bottom traces). In some ECoG channels, close inspection of the unfiltered ongoing ECoG revealed relatively low- amplitude chewing-related bursts (Fig. 1a, highlighted by blue boxes). Chewing related EEG spectral power showed strong broadband increases up to the recording’s Nyquist frequency (max. ca. 500 Hz, Fig. 1b), relative to baseline. ECoG spectral power showed a similar chewing-related pattern, albeit about one order of magnitude weaker (Fig. 1c). Following statistical testing and FDR correction (p<0.001), 406 out of 410 (99%) electrodes showed significant ChRE power increases (all electrodes in S1, S3, and S4, and all but two electrodes in S2 and S5). 3.2 The topography of intracranial ChREs and ECoG responses of neuronal origin In all cases, distribution of ChRE spectral power modulations were spatially widespread over the grid and aligned better with grid geometry than with brain anatomy. In all patients we found the maximal power in the anterolateral corner of the grid. We consistently found

42 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

intermediate power at other positions close to the edge of the grid, while the smallest power increases were located in the center of the grid. These widespread effects were not focalized to oro-facial electrodes (Fig. 3b as defined by electrical cortical stimulation) and extended without any interruption over the anatomical borders of the lateral sulcus (LS) and central sulcus (CS) (Fig. 3a). Physiological neuronal activity was, in contrast, more localized, and it showed spatial selectivity. The finger movement tasks elicited effects in the ESM-identified hand sensorimotor cortex, speech production in the articulatory cortex, and music perception in superior temporal cortex implicated in auditory perception. The fact that gamma-band ChRE had higher amplitudes at the edges than in the center of the ECoG grids is further demonstrated in Fig. 4a, where the distribution of power modulations is shown as a function of electrode distance to the nearest edge of the grids. Correlation coefficients (Spearman's ρ) between relative power and distance to the edge were S1: r=- 0.469; S2: r=-0.580; S3: r=-0. 588; S4: r=-0. 303; S5: r=-0.467 (ρ being negative as small distances were associated with high power).

Fig. 3. A comparison of the topographic distribution of chewing- and speech-related spectral responses (S4). (a) Time-frequency spectra of chewing-related responses. The course of the lateral sulcus (LS) and the central sulcus (CS) are depicted by white lines. Note the spatially widespread distribution bridging the LS. (b) ECoG grid position in relation to the brain surface obtained from individual patient’s MRI data. The position of the mouth sensorimotor cortex as determined by direct cortical electrical stimulation mapping (ESM) is indicated in orange. (c) Time-frequency spectra obtained in the same patient during speech production show spatially focalized responses in areas predominantly in or adjacent the ESM-defined mouth sensorimotor cortex (orange outline). All conventions as in (a). In contrast to the ChREs in (a), speech production-related responses were most prominent in frequencies below 200 Hz and showed an accompanying power decrease in the lower frequencies (dark blue, approx. 8 to 32 Hz). (a) & (b) adapted from (Fiederer et al., 2016).

3.3 Amplitudes of intracranial ChREs To quantify the importance of chewing-related effects (i.e., mainly EMG artifacts of extracranial origin, see Discussion) on both ECoG and EEG signals, the ratio of the maximal chewing-induced power (MP) increase in ECoG and EEG was determined for all patients. The power increases were stronger in the EEG, on average by a factor of approx. 25 (S1: 28.75; S2: 11.03; S3: 20.12; S4: 16.03; S5:49.70). The maximal relative power amplitudes (MP) of ChRE in the ECoG were, on average, larger by a factor of approx. 4 than those of neural ECoG effects (Fig. 4d). 3.4 Frequency profile of intracranial ChREs To characterize and compare the frequency profiles of ChRE and neural ECoG responses, the frequency with the maximal power was determined (FMP). Across patients, the FMP in the ECoG electrodes ranged between 52 and 372 Hz in the ChRE condition and between 32 and 152 Hz in the control data representing neural responses (Fig. 4e). The FMP derived from chewing-related spectra were generally higher than in the neural responses, but there was an

43 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

overlap in the frequency range between 52 and 152 Hz (Fig. 4e). A further distinctive feature between ChRE and neural effects was that broadband gamma increases were only in the latter case typically accompanied by power decreases in the range between 8 and 32 Hz (e.g., the dark blue area in the grid electrode F4 in Fig. 3c). 3.5 The intracranial ChRE topography is largely independent of the configuration of craniotomy defects The positions of burr holes and saw lines were obtained from CT scans to determine the role of craniotomy-related skull defects (burr holes and saw lines) on the intracranial topography of ChRE (Fig. 2). Despite the very different positions of both burr holes and saw lines in patients S4 and S5, the topography of relative power changes in the gamma frequency range (32 to 400 Hz) remained remarkably constant (Fig. 5). The maximal relative power change in each patient was observed in the anterolateral corner of the grid, i.e., in a region which was situated directly underneath the belly of the temporal muscle, regardless of whether a burr hole or a saw line was present in this region (Fig. 5a) or not (Fig. 5c).

4. DISCUSSION 4.1 ChREs are clearly present in intracranial recordings This study presents the first systematic evaluation of the characteristics of non-ocular EMG effects in ECoG based on a larger sample of patients (in contrast to previous single-case studies, such as (Liu et al., 2004b; Otsubo et al., 2008b), and on ocular effects (Ball et al., 2009b; Jerbi et al., 2009b; Kovach et al., 2011b)). In all patients investigated, we found ChRE with a broad frequency distribution (from <25 to above 400 Hz) which were significant in nearly all (99%) of the inspected ECoG channels. The peak-to-peak amplitudes of the ChRE in ongoing recordings were approx. 50µV and, thus, in the order of magnitude of the ictal muscle activity reported by (Otsubo et al., 2008b). 4.2 ChREs are mainly EMG activity Based on the results of this study, it appears most plausible to assume that ChREs are EMG activity of the masticatory muscles, For a number of reasons it is very improbable that ChREs result from neural activity related to the act of chewing. Firsly, Malandraki and colleagues as well as Onozuka and colleagues have shown using functional magnetic resonance imaging (fMRI) that BOLD signal changes related to chewing, tongue tapping, or swallowing can be found focally in regions of the primary sensory and motor cortex (Malandraki et al., 2009; Onozuka et al., 2002). This BOLD spatial response pattern is clearly different from the spatially widespread distribution of ChRE in our study, which extended smoothly over functional and structural boundaries (Fig. 3a). Second, the frequencies with the highest relative power of the ChRE were, in most cases, higher (> 150 Hz) than in any of the investigated examples of control responses representing neural activity (Fig. 4e). The spectral profile of the latter typically showed low-frequency suppression, together with gamma-band increases (Fig. 3c), a well-established feature of event-related neural population responses of the cortex (Crone et al., 1998; Pfurtscheller and Lopes da Silva, 1999), whereas ChRE failed to display this feature (Fig. 3a); see also below). Finally, in our data, ChRE power in EEG was approx. 20 times higher than in ECoG (Fig. 1), whereas the opposite would be expected for recordings of neural activity. Nevertheless, it is likely that a small focal neural signal is present albeit masked by high-amplitude extracranial EMG. Further work will be necessary to isolate this presumably weak neural signal, if possible at all.

44 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

4.3 ChREs involve a broad frequency range In previous reports on surface EMG of facial and masticatory muscles, EMG was characterized by signal increases in a broad frequency range from several Hz to over 500 Hz, with peak frequencies between 80 and 160 Hz for the temporal muscle ((Boxtel et al., 1983): peak frequency ~130 Hz; (Palla and Ash Jr, 1981): ~80 Hz ; (Yuen et al., 1989): ~160 Hz), matching well the ChRE spectra we observed extracranially in the present study (Fig. 1). As extra-to-intracranial volume conduction is largely frequency-independent (Paul L. Nunez and Srinivasan, 2006), we expected a similar spectral profile in intracranial data. Consistent with this expectation, we found ChRE in ECoG to involve a broad frequency range from <25 Hz to above 400 Hz (limited by the anti-aliasing filters and sampling rate of our recordings). Across channels, the frequency with the maximal ECoG ChRE power increase was observed in the range of approx. 100–350 Hz (Fig. 4). However, in a study (Boxtel et al., 1983) which analyzed the masseter and temporalis muscles, as well as five other facial muscles in terms of their spectral properties, the temporal-muscle EMG spectrum was found to be atypical in comparison with the other muscles in that it had an exceptionally high peak frequency, whereas the peak frequencies of other facial muscles (e.g., buccinator or frontalis) were below 50 Hz. Thus, artifacts which are, e.g., related to facial expressions or speaking and hence originate predominantly from the mimic muscles may be also present in frequency bands other than those characterized in our study.

Fig. 4. Comparison of topographic and spectral profiles of ChRE and neural ECoG responses. Note that the relative power is used in all subfigures. Chewing-related (a), but not neural effects (b) showed the highest amplitudes at the edge of the electrode grids. (a,b) The amplitude of chewing-related relative power decreased as a function of electrode distance to the edges of the grids (S1). This was not the case for ECoG responses of neural origin (b, speech production of S4 as a representative example). (c) The maximal power (MP) and the frequency with the maximal power (FMP) were determined from the median spectral power in a given time window separately for each channel for the ChRE and for the control ECoG data representing neural responses (NR) separately. Boxplots are shown in (d) for the MP of neural ECoG responses and ChRE and in (e) for the frequencies of the FMP. Red vertical lines indicate the median across patients (d) and channels (e) (see methods for further details), the box margins the interquartile range (IQR), whiskers extend to the most extreme value within box margins extended by 1.5 times the IQR, data points outside this range are indicated by red crosses.

4.4 ChREs spatial patterns are largely independent of craniotomy defects The basic spatial pattern of gamma-band ChRE in our study was observed largely independent of the individual configurations of craniotomy defects (Fig. 5). We found only

45 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

limited evidence for a pronounced ‘reverse breach’ effect with focalized responses close to burr holes that might be mistaken for typically focalized neural event-related effects as shown in Fig. 3c and in many previous ECoG studies (Aoki et al., 1999; Crone et al., 1998, 2001a). Nevertheless, increased extra-to-intracranial signal conduction must be expected below craniotomy defects as long as they are not filled by low-conductance material (air, silicone- embedded electrode wires), which is at least as insulating as the surrounding skull. Some of our observed details of the ChRE-related intracranial gamma-band topographies may, indeed, be due to such effects, e.g., the increased power at the intersection of the posterior edge of the subdural grid and a saw line in S4 (marked by a white arrow in Fig. 5d). Notably, such cases were restricted to the edge of the grid and did not extend along the whole course of the saw line. This independence of craniotomy defects in particular, and the intracranial spatial distribution of ChREin general, may be explained by a model summarized in Fig. 5e: EMG activity from masticatory muscles passes through both craniotomy defects and intact parts of the skull and around the insulating substrate of the ECoG grid, which leads to strong effects near the edges of the grid. The overall spatial distribution is smooth due to the dominating effect of currents passing through the skull with spatial low-pass filtering properties (Paul L Nunez and Srinivasan, 2006). The strongest ChRE in the anterolateral corner of the ECoG grids in all investigated patients can be explained by its proximity to the belly of the temporal muscle (Fig. 5a–d), one of the main active muscles during chewing. In the opposite direction (intra- to-extracranial, i.e. from neural activity to EEG), previous simulations indicated that substantial spatial distortions of scalp surface EEG are to be expected due to the fact that currents have to pass around the insulating material of the silicone ECoG grid (Lanfer et al., 2012a; Paul L. Nunez and Srinivasan, 2006; von Ellenrieder et al., 2014; Zhang et al., 2006). To verify our assumptions, forward modeling based on detailed volume conduction simulations was performed, with results reported in the companion paper (Fiederer et al., 2016). Briefly, the simulations showed that the ChREs indeed mainly propagate directly through the skull. Furthermore, the simulations also confirmed that the silicone ECoG grids substantially attenuated and distorted the measured ChREs. Without craniotomy defects and without grid, ChREs were 21% stronger than in our measurements. By contrast, removing craniotomy defects and keeping the silicone grid decreased ChREs by only 6%. The silicone grid thus attenuated ChREs by 27%.

46 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

Fig. 5. The relation of craniotomy defects to intracranial ChRE topography. (a, c) Lateral X-ray with superimposed positions of implanted electrodes (blue), burr holes (white dashed discs), saw lines (white dashed lines), and the temporal muscle (red) with the temporal line (red dashed line) as its origin and the coronoid process of the mandibular bone (red asterisk) as its insertion. The variation in transparency reflects the thickness of the temporal muscle, which increases toward the coronoid process. (b, d) Intracranial topography of ChRE in the gamma frequency range (32 to 400 Hz). Electrode positions are marked with black circles. The saw lines and burr holes are indicated by white dashed lines and discs, and the lateral (LS) and central sulci (CS) are indicated by continuous white lines, respectively. The white arrow in (d) indicates a power increase on the edge of the grid, possibly caused by the saw line crossing the edge of the grid at this position. (e) A summary illustration of extracranial-to-ECoG signal propagation in a schematic anterior view of the brain and the different elements of the head volume conductor. The electrode grid (green) acts as an isolator forcing electric currents to pass around its edges. As the basic spatial pattern of chewing-related signals in the ECoG was largely independent of the individual configuration of craniotomy defects (a - d) and as indicated by the modeling results, the direct pathway through the bone (solid red arrows) plays an important role for extra-to- intracranial volume conduction, in addition to the pathway through the craniotomy defects in the skull (dashed red arrow).

47 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

5. CONCLUSION By comparing intracranially-recorded chewing-related events (ChREs) with physiological neural activity underlying several tasks in epilepsy patients, we could show that ChREs are not of neuronal origin but generated by EMG activity. The topographies and frequency characteristics of the ChREs were markedly differed from neural activity and similar to those of reported EMG activity. The differences in topography and preferred frequencies that we found are useful for further development of methods for EMG removal from ECoG. For instance, spatial high-pass filtering could be used to separate the focalized neural from the widespread EMG components. The silicone grids used in ECoG measurements produced an interesting shielding effect, which has implications for the design of future implanted BMI recording devices. Advantages of such a silicone grid to improve BMI decoding accuracies, for example with respect to arm movements during self-feeding, could be further investigated with our data. The possibility that ChREs described in this study are strong enough to influence ongoing brain activity is investigated in the companion paper (Fiederer et al., 2016). ACKNOWLEDGMENT Funding: This work was supported by the German Federal Ministry of Education and Research (BMBF) grants 01GQ0420 to BCCN Freiburg, 0313891 GoBio, 16SV5834 NASS, 01GQ1510 OptiStim and German Research Foundation (DFG) grant EXC 1086 BrainLinks- BrainTools. REFERENCES Aoki, F., Fetz, E.E., Shupe, L., Lettich, E., Ojemann, G.A., 1999. Increased gamma-range activity in human sensorimotor cortex during performance of visuomotor tasks. Clin. Neurophysiol. Off. J. Int. Fed. Clin. Neurophysiol. 110, 524–537. Ball, T., Kern, M., Mutschler, I., Aertsen, A., Schulze-Bonhage, A., 2009a. Signal quality of simultaneously recorded invasive and non-invasive EEG. NeuroImage 46, 708–16. doi:10.1016/j.neuroimage.2009.02.028 Ball, T., Kern, M., Mutschler, I., Aertsen, A., Schulze-Bonhage, A., 2009b. Signal quality of simultaneously recorded invasive and non-invasive EEG. NeuroImage 46, 708–716. doi:10.1016/j.neuroimage.2009.02.028 Ball, T., Nawrot, M.P., Pistohl, T., Aertsen, A., Schulze-Bonhage, A., Mehring, C., 2004. Towards an implantable brain-machine interface based on epicortical field potentials. Biomed Tech 49, 756–759. Ball, T., Schulze-Bonhage, A., Aertsen, A., Mehring, C., 2009c. Differential representation of arm movement direction in relation to cortical anatomy and function. J. Neural Eng. 6, 16006. doi:10.1088/1741- 2560/6/1/016006 Benjamini, Y., Hochberg, Y., 1995. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. J. R. Stat. Soc. Ser. B Methodol. 57, 289–300. Boxtel, A. van, Goudswaard, P., Molen, G.M. van der, Bosch, W.E. van den, 1983. Changes in electromyogram power spectra of facial and jaw-elevator muscles during fatigue. J. Appl. Physiol. 54, 51–58. Canolty, R.T., Soltani, M., Dalal, S.S., Edwards, E., Dronkers, N.F., Nagarajan, S.S., Kirsch, H.E., Barbaro, N.M., Knight, R.T., 2007. Spatiotemporal dynamics of word processing in the human brain. Front. Neurosci. 1, 185–196. doi:10.3389/neuro.01.1.1.014.2007 Crone, N.E., Boatman, D., Gordon, B., Hao, L., 2001a. Induced electrocorticographic gamma activity during auditory perception. Brazier Award-winning article, 2001. Clin. Neurophysiol. Off. J. Int. Fed. Clin. Neurophysiol. 112, 565–582. Crone, N.E., Miglioretti, D.L., Gordon, B., Sieracki, J.M., Wilson, M.T., Uematsu, S., Lesser, R.P., 1998. Functional mapping of human sensorimotor cortex with electrocorticographic spectral analysis. I. Alpha and beta event-related desynchronization. Brain 121, 2271–2299. doi:10.1093/brain/121.12.2271 Crone, N.E., Sinai, A., Korzeniewska, A., 2006. High-frequency gamma oscillations and human brain mapping with electrocorticography. Prog. Brain Res., Event-Related Dynamics of Brain Oscillations 159, 275–295. Dannhauer, M., Lanfer, B., Wolters, C.H., Knösche, T.R., 2011. Modeling of the human skull in EEG source analysis. Hum. Brain Mapp. 32, 1383–1399. doi:10.1002/hbm.21114 Dworetzky, B., Herman, S., Tatum, W.O., 2010. Artifacts of Recording, in: Niedermeyer, E., Silva, F.L. da (Eds.), Electroencephalography: Basic Principles, Clinical Applications, and Related Fields. Lippincott Williams & Wilkins, Philadelphia, pp. 239–266.

48 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

Fiederer, L.D.J., Lahr, J., Vorwerk, J., Lucka, F., Aertsen, A., Wolters, C.H., Schulze-Bonhage, A., Ball, T., 2016. Electrical Stimulation of the Human Cerebral Cortex by Extracranial Muscle Activity: Effect Quantification With Intracranial EEG and FEM Simulations. IEEE Trans. Biomed. Eng. 63, 2552–2563. doi:10.1109/TBME.2016.2570743 Goncharova, I.I., McFarland, D.J., Vaughan, T.M., Wolpaw, J.R., 2003. EMG contamination of EEG: spectral and topographical characteristics. Clin. Neurophysiol. 114, 1580–1593. doi:10.1016/S1388- 2457(03)00093-2 Greenberg, M.S., 2010. Handbook of Neurosurgery, 7th ed. Thieme, Stuttgart. Jerbi, K., Freyermuth, S., Dalal, S., Kahane, P., Bertrand, O., Berthoz, A., Lachaux, J.-P., 2009a. Saccade related gamma-band activity in intracerebral EEG: dissociating neural from ocular muscle activity. Brain Topogr. 22, 18–23. doi:10.1007/s10548-009-0078-5 Jerbi, K., Ossandón, T., Hamamé, C.M., Senova, S., Dalal, S.S., Jung, J., Minotti, L., Bertrand, O., Berthoz, A., Kahane, P., Lachaux, J.-P., 2009b. Task‐related gamma‐band dynamics from an intracerebral perspective: Review and implications for surface EEG and MEG. Hum. Brain Mapp. 30, 1758–1771. doi:10.1002/hbm.20750 Kern, M., Aertsen, A., Schulze-Bonhage, A., Ball, T., 2013. Heart cycle-related effects on event-related potentials, spectral power changes, and connectivity patterns in the human ECoG. NeuroImage 81, 178– 190. doi:10.1016/j.neuroimage.2013.05.042 Klem, G.H., Lüders, H.O., Jasper, H.H., Elger, C., 1999. The ten-twenty electrode system of the International Federation. The International Federation of Clinical Neurophysiology. Electroencephalogr. Clin. Neurophysiol. Suppl. 52, 3–6. Kovach, C.K., Tsuchiya, N., Kawasaki, H., Oya, H., Howard III, M.A., Adolphs, R., 2011a. Manifestation of ocular-muscle EMG contamination in human intracranial recordings. NeuroImage 54, 213–33. doi:10.1016/j.neuroimage.2010.08.002 Kovach, C.K., Tsuchiya, N., Kawasaki, H., Oya, H., Howard III, M.A., Adolphs, R., 2011b. Manifestation of ocular-muscle EMG contamination in human intracranial recordings. NeuroImage 54, 213–233. doi:10.1016/j.neuroimage.2010.08.002 Lahr, J., Schwartz, C., Heimbach, B., Aertsen, A., Rickert, J., Ball, T., 2015. Invasive brain–machine interfaces: a survey of paralyzed patients’ attitudes, knowledge and methods of information retrieval. J. Neural Eng. 12, 43001. doi:10.1088/1741-2560/12/4/043001 Lanfer, B., Röer, C., Scherg, M., Rampp, S., Kellinghaus, C., Wolters, C.H., 2012a. Influence of a Silastic ECoG Grid on EEG/ECoG Based Source Analysis. Brain Topogr. 26, 212–228. doi:10.1007/s10548-012-0251-0 Lanfer, B., Scherg, M., Dannhauer, M., Knösche, T.R., Burger, M., Wolters, C.H., 2012b. Influences of skull segmentation inaccuracies on EEG source analysis. NeuroImage 62, 418–431. doi:10.1016/j.neuroimage.2012.05.006 Leuthardt, E.C., Schalk, G., Moran, D., Ojemann, J.G., 2006. The emerging world of motor : a neurosurgical perspective. Neurosurgery 59, 1–14. doi:10.1227/01.NEU.0000221506.06947.AC Leuthardt, E.C., Schalk, G., Wolpaw, J.R., Ojemann, J.G., Moran, D.W., 2004. A brain-computer interface using electrocorticographic signals in humans. J. Neural Eng. 1, 63–71. doi:10.1088/1741-2560/1/2/001 Liu, H., Hild, K.E., Gao, J.B., Erdogmus, D., Principe, J.C., Chris Sackellares, J., 2004a. Evaluation of a BSS algorithm for artifacts rejection in epileptic seizure detection. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. IEEE Eng. Med. Biol. Soc. 1, 91–4. doi:10.1109/IEMBS.2004.1403098 Liu, H., Hild, K.E., Gao, J.B., Erdogmus, D., Principe, J.C., Sackellares, J.C., 2004b. Evaluation of a BSS algorithm for artifacts rejection in epileptic seizure detection, in: 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2004. IEMBS ’04. Presented at the 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2004. IEMBS ’04, pp. 91–94. doi:10.1109/IEMBS.2004.1403098 Malandraki, G.A., Sutton, B.P., Perlman, A.L., Karampinos, D.C., Conway, C., 2009. Neural activation of swallowing and swallowing‐related tasks in healthy young adults: An attempt to separate the components of deglutition. Hum. Brain Mapp. 30, 3209–3226. doi:10.1002/hbm.20743 Neuling, T., Wagner, S., Wolters, C.H., Zaehle, T., Herrmann, C.S., 2012. Finite-Element Model Predicts Current Density Distribution for Clinical Applications of tDCS and tACS. Front. Psychiatry 3. doi:10.3389/fpsyt.2012.00083 Nunez, P.L., Srinivasan, R., 2006. Electric Fields of the Brain: The Neurophysics of EEG. Oxford University Press, New York, USA. Nunez, P.L., Srinivasan, R., 2006. Electric Fields of the Brain: The Neurophysics of EEG. Oxford University Press. Onozuka, M., Fujita, M., Watanabe, K., Hirano, Y., Niwa, M., Nishiyama, K., Saito, S., 2002. Mapping Brain Region Activity during Chewing: A Functional Magnetic Resonance Imaging Study. J. Dent. Res. 81, 743–746. doi:10.1177/154405910208101104

49 Spatial and Frequency-domain Characteristics of Intracranially-measured EMG

Otsubo, H., Ochi, A., Imai, K., Akiyama, T., Fujimoto, A., Go, C., Dirks, P., Donner, E.J., 2008a. High- frequency oscillations of ictal muscle activity and epileptogenic discharges on intracranial EEG in a temporal lobe epilepsy patient. Clin. Neurophysiol. 119, 862–8. doi:10.1016/j.clinph.2007.12.014 Otsubo, H., Ochi, A., Imai, K., Akiyama, T., Fujimoto, A., Go, C., Dirks, P., Donner, E.J., 2008b. High- frequency oscillations of ictal muscle activity and epileptogenic discharges on intracranial EEG in a temporal lobe epilepsy patient. Clin. Neurophysiol. 119, 862–868. doi:10.1016/j.clinph.2007.12.014 Palla, S., Ash Jr, M.M., 1981. Effect of bite force on the power spectrum of the surface electromyogram of human jaw muscles. Arch. Oral Biol. 26, 287–295. doi:10.1016/0003-9969(81)90049-2 Pfurtscheller, G., Lopes da Silva, F.H., 1999. Event-related EEG/MEG synchronization and desynchronization: basic principles. Clin. Neurophysiol. 110, 1842–1857. doi:10.1016/S1388-2457(99)00141-8 Pistohl, T., Ball, T., Schulze-Bonhage, A., Aertsen, A., Mehring, C., 2008. Prediction of arm movement trajectories from ECoG-recordings in humans. J. Neurosci. Methods 167, 105–14. doi:10.1016/j.jneumeth.2007.10.001 Ruescher, J., Iljina, O., Altenmüller, D.-M., Aertsen, A., Schulze-Bonhage, A., Ball, T., 2013. Somatotopic mapping of natural upper- and lower-extremity movements and speech production with high gamma electrocorticography. NeuroImage 81, 164–177. doi:10.1016/j.neuroimage.2013.04.102 Sammler, D., Koelsch, S., Ball, T., Brandt, A., Elger, C.E., Friederici, A.D., Grigutsch, M., Huppertz, H.-J., Knösche, T.R., Wellmer, J., Widman, G., Schulze-Bonhage, A., 2009. Overlap of Musical and Linguistic Syntax Processing: Intracranial ERP Evidence. Ann. N. Y. Acad. Sci. 1169, 494–498. doi:10.1111/j.1749- 6632.2009.04792.x Sammler, D., Koelsch, S., Ball, T., Brandt, A., Grigutsch, M., Huppertz, H.-J., Knösche, T.R., Wellmer, J., Widman, G., Elger, C.E., Friederici, A.D., Schulze-Bonhage, A., 2013. Co-localizing linguistic and musical syntax with intracranial EEG. NeuroImage 64, 134–146. doi:10.1016/j.neuroimage.2012.09.035 Schalk, G., 2010. Can Electrocorticography (ECoG) Support Robust and Powerful Brain-Computer Interfaces? Front. Neuroengineering 3, 9. doi:10.3389/fneng.2010.00009 Schalk, G., Kubánek, J., Miller, K.J., Anderson, N.R., Leuthardt, E.C., Ojemann, J.G., Limbrick, D., Moran, D., Gerhardt, L. a, Wolpaw, J.R., 2007. Decoding two-dimensional movement trajectories using electrocorticographic signals in humans. J. Neural Eng. 4, 264–75. doi:10.1088/1741-2560/4/3/012 Tong, S., Thankor, N.V., 2009. Quantitative EEG analysis methods and clinical applications. Artech House, Boston. Towle, V.L., Hunter, J.D., Edgar, J.C., Chkhenkeli, S.A., Castelle, M.C., Frim, D.M., Kohrman, M., Hecox, K.E., 2007. Analysis of Human Subdural Recordings. J. Clin. Neurophysiol. 24, 205– 213. doi:10.1097/WNP.0b013e318039b191 Velliste, M., Perel, S., Spalding, M.C., Whitford, A.S., Schwartz, A.B., 2008. Cortical control of a prosthetic arm for self-feeding. Nature 453, 1098–1101. doi:10.1038/nature06996 von Ellenrieder, N., Beltrachini, L., Muravchik, C.H., Gotman, J., 2014. Extent of cortical generators visible on the scalp: Effect of a subdural grid. NeuroImage 101, 787–795. doi:10.1016/j.neuroimage.2014.08.009 Yuen, S.W.H., Hwang, J.C.C., Poon, P.W.F., 1989. EMG Power Spectrum Patterns of Anterior Temporal and Masseter Muscles in Children and Adults. J. Dent. Res. 68, 800–804. doi:10.1177/00220345890680050901 Zhang, Y., Ding, L., van Drongelen, W., Hecox, K., Frim, D.M., He, B., 2006. A cortical potential imaging study from simultaneous extra- and intracranial electrical recordings by means of the finite element method. NeuroImage 31, 1513–1524. doi:10.1016/j.neuroimage.2006.02.027

50 Published in IEEE Trans. Biomed. Eng. December 2016. Highlighted online & featured in editorial

VI. ELECTRICAL STIMULATION OF THE HUMAN CEREBRAL CORTEX BY EXTRACRANIAL MUSCLE ACTIVITY: EFFECT QUANTIFICATION WITH INTRACRANIAL EEG AND FEM SIMULATIONS

51 2552 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 63, NO. 12, DECEMBER 2016 Electrical Stimulation of the Human Cerebral Cortex by Extracranial Muscle Activity: Effect Quantification With Intracranial EEG and FEM Simulations

Lukas Dominique Josef Fiederer∗, Jacob Lahr, Johannes Vorwerk, Felix Lucka, Ad Aertsen, Carsten Hermann Wolters, Andreas Schulze-Bonhage, and Tonio Ball

Abstract—Objective: Electric fields (EF) of approx. 0.2 V/m have poles, the expected EF strength may reach amplitudes in the order been shown to be sufficiently strong to both modulate neuronal of 0.1–1 V/m. Conclusion: The cortical EF caused by natural activity in the cerebral cortex and have measurable effects on chewing could be large enough to modulate ongoing neural activity cognitive performance. We hypothesized that the EF caused by the in the cerebral cortex and influence cognitive performance. Signif- electrical activity of extracranial muscles during natural chewing icance: The present study lends first support for the assumption may reach similar strength in the cerebral cortex and hence might that extracranial muscle activity might represent an endogenous act as an endogenous modality of brain stimulation. Here, we source of electrical brain stimulation. This offers a new potential present first steps toward validating this hypothesis. Methods: explanation for the puzzling effects of gum chewing on cognition, Using a realistic volume conductor head model of an epilepsy which have been repeatedly reported in the literature. patient having undergone intracranial electrode placement and utilizing simultaneous intracranial and extracranial electrical Index Terms—Brain stimulation, electrical stimulation, elec- recordings during chewing, we derive predictions about the trocorticography, electroencephalography, electromyography, en- chewing-related cortical EF strength to be expected in healthy dogenous stimulation, finite element analysis, volume conductor individuals. Results: We find that in the region of the temporal head modeling.

I. INTRODUCTION Manuscript received December 15, 2015; revised April 11, 2016; accepted May 12, 2016. Date of publication July 19, 2016; date of current version Novem- NDOGENOUS modulation of neuronal activity through ber 18, 2016. This work was supported by the German Federal Ministry of E ephaptic coupling at the cellular level has received increas- Education and Research under Grants 01GQ0420 to BCCN Freiburg, Grant ing attention during the last years. Multiple groups could show 0313891 GoBio, Grant 16SV5834 NASS, Grant 01GQ1510 OptiStim, the Ger- man Research Foundation under Grant BrainLinks-BrainTools EXC 1086, Grant that the local electric fields (EF) generated by active neurons WO1425/2-1, and by the DFG Priority Program 1665 under Project WO1425/5- feed back onto themselves [1]–[15]. This ephaptic coupling is 1. Asterisk indicates corresponding author. especially effective for naturalistic EF [12]. EF strength in the ∗L. D. J. Fiederer is with the Intracranial EEG and Brain Imaging Lab, Epilepsy Center, Medical Center – University of Freiburg, Freiburg, Ger- order of magnitude of 0.2 V/m may be sufficient to elicit these many, the Neurobiology and Biophysics, Faculty of Biology, University effects [13]. Transcranial electric stimulation (TES) also influ- of Freiburg, Freiburg Germany, the BrainLinks-BrainTools Cluster of Ex- ences the EF of the brain [16] and has been shown to have an cellence, University of Freiburg, Freiburg, Germany, and with the Bern- stein Freiburg Center, Freiburg, Freiburg (e-mail: lukas.fiederer@uniklinik- impact on diverse brain functions [17]–[24], including working freiburg.de). memory and learning, at similar cortical EF strength as in the J. Lahr is with the Intracranial EEG and Brain Imaging Lab, Epilepsy endogenous case [21], [25]. Center, Medical Center – University of Freiburg, Freiburg, Germany, and also with the Department of Psychiatry and Psychotherapy as well as Freiburg Brain Besides neuronal activity, electrical muscle activity is another Imaging, Medical Center – University of Freiburg, Freiburg, Germany. source of endogenous EF [26], [27]. Particular strong muscle J. Vorwerkis with the Institute for Biomagnetism and Biosignalanalysis, Uni- activity close to the brain occurs during chewing. Interestingly, versity of Munster,¨ Munster,¨ Germany, and also with the Scientific Computing and Imaging (SCI) Institute, University of Utah, Salt Lake City, UT-84112, using different batteries of cognitive tests, it was shown that USA. cognitive performance is enhanced for 15–20 min after gum F. Lucka is with the Center for Medical Image Computing, University Col- chewing [28]. Chewing during the cognitive testing itself sig- lege London, London, England and with the Institute for Biomagnetism and Biosignalanalysis, University of Munster,¨ Munster,¨ Germany. nificantly reduced test performance [28]. These findings were A. Aertsen is with the Neurobiology and Biophysics, Faculty of Biology, previously explained by indirect effects, such as unspecific psy- University of Freiburg and also with the Bernstein Center Freiburg. chological arousal induced by the chewing activity. Here, we A. Schulze-Bonhage is with the Epilepsy Center, Medical Center–University of Freiburg, Freiburg, Germany, the BrainLinks-BrainTools Cluster of Excel- consider the alternative hypothesis that the cognitive effects of lence, University of Freiburg, Freiburg, Germany and also with the Bernstein gum chewing are at least in part a direct consequence of cor- Center Freiburg, Freiburg, Germany. tical electrical endogenous stimulation caused by the electrical T. Ball is with the Intracranial EEG and Brain Imaging Lab, Epilepsy Center, Medical Center – University of Freiburg, Freiburg, Germany, the activity of muscles during mastication. BrainLinks-BrainTools Cluster of Excellence, University of Freiburg, and also However, assessing the cortical EF caused by muscle activity with the Bernstein Center Freiburg. to be expected in healthy individuals is a challenging task. It is Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. not possible to directly measure the intracranial signal gener- Digital Object Identifier 10.1109/TBME.2016.2570743 ated by extracranial muscles in healthy individuals, as this would

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/

52 FIEDERER et al.:ELECTRICALSTIMULATIONOFTHEHUMANCEREBRALCORTEXBYEXTRACRANIALMUSCLEACTIVITY 2553

TABLE I PATIENT OVERVIEW

Pat. Age Sex Diagnosis/Lesion Grid Localization Seizure Onset

S1 34 f Temporal-lobe Fronto-parietal (R) Frontal epilepsy (R), FCD S2 50 f Frontal-lobe epilepsy Frontal (R) Frontal (R), FCD S3 48 f Fronto-central Frontal (L) Frontal epilepsy (L), FCD S4 21 f Frontal-lobe epilepsy Frontal (R) Frontal (R), FCD S5 54 m Epilepsy (L), Fronto-temporal (L) Frontal post-trauma substance defect pat.: patient; m: male, f: female; FCD: focal cortical dysplasia; R: right; L: left, temp.: temporal, med.: medial, ant.: anterior. require implanted electrodes. Therefore, to derive quantitative predictions on the extent of such signals, we proceeded in the following steps: first, we utilized the unique opportunity offered by patients with diagnostically implanted electrodes where it is possible to simultaneously measure both intra- and extracranial electrical signals. These measurements were obtained during chewing of typical soft hospital food. Next, we addressed the problem that the results from these patient measurements cannot be directly transferred to the case of healthy individuals, as in the former but not the latter the skull is breached by craniotomy defects as a consequence of the surgical electrode implantation. Such skull defects can have a substantial impact on volume Fig. 1. CT and MRI imaging data and volume conductor head model. (a)–(d) conduction that has to be taken into account. To do so, here we Axial CT images taken after subdural electrode implantation. (e) and (f) Axial used detailed finite element method (FEM) volume conductor slices through preoperative T1 and T2 weighted MRI data, respectively. (g) and (h) Coronal slices through preoperative T1 and T2 weighted MRI data, head modeling calibrated with the patient data to estimate the respectively. (i) Axial slice through segmented data of Head Model 1 (HM 1, strength of effects to be expected in the absence of craniotomy with craniotomy defects and with grid, see Section II). For comparison with the defects, by closing the skull defects in the otherwise identical MRI data the slice was taken at the same position as in (e) and (f). Soft tissue: light pink; air: black; temporalis muscle: dark pink; skull: light gray; craniotomy FEM model. Finally we performed an experiment to determine defects: red; ECoG grid: green; CSF: blue; gray matter: dark gray; and white the range of electromyogram (EMG) strength during chewing matter: light gray. (j) 3-D visualization of HM 1. Gray matter surface: pink; of food with a range of consistencies, including chewing gum. electrodes: blue; skull: transparent gray. (k) 3-D coronal slice through volume conductor model (HM 1). For comparison with the MRI data, the slice was taken In summary, by this procedure we arrived at quantitative predic- at the same position as in (g) and (h). Conventions as in (i). The red, turquoise, tions on the strength of chewing-related (ChR) cortical EF to be and green arrowheads indicate the burr holes, saw lines, and the electrode grid, expected in healthy individuals. respectively. Our results show that particularly in the region of the temporal poles, which are geometrically close to the masticatory muscles, the strength of ChR cortical EF to be expected in healthy in- and motor control. The electrode contacts were stainless steel dividuals may well reach relevant levels that could modulate or platinum discs 4 mm in diameter, mounted on a flexible cortical activity and have functional consequences. Thus, our silicone substrate (Ad-Tech, Racine, WI, USA) at a 10-mm findings lend first support to the assumption that extracranial center-to-center interelectrode distance. Most patients had ad- muscles can act as endogenous brain stimulators. ditional linearly arranged strip electrodes or penetrating depth electrodes in the hippocampus (1-mm diameter, ten contacts II. METHODS with a 5-mm contact-to-contact distance), though the effects in the depth electrodes were not of the object of the present study. A. Intracranial ChR Potentials During Weak Chewing The type and placement of all electrodes were solely deter- 1) Patients: Five patients under evaluation for neurosurgi- mined by the requirements of preneurosurgical diagnostics. All cal treatment of medically intractable epilepsy were included patients provided written informed consent prior to the study. in the present study (see Table I). Electrodes were implanted 2) Data Acquisition: Electrocorticogram (ECoG) and elec- subdurally for a period of 5–10 days, depending on the indi- troencephalogram (EEG) (standard 10–20 positions [29] as vidual clinical requirements, to localize seizure onset zones and far as allowed by the wounds) were simultaneously recorded determine eloquent brain areas to be preserved during surgical at a sampling rate of 1024 Hz, with a high-pass filter of intervention, such as those responsible for language functions 1Hzandalow-passfilterof344Hz,usingaclinical

53 2554 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 63, NO. 12, DECEMBER 2016

AC EEG-system (IT-Med, Usingen, Germany). Digital video, synchronized with neural data, was recorded at 25 frames per second at VGA resolution. Channels with technical record- ing problems (e.g., broken wires) were excluded from further analyses. 3) Trial Selection: Trials were acquired during natural food intake of the patients without any prior instruction. Chewing events were marked manually within interictal time periods based both on the digital video recordings and on the typi- cal, pronounced ChR EMG bursts of the masticatory muscles visible in the EEG (e.g., in channels T4 and F8). The EMG onset and end were marked for each chewing event [c.f. Fig. 2(a), for an example],andtheirarithmeticmeanwasdefinedasthe0-s time point for each trial. In this way, a total of 1652 trials were acquired from five patients (S1: 551 trials; S2: 438 trials; S3: 252 trials; S4: 264 trials; S5: 147 trials). 4) Analysis: The ECoG data were separately re-referenced to a common average reference (CAR), as it is common in ECoG studies [30]–[32]. The EEG data were re-referenced to Cz, as the clinical environment did not allow for a clean CAR reference and Cz was least susceptible. Trials were excerpted from the contin- uous data from –2 to 2 s with respect to the 0-s time point in the chewing event. In this time window, sliding-window fast Fourier transformations were performed with a window length of 250 ms and a step width of 24.41 ms (corresponding to 256 and 25 sampling points, respectively). A baseline period was defined in a pre-event time window (200 ms) selected around the center between consecutive chewing events [see Fig. 2(b)].Therela- Fig. 2. Chewing–related (ChR) EEG and ECoG data recorded in patients. tive time–frequency spectra were divided by the median baseline (a) Ongoing EEG from channels C4, T4, and F8 of S1 together with the power averaged across trials and then scaled logarithmically. A data from three ECoG channels (F6, G7, and H8) simultaneously recorded two-tailed sign test was employed for statistical analysis, and in the same patient. The time epoch of a chewing event, as marked for the analysis, is indicated by a blue box. The EEG traces reveal distinct EMG correction for multiple testing was performed following the false bursts, and close inspection of the ECoG channel H8 also reveals ChR discovery rate (FDR) approach suitable for correlated p-values high-frequency bursts, albeit of much lower amplitude than in EEG. The (as for neighboring time and frequency bins), with a q-level three lower traces show the high-pass-filtered ECoG signal from the same channels, enhancing the visibility of ChR high-frequency bursts. (b) Time- of 0.001 [33]. resolved ChR relative spectral power changes in the EEG channel T4 and To compare intra- and extracranial ChR EMG amplitudes, we ECoG channel H8 involving a broad frequency range. Median time points high-pass filtered the data at 100 Hz and, for each chew event, of the preceding and following chewing event are indicated above the time– frequency plot (error bars: interquartile range). Color encodes the logarith- calculated the ChR EMG amplitude as the difference between mic power change relative to the baseline (see Section II for further details). the 10th and 90th percentile in a 100-ms time window around (c) ECoG grid position in relation to the brain surface obtained from pa- the center of each trial. To test the influence of these parameters tient S3’s MRI data. (d) Time–frequency spectra of ChR responses. The course of the lateral sulcus (LC) and the central sulcus (CS) are depicted on the results, we also performed the analysis with 55 Hz high- by white lines. Note the spatially widespread distribution bridging the LS. passed data, extracted peak-to-peak amplitudes, and varied the (e) Patient S5: Lateral X-ray with superimposed positions of implanted window length from 50 to 300 ms. electrodes (blue), burr holes (white dashed discs), saw lines (white dashed lines), and the temporal muscle (red) with the temporal line (red dashed line) as its origin and the coronoid process of the mandibular bone (red asterisk) as its insertion. The variation in transparency reflects the thick- B. Volume Conductor Modeling ness of the temporal muscle, which increases toward the coronoid process. (f) Intracranial topography of chewing–related events (ChREs) in the gamma 1) FEM Head Models: Avolumeconductorheadmodelof frequency range (32–400 Hz). Electrode positions are marked with black circles. patient S3 was used to model the extra- to intracranial conduc- The saw lines and burr holes are indicated by white dashed lines and discs. tion of electric potentials caused by dipolar sources located in the left temporal muscle. Patient S3 was chosen because here The head model was created using the brain extraction tool [34] we had the best imaging data for building the FEM model. Ad- and the FMRIB Automated Segmentation Tool [35] provided ditional control simulations were performed using head mod- by the FMRIB Software Library toolbox [36]. It included white els adapted to the burr hole configuration of the other patients matter, gray matter, cerebrospinal fluid (CSF), skull, and soft (S1, 2, 4, and 5). Whole-head MRI volumes were acquired be- tissue. Anatomically unrealistic segmentation outcomes were fore surgery in a Siemens Vision scanner at 1.5 T using a T1 corrected manually. MPRAGE sequence and in a Siemens TrioTrim using a T2 SPC The model was then extended semiautomatically to include sequence, both at a 1-mm isotropic resolution [see Fig. 1(e)–(h)]. facial soft tissue and internal air. The left temporal muscle was

54 FIEDERER et al.:ELECTRICALSTIMULATIONOFTHEHUMANCEREBRALCORTEXBYEXTRACRANIALMUSCLEACTIVITY 2555 manually segmented based on the T1 and T2 data. The posi- 6) Finally, transforming the resulting 2-D triangulation back tions of burr holes, saw lines, and of the electrode grid were into a 3-D triangulation using the original 3-D coordinates determined based on the postimplantation T1 MRI and CT of the patch. scans. Because iatrogen air cavities and metal artifacts, made This created an accurate representation of the electrodes coregistration and segmentation unreliable, the craniotomy de- within the ECoG grid, molded onto the surface of the cortex, fects were included in the following way. Burr holes in the skull while respecting electrode array geometry and in the correct model were created by calculating the position of cylinders (12- position as verified using postimplantation imaging data. As, and 16-mm diameters, determined from CT) around the burr in FEM simulations, contacts over edges or corners lead to hole centers and by replacing the skull tissue within the cylin- current leakage, special care was taken to ensure that the re- der volume by CSF. The saw lines were generated based on constructed grid was “sealed” by face-to-face contacts, thus path nodes set on a surface mesh of the outer skull surface. preserving the grid’s insulating properties. Geometry-adapted The connection line of these points was then projected onto a hexahedral meshes were generated based on the segmented im- mesh of the inner skull surface. All skull points between these ages with Vgrid [48] and visualizations were performed using trajectories were replaced by CSF. SCIRun (freely available from the SCIRun Development Team). The sphenoidal and oval foramina have clinical relevance as Every 3-D surface visualized using SCIRun was smoothed us- they act as high-conductance tunnels, facilitating the recording ing the default settings of the “FairMesh” module. Based on the of brain signals [37]–[40]. Moreover, multiple studies report on procedures described above, three different head models were the importance of skull foramina in conducting epileptic spikes created: to the scalp surface [37], [41]–[44]. Therefore, we assumed that Head Model 1 (HM 1): The complete head model with burr these foramina could also play a role in the opposite direction, holes, saw lines and the insulating grid; Head Model 2 (HM 2): facilitating the propagation of EMG potentials to the brain, par- identical to HM 1, but without burr holes and saw lines, to model ticularly in the case of the pterygoid masticatory muscles, which the effects of the insulating ECoG grid separately; Head Model are very close to some major foramina. Hence, we manually 3(HM3):identicaltoHM1,butwithoutboththecraniotomy added (bilaterally) the following foramina of the skull base to the defects and the insulating electrode grid, thus, representing the model: the foramen ovale, rotundum and spinosum, the fissure situation in healthy individuals. orbitalis superior, and the carotid canals. Foramina and fissures 2) FEM Simulations and Source Models: FEM forward cal- were modeled as cylinders filled with white matter or blood culations were computed with SimBio [49] using the St. Venant as anatomically appropriate, and with diameters of 1–7 mm, dipole modeling approach [50], [51]. The conductivity values based on [45]. Carotid canals were manually segmented from used were derived from the resistivity values used in [52], the MRI data using Seg3D (Seg3D Development Team). namely white matter 0.14 S/m, gray matter 0.33 S/m, CSF Due to substantial swellings and shifts of brain tissue follow- 1.54 S/m, blood 0.63 S/m, skull 0.0063 S/m, muscle 0.11 S/m, ing surgery, as well as due to iatrogen air cavities and metal soft tissue 0.17 S/m, and internal air 0.002 S/m. Foramina filled artifacts, an automatic coregistration of the electrode grid (de- with both blood and nerves were modeled with 0.38 S/m, which termined in postimplantation 3-D images) to the preoperative is the average of blood and white matter conductivities. Burr MRI used for the volume conductor model was not reliable. holes and saw lines, as determined from CT data, were filled Thus, the position of the electrode contacts was reconstructed with CSF. For the insulating silicone ECoG grid a conductivity on the 3-D surface taking into account the positional information of 1e-45 S/m was used, which is the numerical conductivity from the postimplantation MRIs, CT, and a lateral 2-D X-ray closest to 0 S/m that SimBio could model. image. The main challenge in constructing the grid model was We used the following source models to represent the elec- to adapt it to the local gyral geometry constrained by the phys- trical activity of the chewing muscles: Source Model 1 (SM 1): ical properties of the grid. This was achieved by the following asingledipolecentralinthebellyofthetemporalmuscle(i.e., steps. the muscle contributing most force to jaw closure in chewing); 1) Creating a triangulated hull around the brain that followed Source Model 2 (SM 2): to account for the thin, superior part the outer brain surface but not the individual gyri. To this of the temporal muscle, seven dipoles were placed within the end, we used the “mesh_shrinkwrap” algorithm (Bioelec- belly of the temporal muscle and one dipole in the superior part tromagnetism MATLAB Toolbox, [46]). in front of a burr hole; Source Model 3 (SM 3): to investigate 2) Selecting the corners of the electrode grid on the hull, the impact of the pterygoid muscles, in particular of the medial based on the CT, X-ray, and MRI data. pterygoid which also contributes significant force to jaw closing 3) Extracting a 3-D patch defined by the corners from the in chewing and which is situated adjacent to major skull foram- hull. ina (e.g., the foramen ovale), a dipole was placed in the medial 4) Projecting the 3-D patch coordinates into 2D using the pterygoid muscles in front of the formaen ovale. isomap algorithm (MATLAB Toolbox for Dimensionality Reduction, [47]). This algorithm was especially suitable for this task as it is designed to well preserve the geodesic C. Noninvasive ChR Potential Measurements and Analysis distances between neighboring data points [47]. The intracranial data of the present study were acquired while 5) Finding the closest three neighbors within the 2-D patch patients ate the typically soft hospital food (soup, cake, etc.) of each electrode center. that is served to patients after having undergone major head

55 2556 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 63, NO. 12, DECEMBER 2016 surgery, during which a partial incision of the temporal muscle plitude of the noninvasive scaling data to the median EEG is likely. Thus, the ChR EMG amplitudes were relatively low and ChR amplitude of each trial [see Fig. 3(p)].Thisthird largest contralateraly to the side of surgery. To characterize EMG step gave us the single-trial distribution of the peak amplitudes that can be expected during both weak and strong cortical EF expected in healthy individuals during a vari- chewing in the general healthy individuals population, measure- ety of chewing conditions [e.g., Fig. 3(q)]. ments were conducted on three healthy participants (P1–P3) 4) Finally, we determined the percentage of trials with peak under the following six conditions: EF exceeding 0.2 V/m [see Fig. 3(r)]. 1) eating yoghurt with mashed banana (referred to as This analysis was carried out with SMs 1, 2, and 3. Yoghurt); 2) eating banana; III. RESULTS 3) eating a raw carrot; A. Chewing-Related Events (ChREs) are Clearly Present in 4) chewing gum; the ECoG 5) eating a mouthful of hard-to-chew gummi candies; and 6) eating a mouthful of licorice. Examples of simultaneously recorded ChREs in EEG and EMG potentials were recorded from 128 standard electrode ECoG from S1 are shown in Fig. 2(a). Consistent with our expec- positions in the 10–5 system [53], with Cz as reference elec- tations, ChR bursts of high-frequency activity were clearly visi- trode. As in the patients, the amplitude of the chewing events ble in the ongoing EEG recordings in all patients (shown here for was determined as the potential difference between the 10th and S1) from all temporal and fronto-lateral channels [e.g., T4 and 90th percentile of the EMG signal in the 100-Hz high-pass fil- F8 in Fig. 2(a)]. However, similar high-amplitude ChREs were tered data in a time window of 100-ms duration around EMG never observed in the simultaneously recorded ECoG [Fig. 2(a), maximum. Across participants and conditions 1639 chew events middle three traces].Nevertheless,high-frequencyChRbursts were analyzed. For each participant, the median amplitude of with peak-to-peak amplitudes of approx. 30 µVbecamevis- each channel across the chewing events of one condition was ible in the ongoing ECoG after high-pass filtering [cutoff calculated. Then the median amplitude over channels was cal- = 100 Hz, Fig. 2(a), three bottom traces].IntheunfilteredECoG culated for each participant. The mean chewing amplitude over traces, close inspection revealed relatively low-amplitude ChR the three participants was calculated for each condition and used bursts [Fig. 2(a), highlighted by blue boxes] in the ongoing (not as noninvasive scaling data (cf. below). trial-averaged) recordings at some ECoG channels. The time–frequency power spectra of the EEG data typically showed a very pronounced broadband ChR power increase in the D. Cortical EF Analysis frequency range up to the Nyquist frequency of the recordings To estimate the single-trial cortical EF that can be expected [max. ca. 500 Hz, Fig. 2(b)],whichistypicalofEMGactivity. in healthy individuals during weak to strong chewing, we pro- ChR ECoG spectra showed a similar time–frequency pattern, ceeded in the following steps. but with amplitudes smaller by about one order of magnitude 1) We determined the strength of the current dipole(s) in the [Fig. 2(b)], which is typical for extra- to intracranial propagation. masticatory muscles that would be required to generate Significant ChREs-induced power increases (p < 0.001, FDR- ECoG potentials of the same amplitude as measured in the corrected) could be observed in 406 of the 410 (99%) analyzed intracranial calibration data, i.e., in the individual chewing ECoG contacts, including grid and strip electrodes in the five eventsofS3.Tomodelthepotentialreversalexpectedinan patients investigated (all electrodes in S1, S3, and S4, and all amplitude, the simulated ECoG potentials were multiplied but two electrodes in S2 and S5). by a factor of 2 before matching them to the calibration data. These simulations were based on HM 1, i.e., with B. The Topography of Intracranial ChREs craniotomy skull defects (burr holes, saw lines) and the ChREs spectral power modulations revealed a spatially insulating electrode grid [see Figs. 1(e), (f) and 2(a)]. widespread distribution over the grid array [Fig. 2(d)].Also, This first step gave us the distribution of current dipole(s) the maximal power was found in the anterolateral corner of the strength needed to generate the data measured in S3. grid, intermediate power at other positions close to the edge 2) Then, we computed the single-trial cortical EF resulting of the grid, and the smallest power increases in the center of from the current dipole(s) derived in step 1, but using the grid. These widespread effects extended without any in- HM 3 without craniotomy and grid [see Fig. 3(c)].The terruption over the anatomical borders of the lateral sulcus cortical EF was computed for the whole extent of the (LS) and central sulcus (CS), and were not focalized to elec- cerebral cortex and the amplitude and positions of the EF trodes with oro-facial responses elicited by electrical cortical maxima were determined. This second step gave us the stimulation. distribution of the peak cortical EF strength expected in ahealthyindividual(withoutcraniotomyandelectrode C. The Intracranial ChR Power Topography is Reproduced by grid). FEM Volume Conductor Modeling 3) To calculate the EF to be expected during chewing with different muscle strength, the single-trial EF strength val- The basic power topography of intracranial ChREs [see ues determined in step 2 were scaled by the ratio of the am- Figs. 2(f) and 3(h)],withmostpowerintheanterior–inferior

56 FIEDERER et al.:ELECTRICALSTIMULATIONOFTHEHUMANCEREBRALCORTEXBYEXTRACRANIALMUSCLEACTIVITY 2557

Fig. 3. FEM simulation compared to intracranial recordings. (a)–(c) Axial slices through all three head models. Craniotomy (red) and silicone grid (green) indicated by red and green arrows, respectively. Soft tissue: light pink; skull: light gray; CSF: blue; gray matter: dark gray; and white matter: lightgray.(d) Lateral X-ray with superimposed positions of implanted electrodes (blue), burr holes (white dashed discs), saw lines (white dashed lines), and the temporal muscle (red) with the temporal line (red dashed line) as its origin, and the coronoid process of the mandibular bone (red asterisk) as its insertion, the variation in transparency reflects the thickness of the temporal muscle that increases toward the coronoid process. (e)–(g) Interpolated EMG power caused by SM 2 reproducing the power maxima in the anterio-inferior corner of the grid as observed in the recorded ECoG data (h). Electrode positions are marked with black disks. The saw lines and burr holes are indicated by white dashed lines and discs, and the lateral (LS) and central sulci (CS) are indicated by continuous white lines. (h) Intracranial topography of ChRE in the gamma frequency range (32–400 Hz). Conventions as in (e)–(g). (i)–(k) Skin: beige, skull: dark gray; dipole: magenta; ECoG grid: green. Outline of inner skull surface is marked by pink line (interrupted at the positions of the saw lines in HM 1). (l)–(n) Magnifications of the regions indicated by black boxes in (i)–(k) showing both the normalized potential (the background colors using a blue–white–red color scale) and the normalized EF (foreground cones using a red–yellow–white color scale) around the edge of the silicone grid. (o) Cortical EF (median across trials) expected in healthy individuals during chewing of licorice using SM 1. Maximal EF strength was found in the temporal pole and anterior medial and lateral temporal cortex. (p) ChR median EEG amplitudes of patient S3 and, for each chewing condition, of participants P1–3. (q) Distribution of the peak cortical EF strength across trials expected in healthy individuals for gum chewing (red units) and licorice (black units) chewing. All trials to the right of the red and black bars exceeded 0.2 V/m. Previous studies suggest modulatory effects on ongoing brain activity above this threshold (see Section IV). (r) Percentage of trials producing peak cortical EF exceeding 0.2 V/m for each chewing condition (yoghurt not shown as the percentage was always 0%), Source Models (SM) 1 and 2 (SM 3 not shown as always 100%) and different analysis parameters for high-pass frequency and amplitude window. corner of the grid could be well reproduced by an FEM for- D. The Silicone Grid has a Strong Shielding Effect ward simulation, based on dipole sources in the belly and in Comparing the results of HM 1 with those of HM 2 [cf. the thin superior part of the temporal muscle (SMs 1 and 2), Fig. 3(e), (f)], it could be seen that the craniotomy defects have including both skull defects (burr holes, saw lines) and the only a small impact on intracranial EMG power topography and insulating ECoG grid [HM 1, Figs. 1(e), (f) and 2(a)].EFvec- power amplitudes (accounting for power amplitude differences tors were forced around the edges of the silicone substrate [see of only approx. 6%). However, comparing the results of HMs 1 Fig. 3(i)–(k), (l)–(n)].Thisbasictopographywasalsorepro- [see Fig. 3(e), (i), (l)] and 2 [see Fig. 3(f), (j), (m)] with those duced with the other control simulations (SM 3 and burr holes of HM 3 [see Fig. 3(g), (k), (n)],itbecameapparentthatthe adapted to other patients, see Section II). insulating ECoG grid has a strong shielding effect. Its removal

57 2558 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 63, NO. 12, DECEMBER 2016 accounts for an 27% increase of intracranial EMG power (re- sion and gamma-band increases [56], [57]. Nevertheless, since ferring to the peak∼ EMG power across the grid). This is best previous fMRI studies have shown a cortical involvement in the illustrated in Fig. 3(l), (m), where one can see how EF vectors motor control of chewing (see above), the presence of a small, are forced to run parallel to the ECoG grid, and in Fig. 3(g), with focal neural signal component masked by the high-amplitude substantially increased EMG power in the head model without extracranial EMG seems likely, although nonexperimentally ECoG grid. performed chewing might produce much less cortical involve- ment than its experimental counterpart. Further work will be E. EEG and ECoG Amplitudes of ChREs necessary to isolate this presumably weak neural signal compo- nent, if possible at all. The median ChRE amplitude was determined across EEG and ECoG channels for all patients and subjects using the differ- ence between the 10th and 90th percentile in a 100-ms window B. FEM Modeling Predictions of ChR Cortical EF in Heathy relative to the center of the events. For S1-5 median EEG ampli- Subjects tudes were 24.9, 25.1, 33.7, 38.8, and 29.4 µV, respectively, with. From our FEM simulations based on three head models as 30.4 µVmean.MedianECoGamplitudeswere5.0,5.4,6.4,4.3, summarized in Fig. 3, it follows that high-amplitude extra-to- and 7.6 µV, respectively, with 5.7 µVmean.Thusmedianchew- intracranial signal conduction should also take place in healthy ing event amplitudes were attenuated by a factor of 5.0, 4.6, 5.3, individuals with an intact skull. This assumption was tested 9.0, and 3.9, respectively, with 5.5 mean, from EEG to ECoG. through volume conductor modeling determining the ampli- Mean ChR EEG amplitudes across healthy participants in tudes of signals resulting from extra-to-intracranial EMG prop- the different chewing conditions were yoghurt 46.6 µV, banana agation if craniotomy defects and insulating silicone grid were 45.7 µV, raw carrot 116.4 µV, gum 107.3 µV, candy 139.9 µV, removed from the head model, while keeping all other factors and licorice 155.2 µV. Results are summarized in–Fig. 3(p). constant (see Fig. 3). Not surprisingly, craniotomy defects fa- cilitated extra-to-intracranial EMG propagation and hence their F. Cortical EF Expected in Healthy Individuals removal from the head model slightly reduced the amplitudes of the EMG signals that reach the brain [compare Fig. 3(e) The strongest EF were, irrespective of Head and Source and (f)].However,whenadditionallyremovingtheinsulating Model, located at the temporal pole [see Fig. 3(o)].Forthe ECoG grid [see Fig. 3(g)],itbecameevidentthatthegridactsas EF expected in healthy individuals, depending on the source astrongelectricalshieldandthatremovalofthegridtherefore model and chewing condition, the percentage of chewing events leads to substantially increased intracranial EMG amplitudes. generating EF strengths above 0.2 V/m varied from 0 to 100%. The signal gain by removal of the insulator outweighs the sig- The predicted gum-ChR EF strengths were above 0.2 V/m in nal loss by closing the craniotomy, resulting in a net signal 27.5% of trials in SM 1 (one dipole in the belly of the tempo- increase in the “healthy” head model (HM 3) as compared to ralis muscle), 25.9% in SM 2 (seven dipoles in the belly of the HM 1 with craniotomy and with grid [see Fig. 3(e), (g)]. This ef- temporalis muscle and one in the superior part), and 100% in fect was observed consistently in a range of control simulations SM 3 (one dipole in the medial pterygoid muscle in front of with source configurations with different levels of spatial detail. the foramen ovale). For details relating to the other conditions, These results also imply that, in the opposite direction, cortical as well as parameter variations, cf. Fig. 3(r). Median chewing potentials generated below the ECoG grid should be attenu- repetition rate ranged, across all patients, participants, and con- ated in EEG recordings above the insulating grid, even in the ditions, from 0.82 to 1.8 Hz. presence of craniotomy defects as indeed shown by [58]–[60] (however, see also [61], [62]). The assumption that signals in the IV. DISCUSSION gamma-frequency range, in which EMG has high amplitudes, A. ChREs Mainly Arise From EMG Activity can indeed overcome the intact skull is further supported by earlier studies showing that, in the other direction, task-related For a number of reasons, it appears most plausible to as- gamma responses originating from the brain can be detected in sume that the ChREs observed in the present study, for the scalp EEG in healthy individuals [63], [64]. most part, arise from the EMG activity of the masticatory mus- cles, rather than result from neural activity related to sensory processing or motor control of the act of chewing. In two pre- C. Could Cortical EF Induced by Chewing Modulate Brain vious functional magnetic resonance imaging (fMRI) studies, Activity? BOLD signal changes related to chewing, tongue tapping, or 1) Cortical EF Induced by Chewing Are in the Proper Ampli- swallowing [54], [55] were found focally in regions of the pri- tude Range to Modulate Brain Activity: Recent evidence sug- mary sensory and motor cortex with a spatial response pat- gests that even weak EF (in the range of 0.2 V/m) can have a tern clearly different from the spatially widespread distribution direct influence on the activity of neocortical neural networks of ChREs in our study, which extended smoothly over func- [13]. While low-amplitude EF did not trigger additional action tional and structural boundaries [see Fig. 2(d)].Furthermore, potentials, they did induce substantial shifts in the timing of ac- the spectral profile of the ChREs showed broadband frequency tion potentials [12]–[14]. Neuronal networks have been shown increases instead of the typical of event-related neural popula- to be even more sensitive to EF than single neurons [15]. The tion responses of the cortex with both low-frequency suppres- theoretical sensitivity limit of elongated neurons was calculated

58 FIEDERER et al.:ELECTRICALSTIMULATIONOFTHEHUMANCEREBRALCORTEXBYEXTRACRANIALMUSCLEACTIVITY 2559 to be in the order of 0.01 V/m [65] but no empirical study has asinefunction,likelybecausetheformerconsistedofsharp yet confirmed this prediction. rising ramps with high slopes, similar to the time course of Typical stimulation intensities used in previous transcranial EMG activity in our study. random noise stimulation (tRNS) studies were in the 1-mA 3) Role of Chewing Repetition Rate: Besides the frequency peak-to-peak amplitude range [20], but already 0.4 mA tRNS contents of the EMG generated with each individual chewing has been shown to modulate cortical function [21]. The maximal event, the repetition rate of these events (how fast or slow one cortical EF strength directly beneath a stimulation pad and at chews) may also play a role in our context. Anastasious et al. 1mAwasfoundtobe0.45V/m[25],hence0.18V/mEFcanbe [14] reported that weak EF oscillating at low (<8Hz)frequen- expected to be responsible for the effects observed with 0.4 mA cies are particularly effective for entraining action potentials tRNS, which matches well the threshold of 0.2 V/m determined in rat cortical slices. Similarly, Ozen et al. [16] demonstrated empirically by Reato and colleagues [13]. The assumption that that TES at 0.8–1.7 Hz significantly entrained neuronal activ- cortical EF in this order of magnitude has a modulatory effect on ity in anesthetized and sleeping, but not in behaving, rats. In neuronal network function is strongly supported by data from humans, Marshall et al. [17] could show that TES oscillating recent in vitro experiments [12], [13]. at 0.75 Hz during non-rapid-eye-movement sleep significantly With our SM 2 (seven dipoles in the belly of the temporalis increased declarative memory retention rates. By contrast, 5-Hz muscle and one in the superior part), 25.9% of chewing events TES did not induce any changes in declarative memory retention scaled for gum chewing in healthy individuals produced peak EF rates. Kirov et al. [18] could consequently extend the results of strengths larger than the empiric threshold of 0.2 V/m. When Marshall et al. to wakefulness, also using 0.75-Hz TES. Across varying the window length used to calculate the ChR ampli- all patients and healthy participants, the median chewing rep- tudes from 50 to 300 ms, this percentage ranged from 33.1% to etition rate ranged from approx. 0.8 to 1.8 Hz. This repetition 14.7%, respectively [cf. Fig. 3(r) for more details].Aswegrad- rate range is further supported by literature [68] and quite close ually increased the firmness of the chewed food the proportion to the stimulation frequencies described above and could thus of chewing events above 0.2 V/m also increased: carrot 35.5%, favor the entrainment of neuronal activity. candy 66.5%, and licorice 80.1%. These strong EFs involved 4) Cortical EF Induced by Chewing May Modulate Brain the temporal poles, extending to the medial and lateral anterior Activity and Influence Cognitive Performance: Together, these temporal regions [see Fig. 3(o)].SM1(onedipoleinthebellyof results show that on the one hand, the cortical EF to be expected the temporalis muscle) produced slightly larger values as SM 2 in healthy individuals should depend on the exact recruitment while SM 3 (one dipole in the medial pterygoid muscle in front pattern of the masticatory muscles. At the same time, though, of the foramen ovale) continuously produced EF above 0.2 V/m. our findings indicate that the effects to be expected in healthy in- These differences are understandable as dipoles in the superior dividuals might be in the same order of magnitude (0.1–1 V/m), part of the temporalis muscles are in a “good” (spatially close) frequency range (100–500 Hz), and repetition rate (1–2 Hz) as position to generate potentials measurable in the ECoG grid, EF caused by external technical (tRNS) and endogenous neu- but contribute little to the anterior temporal EF, which is mainly ronal sources that have both been shown to have an impact on caused by dipoles in the belly of the temporal muscle. The op- neural network activity. posite is true for dipoles representing activity of the pterygoid Thus, taking together previous insights that even weak EF masticatory muscles. Due to their position, dipole sources here have a modulating impact on cortical network dynamics, find- must be of relatively high amplitudes to generate appreciable ings from tRNS stimulation, and our present findings on how ECoG potentials but they can “easily” cause high anterior tem- endogenous EF propagate to the human cortex during chew- poral cortical EF, because they are situated close to the foramina ing, it appears possible that ChR EMG acts as an endogenous of the skull base, which act as high-conductance tunnels con- type of brain stimulation, potentially exerting similar effects on necting the extracranial and intracranial space [66], [67]. brain functions as are elicited by exogenous brain stimulation, 2) Cortical EF Induced by Chewing Are in the Proper in particular tRNS. Frequency Range to Modulate Brain Activity: tRNS, i.e., brain stimulation with a broadband signal similar to the EMG examined here, is particularly effective in modulating cortical D. Cortical EF Induced by Chewing: A Possible Explanation network function. tRNS can improve underlying for Gum Chewing Effects on Cognition motor and perceptual learning with effects lasting at least 60 Chewing gum has repeatedly been reported to have effects on min after stimulation [19], [20]. The effect of tRNS appeared to cognitive functions [28], [69]–[71]. By administering a battery depend mainly on the high-frequency (100–600 Hz) component of cognitive tasks to participants who chewed gum either prior to of the stimulation signal, whereas the lower frequencies seem to or during testing, it was recently confirmed that chewing is asso- be less important [20]. In addition to producing lower frequency ciated with changes in cognitive performance that are not present components in the ECoG, the ChR EMG had pronounced in nonchewing controls. Critically, in chewing subjects, a wors- effects in the ECoG in the range from 100 to at least 500 Hz ening in cognitive performance was observed during chewing, [see Fig. 2(a), (b), (d), (f)].Moreover,Frohlich¨ and McCormick whereas a consecutive enhancement in performance took place [12] presented strong evidence that naturalistic stimulation when the chewing preceded the cognitive measurements [28]. (using previously recorded ongoing EF) was more effective The beneficial effects of chewing were reported to last for a time at entraining network activity than artificial EF modulated by period of 15–20 min after the subjects had chewed gum. These

59 2560 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 63, NO. 12, DECEMBER 2016 effects were previously explained by indirect psychological ef- Amoredetailedrepresentationoftheskull-basechewing fects, in particular by unspecific arousal. In contrast, based on muscles would be desirable but could probably further increase the findings of the present study we propose that the observations the ChR EF. on cognitive performance may at least partly be explained by The head model used in our study is also limited. Due to the direct electrical stimulation of the brain by one’s own EMG. The brain shift that occurs when the skull of the patient is opened cortical EF to be expected, especially in the anterior temporal during surgery, the alignment of the preoperative MRI and the lobe [see Fig. 3(o)],inhealthyindividualsduringgumchewing postimplantation images was most likely suboptimal. Therefore, might be in the same order of magnitude as both exogenously we must expect some inaccuracy in our head model. It would and endogenously caused EF that modulate cortical neuronal be advantageous to use the postimplantation MRI for model function (see above). The temporal pole and the adjacent area construction, but this was hindered by large iatrogen air cavities of the anterior and medial temporal lobe have been implicated as well as by large metal artifacts. We see two possibilities to in a wide range of cognitive functions [72]–[74] and (subtle) improve our modeling in future work. First, following [76], if modulation of neuronal activity in these regions by masticatory postimplantation MRI with inverted phase-encoding direction EMG may, therefore, indeed contribute to the reported cognitive has additionally been measured, it should be possible to correct effects of chewing. postimplantation MRI artifacts using a reversed gradient artifact Generally, the underlying mechanisms and hence the range correction approach [77]. Another strategy could be to model of effects that can be achieved with brain stimulation techniques the brain shift as reported by [78] and subsequently use it for goes far beyond the consequences of unspecific effects such as an improved registration of a preoperative MRI and a postim- arousal [75]. The effects of tRNS have, for example, been linked plantation CT. This procedure would also make it possible to to the phenomenon of stochastic resonance [20]. The notion model the metal contacts of the electrodes that could introduce of endogenous brain stimulation presents a novel principle by local EF distortions. As only 4.15% of the silicone grid would which interfering with and modulation of neural activity in the be replaced by open metal contacts, we however anticipate that human brain may be possible. Among the many topics for further results should be influenced rather minimally. research that arise, evaluating the potential of endogenous brain The conductivities used in our study are widely used, but stimulation as a new experimental tool and even for clinical their accuracy could be further improved, such as by taking application, complementary to the exogenous, technical brain into account their inter- and intraindividual variabilities [79]– stimulation currently used exclusively for this purpose, will be [81] and frequency dependence [80], [82]. Moreover, we could of particular importance. try to incorporate the known inhomogeneous and anisotropic conductivity of skull and brain [83]–[86]. As shown by [84] E. Limitations and [85], however, brain anisotropy only plays an important Although we took great care to construct a detailed and role for sources deep in the brain while we investigated sources precise analysis, some limitations of our results need to be outside of the brain. Taken together, therefore, we do not expect discussed. significant differences in the results for our specific simulation 1) Sample Size and Calibration/Scaling Procedure: The setup, due to these various model simplifications. results are based on a small sample, five epilepsy patients, only one of which we used for volume conduction modeling, and three healthy participants, which obviously restrains the gener- V. C ONCLUSION alization of our results. These should therefore be considered as We presented our first results toward clarifying whether en- tentative until confirmed in a larger sample. We took great care dogenously produced EF beyond those arising from neuronal to use conservative parameters for the calibration and scaling activity, e.g., in our case ChR EMG, can influence brain activ- procedure. By using the difference between the 10th and 90th ity and function. Using an FEM head model, calibrated with percentile as chewing amplitude, we increased the robustness intracranial ECoG data from an epilepsy patient and noninva- against noise but likely underestimated the true peak-to-peak sive EEG data from healthy participants, we could show that amplitude of the chewing events. Progressive pooling of the the amplitude of the ChR EMG expected to reach the cortex of noninvasive scaling data using the median instead of mean healthy individuals during strong chewing might indeed be suf- further increased our robustness against outliers but reduced the ficiently strong to have such effects. The simulated amplitudes final percentage of trials above 0.2 V/m by an average of 7.8%. of the ChR cortical EF that we found were very close to the Similarly, by 100 Hz high-passing the ECoG signal before stimulation thresholds previously suggested for both endoge- ChR amplitude analysis, the low-frequency components of nous and exogenous brain stimulation [12], [13], [21], [25]. The the muscle activity (30–100 Hz) were discarded, again giving present study demonstrates that the combination of simultane- conservative estimates. To illustrate this, we show results of the ous intra- and extracranial EEG recordings with detailed FEM calibration and scaling procedure for 55 Hz (above line-noise) volume conductor modeling is a powerful approach to assess high-passed data in Fig. 3(r). the impact of muscle activity on the human brain. We believe 2) Source and Head Modeling: Our simple source models that this approach will also be useful in further studies on the qualitatively reproduced the measured intracranial topographies electrical muscle effects on the brain. For example, such future well, but more detailed EMG source models would further research might gain further insight by using data from intracra- approximate the real electrical activity induced by chewing. nial stereotactic EEG recordings alongside ECoG. Stereotactic

60 FIEDERER et al.:ELECTRICALSTIMULATIONOFTHEHUMANCEREBRALCORTEXBYEXTRACRANIALMUSCLEACTIVITY 2561

recordings may offer additional information as such electrodes with higher intensities,” Brain Stimulation,vol.5,no.4,pp.505–511, are sometimes implanted close to the temporal poles and skull Oct. 2012. [22] B. Pollok et al., “The effect of transcranial alternating current stimulation base, where according to our findings, muscles effects should (tACS) at alpha and beta frequency on motor learning,” Behav. Brain Res., be especially strong. vol. 293, pp. 234–240, Oct. 2015. [23] M. Meinzer et al., “Electrical brain stimulation improves cognitive perfor- mance by modulating functional connectivity and task-specific activation,” ACKNOWLEDGMENT J. Neurosci.,vol.32,no.5,pp.1859–1866,Feb.2012. [24] C. A. Dockery et al., “Enhancement of planning ability by transcranial L. D. J. Fiederer thanks his wife and his daughter for their direct current stimulation,” J. Neurosci.,vol.29,no.22,pp.7271–7277, Jun. 2009. continuous and loving support and without whom the hardships [25] A. Datta et al.,“Gyri–preciseheadmodeloftranscranialDCstimu- of research would be far less bearable. lation: Improved spatial focality using a ring electrode versus conven- tional rectangular pad,” Brain Stimulation,vol.2,no.4,pp.201–207, Oct. 2009. REFERENCES [26] H. Otsubo et al., “High-frequency oscillations of ictal muscle activity and epileptogenic discharges on intracranial EEG in a temporal lobe [1] A. Arvanitaki, “Effects evoked in an axon by the activity of a contiguous epilepsy patient,” Clin. Neurophysiol.,vol.119,no.4,pp.862–868, one,” J. Neurophysiol.,vol.5,no.2,pp.89–108,Mar.1942. Apr. 2008. [2] J. G. R. Jefferys and H. L. Haas, “Synchronized bursting of CA1 [27] H. Liu et al.,“EvaluationofaBSSalgorithmforartifactsrejectionin hippocampal pyramidal cells in the absence of synaptic transmission,” epileptic seizure detection,” in Proc. 26th Annu. Int. Conf. IEEE Eng. Nature,vol.300,no.5891,pp.448–450,Dec.1982. Med. Biol. Soc.,Jan.2004,vol.1,pp.91–4. [3] R. D. Traub et al., “Simulation of hippocampal after discharges syn- [28] S. V. Onyper et al.,“Cognitiveadvantagesofchewinggum.Nowyousee chronized by electrical interactions,” Neuroscience, vol. 14, no. 4, them, now you don’t,” Appetite,vol.57,no.2,pp.321–328,Oct.2011. pp. 1033–1038, Apr. 1985. [29] G. H. Klem et al., “The ten-twenty electrode system of the international [4] C. Y. Chan and C. Nicholson, “Modulation by applied electric fields of federation. The international federation of clinical neurophysiology,” Purkinje and stellate cell activity in the isolated turtle cerebellum,” J. Electroencephalogr. Clin. Neurophysiol. Suppl.,vol.52,no.suppl., Physiol.,vol.371,no.1,pp.89–114,Feb.1986. pp. 3–6, 1999. [5] C. Y. Chan et al., “Effects of electric fields on transmembrane potential [30] T. Ball et al.,“Signalqualityofsimultaneouslyrecordedinvasiveand and excitability of turtle cerebellar Purkinje cells in vitro,” J. Physiol., non-invasive EEG,” NeuroImage,vol.46,no.3,pp.708–716,Jul.2009. vol. 402, no. 1, pp. 751–771, Aug. 1988. [31] N. E. Crone et al., “High-frequency gamma oscillations and human [6] J. G. Jefferys, “Nonsynaptic modulation of neuronal activity in the brain: brain mapping with electrocorticography,” Progress Brain Res.,vol.159, Electric currents and extracellular ions,” Physiol. Rev.,vol.75,no.4, pp. 275–295, 2006. pp. 689–723, Oct. 1995. [32] V. L. Towle et al., “Frequency domain analysis of human subdural record- [7] M. Bikson et al.,“EffectsofuniformextracellularDCelectricfieldson ings,” J. Clin. Neurophysiol.,vol.24,no.2,pp.205–213,Apr.2007. excitability in rat hippocampal slices in vitro,” J. Physiol.,vol.557,no.1, [33] Y. Benjamini and Y. Hochberg, “Controlling the false discovery rate: A pp. 175–190, May 2004. practical and powerful approach to multiple testing,” J. Roy. Statist. Soc. [8] J. K. Deans et al., “Sensitivity of coherent oscillations in rat hippocampus Ser. B Methodol.,vol.57,no.1,pp.289–300,1995. to AC electric fields,” J. Physiol.,vol.583,no.2,pp.555–565,Sep.2007. [34] M. Jenkinson et al., “BET2: MR-based estimation of brain, skull and scalp [9] T. Radman et al.,“Spiketimingamplifiestheeffectofelectricfieldson surfaces,” in Proc. 11th Annu. Meeting Org. Human Brain Mapping,2005, neurons: Implications for endogenous field effects,” J. Neurosci.,vol.27, pp. 12–16. no. 11, pp. 3030–3036, Mar. 2007. [35] Y. Zhang et al., “Segmentation of brain MR images through a hidden [10] C. A. Anastassiou et al., “The effect of spatially inhomogeneous extracel- Markov random field model and the expectation-maximization algorithm,” lular electric fields on neurons,” J. Neurosci.,vol.30,no.5,pp.1925–1936, IEEE Trans. Med. Imag.,vol.20,no.1,pp.45–57,Jan.2001. Feb. 2010. [36] M. W. Woolrich et al.,“BayesiananalysisofneuroimagingdatainFSL,” [11] M. Zhang et al.,“Propagationofepileptiformactivitycanbeindependent NeuroImage,vol.45,no.1,Supplement1,pp.S173–S186,Mar.2009. of synaptic transmission, gap junctions, or diffusion and is consistent with [37] J. L. Fernandez-Torre´ et al., “Comparison of sphenoidal, foramen ovale electrical field transmission,” J. Neurosci.,vol.34,no.4,pp.1409–1419, and anterior temporal placements for detecting interictal epileptiform Jan. 2014. discharges in presurgical assessment for temporal lobe epilepsy,” Clin. [12] F. Frohlich¨ and D. A. McCormick, “Endogenous electric fields may Neurophysiol.,vol.110,no.5,pp.895–904,May1999. guide neocortical network activity,” Neuron,vol.67,no.1,pp.129–143, [38] A. M. Kanner et al., “The utility of placing sphenoidal electrodes under Jul. 2010. the foramen ovale with fluoroscopic guidance,” J. Clin. Neurophysiol., [13] D. Reato et al.,“Low-IntensityElectricalStimulationAffectsNetwork vol. 12, no. 1, pp. 72–81, Jan. 1995. Dynamics by Modulating Population Rate and Spike Timing,” J. Neu- [39] U. C. Wieshmann et al.,“Foramenovalerecordings:Apresurgicalinves- rosci.,vol.30,no.45,pp.15067–15079,Nov.2010. tigation in epilepsy,” Eur. Neurol.,vol.49,no.1,pp.3–7,2003. [14] C. A. Anastassiou et al., “Ephaptic coupling of cortical neurons,” Nature [40] D. Zumsteg et al.,“Sourcelocalizationofmesialtemporalinterictal Neurosci.,vol.14,no.2,pp.217–223,Feb.2011. epileptiform discharges: Correlation with intracranial foramen ovale elec- [15] J. T. Francis et al., “Sensitivity of neurons to weak electric fields,” J. trode recordings,” Clin. Neurophysiol.,vol.116,no.12,pp.2810–2818, Neurosci.,vol.23,no.19,pp.7255–7261,Aug.2003. Dec. 2005. [16] S. Ozen et al.,“Transcranialelectricstimulationentrainscorticalneu- [41] J. L. Fernandez-Torre´ et al., “Generation of scalp discharges in tem- ronal populations in rats,” J. Neurosci.,vol.30,no.34,pp.11476–11485, poral lobe epilepsy as suggested by intraoperative electrocorticographic Aug. 2010. Recordings,” J. Neurol. Neurosurgery Psychiatry,vol.67,no.1,pp.51–58, [17] L. Marshall et al., “Boosting slow oscillations during sleep potentiates Jul. 1999. memory,” Nature,vol.444,no.7119,pp.610–613,Nov.2006. [42] B. C. Heasman et al.,“Aholeintheskulldistortssubstantiallythedis- [18] R. Kirov et al.,“Slowoscillationelectricalbrainstimulationduringwaking tribution of extracranial electrical fields in an in vitro model,” J. Clin. promotes EEG theta activity and memory encoding,” Proc. Nat. Acad. Sci., Neurophysiol.,vol.19,no.2,pp.163–171,2002. vol. 106, no. 36, pp. 15460–15465, Sep. 2009. [43] D. Nayak et al.,“Characteristicsofscalpelectricalfieldsassociated [19] A. Fertonani et al., “Random noise stimulation improves neuroplasticity with deep medial temporal epileptiform discharges,” Clin. Neurophysiol., in perceptual learning,” J. Neurosci.,vol.31,no.43,pp.15416–15423, vol. 115, no. 6, pp. 1423–1435, Jun. 2004. Oct. 2011. [44] M. Sparkes et al.,“Mechanismsinvolvedintheconductionofante- [20] D. Terney et al., “Increasing human brain excitability by transcranial rior temporal epileptiform discharges to the scalp,” Clin. Neurophysiol., high-frequency random noise stimulation,” J. Neurosci.,vol.28,no.52, vol. 120, no. 12, pp. 2063–2070, Dec. 2009. pp. 14147–14155, Dec. 2008. [45] A. Berlis et al., “Direct and CT measurements of canals and foramina of [21] V. Moliadze et al.,“Closetothresholdtranscranialelectricalstimulation the skull base,” Brit. J. Radiol.,vol.65,no.776,pp.653–661,Aug.1992. preferentially activates inhibitory networks before switching to excitation [46] D. Weber, Bioelectromagnetism MATLAB Toolbox. 2009.

61 2562 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 63, NO. 12, DECEMBER 2016

[47] L. van der Maaten et al., “Dimensionality reduction: A comparative re- [74] I. R. Olson et al., “The enigmatic temporal pole: A review of findings on view,” Tilburg Univ., Tilburg, Netherlands, Tech Rep. TiCC TR 2009–005, social and emotional processing,” Brain,vol.130,no.7,pp.1718–1731, 2009, pp. 1–35. Jul. 2007. [48] C. H. Wolters et al., “Geometry-adapted hexahedral meshes improve accu- [75] S. Borchers et al.,“Directelectricalstimulationofhumancortex—the racy of finite-element-method-based EEG source analysis,” IEEE Trans. gold standard for mapping brain functions?,” Nature Rev. Neurosci., Biomed. Eng.,vol.54,no.8,pp.1446–1453,Aug.2007. vol. 13, no. 1, pp. 63–70, Jan. 2012. [49] SimBio Development Group, SimBio—A generic environment for bio- [76] S. Skare and J. L. R. Andersson, “Correction of MR image distortions in- numerical simulations. 2012. duced by metallic objects using a 3D cubic B-spline basis set: Application [50] J. Vorwerk et al.,“Comparisonofboundaryelementandfiniteelement to stereotactic surgical planning,” Magn. Resonance Med.,vol.54,no.1, approaches to the EEG forward problem,” in Proc. 46 DGBMT Annu. pp. 169–181, Jul. 2005. Conf.,Jena,Germany,2012,vol.57,pp.795–798. [77] L. Ruthotto et al.,“Diffeomorphicsusceptibilityartifactcorrection [51] C. H. Wolters et al., “Numerical approaches for dipole modeling in finite of diffusion-weighted magnetic resonance images,” Phys. Med. Biol., element method based source analysis,” Int. Congr. Series,vol.1300, vol. 57, no. 18, pp. 5715–5731, Sep. 2012. pp. 189–192, Jun. 2007. [78] G. Soza, “Registration and simulation for the analysis of intraopera- [52] J. Haueisen et al., “On the influence of volume currents and extended tive brain shift,” PhD thesis, Friedrich-Alexander-Universitat¨ Erlangen- sources on neuromagnetic fields: A simulation study,” Ann. Biomed. Eng., Nurnberg,¨ 2005, http://opus4.kobv.de/opus4-fau/files/175/Dissertation_ vol. 23, no. 6, pp. 728–739, Dec. 1995. Soza_2005.pdf. [53] R. Oostenveld and P. Praamstra, “The five percent electrode system [79] L. A. Geddes and L. E. Baker, “The specific resistance of biological for high-resolution EEG and ERP measurements,” Clin. Neurophysiol., material—A compendium of data for the biomedical engineer and physi- vol. 112, no. 4, pp. 713–719, Apr. 2001. ologist,” Med. Biol. Eng.,vol.5,no.3,pp.271–293,May1967. [54] G. A. Malandraki et al.,“Neuralactivationofswallowingandswallowing- [80] C. Gabriel et al.,“Thedielectricpropertiesofbiologicaltissues:I. related tasks in healthy young adults: An attempt to separate the compo- Literature survey,” Phys. Med. Biol.,vol.41,no.11,pp.2231–2249, nents of deglutition,” Hum. Brain Mapping,vol.30,no.10,pp.3209–3226, Nov. 1996. Oct. 2009. [81] S. K. Law, “Thickness and resistivity variations over the upper surface of [55] M. Onozuka et al., “Mapping brain region activity during chewing: A the human skull,” Brain Topography,vol.6,no.2,pp.99–109,1993. functional magnetic resonance imaging study,” J. Dental Res.,vol.81, [82] S. Gabriel et al.,“Thedielectricpropertiesofbiologicaltissues:II.Mea- no. 11, pp. 743–746, Nov. 2002. surements in the frequency range 10 Hz to 20 GHz,” Phys. Med. Biol., [56] N. E. Crone et al., “Functional mapping of human sensorimotor cortex with vol. 41, no. 11, pp. 2251–2269, Nov. 1996. electrocorticographic spectral analysis. I. Alpha and beta event-related [83] M. Dannhauer et al., “Modeling of the human skull in EEG source analy- desynchronization,” Brain,vol.121,no.12,pp.2271–2299,Dec.1998. sis,” Hum. Brain Mapping,vol.32,no.9,pp.1383–1399,Sep.2011. [57] G. Pfurtscheller and F. H. Lopes da Silva, “Event-related EEG/MEG syn- [84] M. Rullmann et al.,“EEGsourceanalysisofepileptiformactivityusing chronization and desynchronization: Basic principles,” Clin. Neurophys- a 1 mm anisotropic hexahedra finite element head model,” NeuroImage, iol.,vol.110,no.11,pp.1842–1857,Nov.1999. vol. 44, no. 2, pp. 399–410, Jan. 2009. [58] B. Lanfer et al., “Influence of a silastic ECoG grid on EEG/ECoG based [85] C. H. Wolters et al., “Influence of tissue conductivity anisotropy on source analysis,” Brain Topogr.,vol.26,no.2,pp.212–228,Apr.2013. EEG/MEG field and return current computation in a realistic head model: [59] Y. Zhang et al., “A cortical potential imaging study from simultaneous A simulation and visualization study using high-resolution finite element extra- and intracranial electrical recordings by means of the finite element modeling,” NeuroImage,vol.30,no.3,pp.813–826,Apr.2006. method,” NeuroImage,vol.31,no.4,pp.1513–1524,Jul.2006. [86] J. Vorwerk et al.,“Aguidelineforheadvolumeconductormodelingin [60] N. von Ellenrieder et al., “Extent of cortical generators visible on the scalp: EEG and MEG,” NeuroImage,vol.100,pp.590–607,Oct.2014. Effect of a subdural grid,” NeuroImage,vol.101,pp.787–795,Nov.2014. [61] J. X. Tao et al.,“IntracranialEEGsubstratesofscalpEEGinterictal spikes,” Epilepsia,vol.46,no.5,pp.669–676,May2005. [62] J. X. Tao et al.,“CorticalsubstratesofscalpEEGepileptiformdischarges,” J. Clin. Neurophysiol.,vol.24,no.2,pp.96–100,Apr.2007. [63] T. Ball et al., “Movement related activity in the high gamma range of the human EEG,” NeuroImage,vol.41,no.2,pp.302–310,Jun.2008. Lukas Dominique Josef Fiederer was born in [64] F. Darvas et al.,“HighgammamappingusingEEG,”NeuroImage, Meyrin, Switzerland, in 1988. He received the Dipl. vol. 49, no. 1, pp. 930–938, Jan. 2010. Biol. degree in biology from the University of [65] J. C. Weaver et al., “Theoretical limits on the threshold for the response of Freiburg, Freiburg, Germany, in 2012. He is currently long cells to weak extremely low frequency electric fields due to ionic and working toward the doctoral thesis (Dr. rer. nat.) at molecular flux rectification,” Biophys. J.,vol.75,no.5,pp.2251–2254, the University Medical Center Freiburg, Freiburg, Nov. 1998. Germany. [66] L. D. J. Fiederer et al., “The role of blood vessels in high-resolution Since 2013, he is also in charge of the EEG labora- volume conductor head modeling of EEG,” NeuroImage,vol.128, tory of the Intracranial EEG and Brain Imaging Lab, pp. 193–208, Mar. 2016. Epilepsy Center, University Medical Center Freiburg, [67] B. Lanfer et al., “Influences of skull segmentation inaccuracies on EEG Freiburg, Germany. His main research interests in- source analysis,” NeuroImage,vol.62,no.1,pp.418–431,Aug.2012. clude human electrophysiology, in particular surface EEG and invasive EEG, [68] J. M. C. Po et al., “Time-frequency analysis of chewing activity in the their measurement, analysis and simulation/reconstruction using highly detailed natural environment,” J. Dental Res.,vol.90,no.10,pp.1206–1210, and individualized FEM head models. Oct. 2011. [69] A. Smith, “Effects of chewing gum on mood, learning, memory and per- formance of an intelligence test,” Nutritional Neurosci.,vol.12,no.2, pp. 81–88, Apr. 2009. [70] L. Tucha and W. Simpson, “The role of time on task performance in modifying the effects of gum chewing on attention,” Appetite,vol.56, Jacob Lahr received the M.D. degree from Albert- no. 2, pp. 299–301, Apr. 2011. Ludwigs University, Freiburg, Germany, in 2011. [71] L. Wilkinson et al., “Chewing gum selectively improves aspects of He is currently a Resident in the Department of memory in healthy volunteers,” Appetite,vol.38,no.3,pp.235–236, Psychiatry, University Medical Center of Freiburg, Jun. 2002. Freiburg, Germany. His research fields include signal [72] R. N. Spreng et al., “The common neural basis of autobiographical mem- quality in intracranial electroencephalography and ory, prospection, navigation, theory of mind, and the default mode: A noninvasive brain stimulation. quantitative meta-analysis,” J. Cogn. Neurosci., vol. 21, no. 3, pp. 489– 510, May 2008. [73] R. A. Mar, “The neural bases of social cognition and story comprehen- sion,” Annu. Rev. Psychol.,vol.62,no.1,pp.103–134,2011.

62 FIEDERER et al.:ELECTRICALSTIMULATIONOFTHEHUMANCEREBRALCORTEXBYEXTRACRANIALMUSCLEACTIVITY 2563

Johannes Vorwerk received the M.Sc. degree in Carsten Hermann Wolters received the M.Sc. de- mathematics with a minor in physics from the Uni- gree in mathematics with a minor in medicine from versity of Munster,¨ Munster,¨ Germany, in 2011 and the RWTH Aachen, Aachen, Germany, the Ph.D. de- the PhD degree in Mathematics from the University gree in mathematics from the University of Leipzig, of Munster¨ in 2016. In April 2016, he joined the Sci- Leipzig, Germany, and the Habilitation in mathemat- entific Computing and Imaging (SCI) Institute at the ics from the University of Munster,¨ Munster,¨ Ger- University of Utah, Salt Lake City, USA as a Post- many, in 1997, 2003, and 2008, respectively. doctoral Fellow. From 1997 to 2004, he was with the Max Planck His research interests include EEG/MEG source Institutes for Human Cognitive and Brain Sciences localization with a focus on solving the EEG/MEG and Mathematics in the Sciences, Leipzig, Germany. forward problem using FE methods and tCS In 2004, he joined the Scientific Computing and optimization. Imaging Institute, University of Utah, Salt Lake City, UT, USA. Since 2005, he has been a Research Associate with the Institute for Biomagnetism and Biosig- nalanalysis (IBB), University of Munster,¨ Munster,¨ Germany. Since 2008, he has been heading the research group “Methods in Bioelectromagnetism” at IBB. His research interests include the field of neuroscience with a focus on recon- structing and manipulating neuronal networks in the human brain.

Felix Lucka received the Ph.D. degree in math- ematics from the University of Munster,¨ Munster,¨ Germany, supervised by Martin Burger, the Head of the working group “Imaging,” Institute for Applied Mathematics, and by Carsten H. Wolters, the Head of the working group “Methods in bioelectromag- Andreas Schulze-Bonhage received the M.D. de- netism,” Institute for Biomagnetism and Biosignal- gree in medicine from the University of Munster,¨ analysis. Munster,¨ Germany, in 1998. He was habilitated by In September 2014, he joined the “Center for Med- the University of Freiburg in 2004. ical Image Computing” (CMIC) at the UCL, U.K., as He is the Head of the Epilepsy Center, Univer- aResearchAssociatetoworkwithSimonArridge. sity Medical Center Freiburg, Freiburg, Germany and His research interests include inverse problems, Bayesian inference, and math- aProfessorofneurologyandclinicalneurophysiol- ematical modeling applied to biomedical imaging and brain research. ogy. Dr. Schulze-Bonhage is a Member of the board of directors, University Medical Center, Freiburg, a Member of the Bernstein Center Freiburg, and a Prin- cipal Investigator in the Excellence Cluster BrainLinks-BrainTools. Further- more, he is a Member of the editorial boards of Epilepsia, Seizure, Frontiers in Epilepsy, Epilepsy Research and Treatment,andZeitschrift fur¨ Epileptologie. He has authored more than 240 publications in peer reviewed journals. Ad Aertsen received the M.Sc. degree in physics from the University of Utrecht, The Netherlands, in 1973. He also studied applied mathematics from the same institute. He received the Ph.D. degree in physics in P. Johannesma and J. Eggermont’s Neu- rophysics Laboratory, University of Nijmegen, The Netherlands, in 1981. He pursued his postdoctoral studies in physiology with G. Gerstein at the Univer- sity of Pennsylvania, Philadelphia, PA, USA. Tonio Ball received the M.D. degree in medicine He is a Professor of neurobiology and biophysics, from the University of Freiburg, Freiburg, Germany. Faculty of Biology, University of Freiburg, Freiburg, He has a permanent position as a Research Group Germany. He is a Founding Director of the Bernstein Center Freiburg. He was Leader of the Intracranial EEG and Brain Imag- aResearchGroupLeaderwithV.BraitenbergattheMax-Planck-Institutefor ing Lab, Epilepsy Center, University Medical Center Biological Cybernetics in Tu¨bingen, Germany, a Visiting Professor with E. Freiburg, Freiburg, Germany. His main research in- Vaadia and M. Abeles, Hebrew University, Jerusalem, Israel, a Research Group terests include recording, analysis, and modeling of Leader with W. von Seelen, Ruhr-University, Bochum, Germany, and an Asso- cortical neuronal population activity across multiple ciate Professor at the Weizmann Institute of Science, Rehovot, Israel, working spatial scales, from microECoG to EEG and fMRI, with A. Arieli and A. Grinvald. He moved to Freiburg University, Germany, in and the application of such measurements in clinical 1996. There, his research interests include the recording, analysis, and mod- . elingof neuronal assembly activity in cortical networks, on the improvement of Dr. Ball is a Member of the executive board and a Principal Investigator in computational neuroscience methods, and on the development of brain–machine the Excellence Cluster BrainLinks-BrainTools at the University of Freiburg. He interfaces, including neuronal motor prostheses. is a Founding Member of the Bernstein Center Freiburg.

63 64 Published in NeuroImage March 2016. Featured on cover page

VII. THE ROLE OF BLOOD VESSELS IN HIGH-RESOLUTION VOLUME CONDUCTOR HEAD MODELING OF EEG

65 NeuroImage 128 (2016) 193–208

Contents lists available at ScienceDirect

NeuroImage

journal homepage: www.elsevier.com/locate/ynimg

The role of blood vessels in high-resolution volume conductor head modeling of EEG

L.D.J. Fiederer a,b,c,d,⁎,J.Vorwerke, F. Lucka e,f,l,M.Dannhauerg,m,S.Yangh, M. Dümpelmann a,c, A. Schulze-Bonhage c,d,A.Aertsenb,d, O. Speck h,i,j,k,C.H.Wolterse, T. Ball a,c,d a Intracranial EEG and Brain Imaging Lab, Epilepsy Center, University Hospital Freiburg, Germany b Neurobiology and Biophysics, Faculty of Biology, University of Freiburg, Germany c BrainLinks-BrainTools Cluster of Excellence, University of Freiburg, Germany d Bernstein Center Freiburg, University of Freiburg, Germany e Institute for Biomagnetism and Biosignalanalysis, University of Münster, Germany f Institute for Computational and Applied Mathematics, University of Münster, Germany g Scientific Computing and Imaging Institute, 72 So. Central Campus Drive, Salt Lake City, Utah 84112, USA h Dept. of Biomedical Magnetic Resonance, Otto-von-Guericke University Magdeburg, Germany i Leibniz Institute for Neurobiology, Magdeburg, Germany j German Center for Neurodegenerative Diseases (DZNE), Site Magdeburg, Germany k Center for Behavioral Brain Sciences, Magdeburg, Germany l Department of Computer Science, University College London, WC1E 6BT London, UK m Center for Integrative Biomedical Computing, University of Utah, 72 S. Central Campus Drive, 84112, Salt Lake City, UT, USA article info abstract

Article history: Reconstruction of the electrical sources of human EEG activity at high spatio-temporal accuracy is an important Received 2 October 2015 aim in neuroscience and neurological diagnostics. Over the last decades, numerous studies have demonstrated Accepted 22 December 2015 that realistic modeling of head anatomy improves the accuracy of source reconstruction of EEG signals. For exam- Available online 31 December 2015 ple, including a cerebro-spinal fluid compartment and the anisotropy of white matter electrical conductivity were both shown to significantly reduce modeling errors. Here, we for the first time quantify the role of detailed Keywords: FEM reconstructions of the cerebral blood vessels in volume conductor head modeling for EEG. To study the role of the 7 T MRI highly arborized cerebral blood vessels, we created a submillimeter head model based on ultra-high-field- Blood vessel modeling strength (7 T) structural MRI datasets. Blood vessels (arteries and emissary/intraosseous veins) were segmented Submillimeter volume conductor head model using Frangi multi-scale vesselness filtering. The final head model consisted of a geometry-adapted cubic mesh Forward problem with over 17 × 106 nodes. We solved the forward model using a finite-element-method (FEM) transfer matrix Inverse problem approach, which allowed reducing computation times substantially and quantified the importance of the blood EEG source localization vessel compartment by computing forward and inverse errors resulting from ignoring the blood vessels. Our re- Extended source model sults show that ignoring emissary veins piercing the skull leads to focal localization errors of approx. 5 to 15 mm. Large errors (N2 cm) were observed due to the carotid arteries and the dense arterial vasculature in areas such as in the insula or in the medial temporal lobe. Thus, in such predisposed areas, errors caused by neglecting blood vessels can reach similar magnitudes as those previously reported for neglecting white matter anisotropy, the CSF or the dura — structures which are generally considered important components of realistic EEG head models. Our findings thus imply that including a realistic blood vessel compartment in EEG head models will be helpful to improve the accuracy of EEG source analyses particularly when high accuracies in brain areas with dense vascu- lature are required. © 2016 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Introduction and Murray, 2012; Schneider, 1972; Opitz et al., 2011; Datta et al., 2013; Sadleir et al., 2010; Fernández-Corazza et al., 2013; Bayford and Realistic head models are important tools in neuroscience Tizzard, 2012; Vonach et al., 2012; Carter et al., 2005; Miller et al., (Pascual-Marqui, 1999; Michel et al., 2004; Grech et al., 2008; Michel 2010; Voo et al., 1996; Yang et al., 2009; Panzer et al., 2012; Wendel et al., 2009; Lau et al., 2014; Heers et al., 2012; Rampp and Stefan, 2007). The present paper focuses on realistic head models for EEG ⁎ Corresponding author at: Engesserstr. 4, 5th floor, EEG Lab AG Ball, 79108 Freiburg, Germany. research that are used as volume conductor head models (VCHMs) for E-mail address: lukas.fi[email protected] (L.D.J. Fiederer). computing the electric fields created by electrical sources in the brain.

http://dx.doi.org/10.1016/j.neuroimage.2015.12.041 1053-8119/© 2016 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

66 194 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208

VCHMs enable to study the influence of detailed anatomy on field high number of tight junctions between endothelial cells (Daneman, propagation (Opitz et al., 2011; Ramon et al., 2006; Haueisen et al., 2012), which should further decrease electrical conductivity. Thus, the 1997) and the optimal spatial sampling of EEG signals (Ramon et al., vessel-wall-related effects previously described in ECG modeling 2009; Slutzky et al., 2010; Srinivasan et al., 1998) and are essential for (Stinstra et al., 2005a, 2005b) may be even more important for the source localization (Pascual-Marqui, 1999; Michel et al., 2004; Grech blood vessels supplying the brain. et al., 2008; Michel and Murray, 2012; Schneider, 1972). To investigate the role of blood vessels in volume conductor modeling, For these applications, simplified spherical shell (Brazier, 1949; we needed to create a detailed reconstruction of the cerebral blood ves- Geisler and Gerstein, 1961; Frank, 1952; Wilson and Bayley, 1950; sels. 7 T MRI can detect blood vessels with a particularly high contrast- Hosek et al., 1978; Meijs and Peters, 1987) models can be used and to-noise ratio (CNR) (Maderwald et al., 2008) not achieved at lower solved with analytical methods, but they neglect the complex anatomy field strengths. We therefore built a VCHM including a detailed blood ves- of the head and the brain. Numerous studies have demonstrated that re- sel compartment based on submillimeter 7 T anatomical sequences. We alistic modeling of anatomical structures such as the skull (Dannhauer assessed the modeling errors induced byneglectingbloodvessels(arter- et al., 2011; Ramon et al., 2004; Chauveau et al., 2004; Lanfer et al., ies and intraosseous/emissary veins) by comparisons with the well- 2012a; Anwander et al., 2002; Ary et al., 1981; Cuffin, 1993; van den established effect of neglecting CSF, as well as with the effect of neglecting Broek et al., 1998; Vorwerk et al., 2014), the dura (Slutzky et al., 2010; the dura. In addition, the feasibility of using 7 T MRI data to build a submil- Ramon et al., 2014; Ramon, 2012), the cerebrospinal fluid (CSF) limeter VCHM needed to model near-microscopic blood vessels had not (Ramon et al., 2006; Haueisen et al., 1997; Slutzky et al., 2010; Ramon been investigated thus far. We therefore implemented this new approach et al., 2004; van den Broek et al., 1998; Vorwerk et al., 2014; Bangera to create the first submillimeter 7 T-based VCHM and solve it using a et al., 2010; Bénar and Gotman, 2002; Lanfer et al., 2012b; Rice et al., Finite Element Method (FEM) transfer matrix approach to minimize com- 2013; Vanrumste et al., 2000; Wendel et al., 2008)andheadextent putational load while maintaining minimal numerical errors. (Lanfer et al., 2012a; Bruno et al., 2003; Vatta et al., 2005)aswellasre- The present paper provides a detailed description of the methods alistic modeling of anisotropy (Chauveau et al., 2004; Anwander et al., used to create our submillimeter FEM model based on 7 T MRI data, in- 2002; Vorwerk et al., 2014; Bangera et al., 2010; Haueisen et al., 2002; cluding the extraction of the blood vessels using spatial filtering Güllmar et al., 2010; Wolters et al., 2006; Hallez et al., 2005, 2008, methods, describes the computational requirements for whole-head 2009; Rullmann et al., 2009; Wolters, 2003), particularly of the white submillimeter FEM modeling, and presents the forward and inverse matter, can substantially improve the accuracy of forward and inverse modeling results on the role of blood vessels in high-resolution volume modeling of EEG signals. The strong concerns related to anisotropy conductor head modeling of EEG. even prompted the development of new modeling methods to enable its implementation (Hallez et al., 2005; Wolters, 2003). Thus, most as- Methods pects of the cranial macro-anatomy have meanwhile been addressed in previous head modeling studies. 7 T MRI data acquisition and pre-processing One exception, though, is the role of cranial blood vessels for EEG forward and inverse solutions which has only been marginally ad- Whole-head 3-D Magnetization Prepared Rapid Gradient Echo dressed so far (Haueisen et al., 1997). As the influences of gray matter, (MPRAGE, T1-weighted) and 3-D Gradient Echo (GE, PD-weighted) se- white matter, CSF, dura and skull have all been addressed, blood vessels quences of one male subject (age: 27, right-handed, no history of neuro- might be the last uninvestigated widespread macroscopic structure psychiatric disease) were acquired on a Magnetom 7 T whole body MRI within the bounds of the skull. One reason for this has been the difficulty system (Siemens, Germany, Erlangen) at a 0.6-mm isotropic resolution in obtaining detailed reconstructions of the complex, highly arborized (Fig. 1a,b). Acquisition parameters are summarized in Table 1. cerebral blood vessels from available imaging data for VCHMs, in partic- The volumes were co-registered using SPM8 (freely available at ular without application of contrast agents. The role of blood vessels in http://www.fil.ion.ucl.ac.uk/spm/) with default parameters and T1 as VCHMs however deserves attention as (i) the brain is strongly reference. Additionally, a third dataset with a more homogenous brain vascularized and, hence, a large number of blood vessels of different cal- was created by dividing the T1 images by the PD images (Van de ibers are present throughout the skull and brain. Blood vessels not only Moortele et al., 2009). The T1/PD data was used for skull stripping and permeate the skull diploe but, at specific locations, directly pierce brain segmentation (cf. Supplementary Methods for a detailed descrip- through the skull bone. As in the case of nerve foramina and surgical tion of the segmentation procedure). skull holes (Chauveau et al., 2004; Lanfer et al., 2012a; van den Broek et al., 1998; Heasman et al., 2002; Bénar and Gotman, 2002; Li et al., Segmentation of blood vessels 2007; Oostenveld and Oostendorp, 2002; Sparkes et al., 2009; Thevenet et al., 1992; Vanrumste et al., 2000), these direct connections To segment cranial blood vessels (intracranial, intraosseous, and (foramina) between brain and head surface may significantly influence extracranial), we utilized a Frangi vesselness filter (Kroon, 2009). This the forward and inverse propagation of electrical fields. However, the filter is designed to enhance tubular structures, indicated by the eigen- impact of these skull foramina due to blood vessels on VCHMs has values of the Hessian of the image data at multiple spatial scales (Frangi thus far not been addressed. (ii) The conductivities previously used to et al., 1998; Manniesing et al., 2006). In our hands, this filter proved simulate blood vessels were quite high (0.417–1.25 S/m) (Haueisen well-suited for segmenting arteries and intraosseous/emissary veins, et al., 1997) and while these values appear appropriate for blood per but not as successful in detecting draining veins. This could be due to se, they may not be adequate for the blood vessel system as a whole, the draining veins’ geometry and lower contrast, because of slower as vessels also include the surrounding layer of endothelium. This blood flow compared to the arteries. Throughout the manuscript, we endothelium, among other tasks, serves as a diffusion barrier with low will use the term “blood vessels” when addressing all segmented electrical conductivity, preventing substances from freely entering and vessels, and “arteries” or “veins” otherwise. Blood vessels were segment- leaving the blood stream. The importance of taking into account the ed from the Frangi-filtered volumes with an in-house regional growth al- low electrical conductivity of blood vessel walls has recently been dem- gorithm (see Supplementary Methods for further details). Intraosseous onstrated for electrocardiogram (ECG) modeling (Stinstra et al., 2005a, vessels, including veins piercing through the skull via foramina, were 2005b). Although direct measurements comparing vessel wall resis- identified by computing the intersection between the blood-vessel and tance in the brain with that in the rest of the body are missing to our skull compartments. Results were manually inspected and compared knowledge, resistance of the former may be even more pronounced, with anatomy atlases (Benninghoff, 1993; Netter, 1987; Nowinski et al., as the endothelium there forms the brain–blood barrier (BBB) with a 2011) to ensure that only blood vessels were segmented. An axial slice

67 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 195

Fig. 1. 7 T structural MRI data and segmentation. (a) 7 T T1 MPRAGE MRI data at 0.6-mm isotropic resolution used to derive the volume conductor head model. Arteries are, for example, visible as bright tubular structures in the insular region (white box). Note that the dataset was acquired without any contrast agent. (b) 3 T T1 MPRAGE dataset obtained in the same subject (see Derix et al., 2014; Lüsebrink et al., 2013 for acquisition parameters). Arteries in the same region (white box) are not clearly visible. (c) Axial slice through the VCHM derived from the 7 T data by tissue segmentation. The white box again highlights the insular region as in (a) and (b). Segmented blood vessels are shown in red. Note that neither the hematopoetic nor the fatty bone marrow was included in the segmentation (see Methods section). (d) 3-D visualization of intracranial and intraosseous blood vessels (cf. Fig. 2 for a 3-D for visualization of extraosseous vessels); the black arrow indicates an example of an intraosseous vein. as well as a 3-D axial cut through the final head model segmentation are conductivity of cerebral blood vessels, i.e., including both vessel walls shown in Fig. 1(c) & (d), respectively. Fig. 2 shows an overview of all seg- and blood-filled vessel lumen. As it is not yet possible to treat vessel mented blood vessels, including major cerebral arteries and their ramifi- walls and lumina separately, we modeled them as one compartment cations (Benninghoff, 1993; Netter, 1987; Nowinski et al., 2011) and set the compound conductivity of this compartment to cover the range of possible scenarios described in the Introduction. Because it is Volume conductor head models highly unlikely that blood vessels as a whole could have a conductivity (σ) higher than that of blood alone (Haueisen et al., 1997), we used the To quantify and compare the model errors induced by ignoring latter as our upper limit in the high-σ-model. Similarly, it is highly blood vessels, the CSF, and the dura, we created one blood-vessel-free unlikely that the combination of blood vessel walls (endothelium) and model, three models including blood vessel, one CSF-free model and BBB would produce a conductivity lower than that of compact bone. two dura-free models (Fig. 3). Therefore, we used compact bone conductivity (Haueisen et al., 1995) The blood-vessel-free model was the model as described above, but as a lower extreme in the low-σ-model. Because the conductivity of car- without the blood vessels, which were replaced by the surrounding tissue diac blood vessel endothelium is known (Stinstra et al., 2005a, 2005b), types, i.e., soft tissue, fat, bone, dura, CSF, GM and WM, depending on the we used this conductivity for our intermediate-σ-model. vessel location. We shall refer to this model as the no-blood-vessel-model. Several previous studies have demonstrated the importance of the In the blood vessel model, all blood vessels derived from the imaging CSF on volume conduction. It is well established that neglecting the data as described in the preceding sections were implemented as one CSF compartment induces severe modeling errors. To directly compare blood vessel compartment. For volume conductor modeling, a conduc- model improvement by including CSF with model improvement by tivity value needs to be assigned to each volume conductor model com- including blood vessels, we generated a no-CSF-model by replacing partment. In contrast to other tissue types such as skin, bone, or gray CSF by gray matter in the no-blood-vessel-model. To also compare matter, there are no conductivity values in the literature for the total blood-vessel-related effects to those related to the dura, we replaced

Table 1 7 T MRI acquisition parameters.

Sequence TR TI TE Flip angle Bandwidth Field of view Voxel size

MPRAGE 2500 ms 1050 ms 2.87 ms 5° 150 Hz/pixel 230.4 mm × 230.4 mm 0.6 mm × 0.6 mm × 0.6 mm GE 1630 ms – 2.87 ms 5° 150 Hz/pixel 230.4 mm × 230.4 mm 0.6 mm × 0.6 mm × 0.6 mm

68 196 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208

Fig. 2. Blood vessels extracted from 7 T MRI by Frangi vesselness filtering and regional growth segmentation. The following cerebral blood vessels are indicated by numbers: (1) internal carotid arteries, (2) vertebral arteries, (3) basilar artery, (4) posterior arteries, (5) medial arteries, (6) anterior artery. (A) Part of the carotid artery above the foramen lacerum. Draining veins, due to the slow flow of their blood, produced insufficient signal for accurate segmentation and are thus not included. For orientation, the inset shows the outer surface of the head model from the same viewing angle as for the blood vessels in the main figure.

the dura of the no-blood-vessel-model by compact bone in the dura-as- memory (RAM) under Linux. For simulations, three different systems bone-model. Finally, as an alternative scenario of dura-related model were used: the same as for building the models, one with errors, the dura was replaced by CSF in the dura-as-CSF-model. Both 16 × 3.1 GHz cores and 256 GB RAM and one with 120 × 2.8 GHz dura models are included because, in our experience, the dura may be cores and 3 TB RAM, the latter two used to run multiple simulations in misclassified as either bone or CSF, depending on which MRI- parallel. weighting the segmentation is based on. Fig. 3 shows axial slices through the different models investigated. Placement of sources

FEM methods For forward EEG simulations, one St. Venant dipole (Wolters et al., 2007) was placed at the center of every gray matter mesh element of FEM forward calculations were computed with SimBio-NeuroFEM the full model (with blood and CSF compartments). The St. Venant di- (SimBio Development Group, 2012) using the Saint-Venant direct ap- rect approach has a high computational efficiency when used in combi- proach (Buchner et al., 1997; Wolters et al., 2007; Vorwerk et al., nation with a FEM transfer matrix (Wolters et al., 2004). To fulfill the St. 2012) based on geometry-adapted cubic meshes (Hartmann et al., Venant condition (Lanfer et al., 2012a; Vorwerk et al., 2014), all dipoles 2010) (cf. Supplementary Methods for details), which improve the pre- neighboring non-gray matter elements were discarded using a cision of the computed potentials by reducing the error due to parallelized version of the sb_check_sources function provided by unsmooth transition edges (Wolters et al., 2007). To achieve good FieldTrip (Oostenveld et al., 2011), resulting in 2,229,036 remaining di- RAM efficiency, we used a conjugate gradient solver with incomplete poles. Inverse localization was performed on a St. Venant-condition- Cholesky preconditioning (IC(0)-CG) (Lew et al., 2009). To maximize fulfilling 1.2-mm isotropic grid (278,565 dipoles). The dipoles were ori- the accuracy of our model, forward solutions were calculated with a re- ented normally to the local gray matter surface (see Supplementary sidual error in the order of 10−11. All models comprised the same Methods for more details). 17,606,835 nodes and 17,349,004 elements with an isotropic resolution Because the dipolar model of brain activity is best used when evalu- of 0.6 mm. To reduce simulation time, a transfer matrix for 329 EEG ating the effect of spatially smooth structures, like dura and CSF, and channel was calculated for each model (Wolters et al., 2004). The posi- blood vessels are heterogeneously distributed within the brain, an tions of the 329 electrodes were defined according to the 10-5 system extended source model could better approximate the effects to expect (Oostenveld and Praamstra, 2001) using the MATLAB script kindly pro- in vivo. Therefore we generated a second source space where the vided by Giacometti et al. (2014) on their website. Conductivity values activity of each entry was taken as the sum of all dipoles within a corti- of the different tissue compartments are listed in Table 2. cal area of approx. 6 cm2 which is often assumed to be the area of cortex For building the models, we used a workstation with 4 × 2.8 GHz required to be active to generate scalp-visible effects (Cooper et al., cores central processing units (CPU) and 16 GB of random access 1965).

69 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 197

Fig. 3. Volume conductor head models investigated. No-blood-vessel-model: Model without any blood vessels, all other segmented tissues are included. Blood vessel model: As before, but with blood vessels. This model was used with three different blood vessel conductivities (see Methods section). No-CSF-model: As the no-blood-vessel-model, but with CSF replaced by gray matter. Dura-as-bone-model: As the no-blood-vessel-model, but with dura replaced by compact bone. Dura-as-CSF-model: As the no-blood-vessel-model, but with dura replaced by CSF. Color-coding as in Fig. 1. Note that the holes in the rendering of the no-CSF-model are due to the very thin 3D slice used, combined with the geometry-adapted mesh described below. These holes are not present in the full volume model.

Error measures The first error measure investigated was the relative difference mea- sure (Lew et al., 2009; Meijs et al., 1989)(RDM),defined as To quantify and compare the effects of ignoring blood vessels, CSF and dura, we calculated threeerrormeasurescommonlyused 2 in the modeling literature. In the following, “reference model” n ref test RDM vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii − i 1 always refers to the more detailed model of a tested pair and u i 10 n 2 n 1 ¼ u ¼ ref test2 ð Þ u j 1 j j 1 j the “test model” to the less detailed model, which is responsible uX B ¼ ¼ C for the investigated error. Seven model pairs were tested, t @qXffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qXffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA which were the no-blood-vessel-model paired with each other where n is the number of electrodes, and refi and testi are the voltages of model. all sources at the ith electrode in the reference model and the test

Table 2 Overview of algorithms and MRI data used for the segmentation of each model compartment. Additionally, the conductivities used for FEM simulations and references for these values are given.

Compartment Segmentation MRI data Conductivity (σ = S/m) References

White matter FAST MPRAGE 0.1429 Haueisen et al. (1995) Gray matter FAST MPRAGE 0.3333 Haueisen et al. (1995) Liquor FAST MPRAGE 1.5385 Haueisen et al. (1995) Blood vessels Frangi filtering + regional growth MPRAGE+GE 0.6250 (high-σ) See Volume conductor head models section 0.02 (intermediate-σ) 0.0063 (low-σ) Dura Masking GE 0.0650 Manola et al. (2005) Compact bone BET2 GE 0.0063 Haueisen et al. (1995) Fat Thresholding MPRAGE 0.0400 Haueisen et al. (1995) Eye Regional growth MPRAGE 0.5051 Haueisen et al. (1995) Soft tissue Regional growth Binary 0.1736 Haueisen et al. (1995) Internal air Regional growth MPRAGE 0.0020 Haueisen et al. (1995) Skin Isosurface Binary 0.4348 Haueisen et al. (1995)

70 198 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 model, respectively. The RDM is used to quantify forward errors and Results was calculated for all 2,229,036 cortical sources of each source model.

In some publications the subtraction of the L2 norms is inverted (test- In the present study, for the first time, a FEM VCHM with an isotropic ref instead of ref-test). From a mathematical point of view this makes submillimeter resolution including a detailed blood vessel compart- no difference and is irrelevant for comparability. ment and skull foramina was used for forward and inverse modeling The second error measure was the goal function scan localization (Figs. 1(c), (d), 2&3). In the following, we will present the forward error (Mosher et al., 1992), defined as and inverse simulation results and also describe the computational requirements of submillimeter FEM modeling.

L test i; i0 Effect of blood vessels GfPos test argmini testi− Á Án Li; 2 ðÞ¼ 0vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiL2 Á Á1 ð Þ u j 1 i; j Bu ¼ C @t X A To understand the role of blood vessels in volume conductor head modeling, three scenarios with different blood vessel conductivities GfError Pos ref −GfPos test 3 ¼ ðÞ ðÞ ð Þ were investigated. In the first one, the high-σ-model, blood vessels were attributed the conductivity of blood (Haueisen et al., 1997). In where GfPos(test) is the position in the source space of the test models the second scenario, the intermediate-σ-model, the conductivity of where the goal function scan is minimal for the ith source, Pos(ref)is the cardiac endothelium was used (Stinstra et al., 2005a). In the third the position in the source space of the reference source and GfError is case, the low-σ-model, blood vessels were modeled with conductivity the Euclidian distance between Pos(ref)andGfPos(test), also known as of bone as the lower extreme. This wide range of conductivities was the localization error, test is the voltages at all electrodes of the ith i used to ensure that effects induced by the real bulk conductivity of cere- source, L is the leadfield matrix of the reference model for the ith i,· bral blood vessels, which can be expected to be somewhere in this spec- source and all electrodes, and n is the number of electrodes. The locali- trum, will be accounted for. To ensure that we did not overestimate the zation error is used to quantify the inverse error and was calculated for a effects of blood vessels due to the use of single dipolar sources we also 1.2-mm grid comprising 278,565 sources, again for both source models. calculated the results for an extended source model (cf. Methods sec- The number of sources was reduced for this error measure because of its tion). Results obtained with dipolar and extended sources were mostly high computational load. As sources were reconstructed using identical very similar regarding the conclusions of this paper. The reported grids perfect source localization (zero localization error) is possible, results thus refer to both source models if not otherwise stated. making our estimation of the inverse error conservative. Because The simulations produced one EEG topography for each model and sources were always reconstructed in a test vs. reference model setting, dipole. The EEG topographies resulting from selected dipoles (with the implying that reconstruction was always performed in a model other 100th strongest RDM) for all models are shown in Fig. 4. The change than the one used for forward simulation, this is not an inverse crime in topographies induced by introducing blood vessels and varying (Kaipio and Somersalo, 2007). their conductivity are quite noticeable for the presented example of We also calculated the logarithmic magnitude error (lnMAG), the topographies with the 100th strongest RDM error for each model. defined as As can be seen, blood vessel-related topography changes become visible to the bare eye above an RDM of approx. 0.2. Following Lanfer and col- n test2 leagues, we consider errors with an RDM value N= 0.1 and/or a i 1 i lnMAG ln ¼ 4 0qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffin 1 mislocalization N= 5 mm as non-negligible (Lanfer et al., 2012a). ¼ X ref 2 ð Þ i 1 i RDM and goal function scan localization errors were computed B ¼ C @qXffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA against a model without any blood vessels (Figs. 5–8). Maximal error, where testi and refi are the voltages of all sources at electrode n for the mean error, the proportion of affected sources and the 5th, 50th and test model and the reference model, respectively. This error measure 95th percentiles of the error distributions are summarized in Table 3. did not provide any additional insights to the other two error measures Forward and inverse errors of both source models showed a similar and was therefore later omitted (Lanfer et al., 2012a). general picture. With the high-σ-model, non-negligible (see above) er- RDM, localization errors and lnMAG were computed using in-house rors were mainly located directly adjacent to points with blood vessels Matlab scripts (The MathWorks Inc., Natick, MA, USA). Because the either passing through or within the skull (emissary or intraosseous RDM is bounded between 0 and 2, it can be converted into a percentage vessels, respectively) (Figs. 5(a) & 7(a)), namely 5 vessel-related skull by dividing by 2 and multiplying by 100. For more information regard- foramina and 3 intraosseous veins. The foramina were the parietal em- ing these error measures, we refer to Lew et al. (2009), Meijs et al. issary foramen, the paired carotid canals, parts of the paired foramen (1989),andMosher et al. (1992)). lacerum, parts of the paired foramen spinosum and two symmetrical fo- ramina located above the anterior part of the Sylvian fissure (Netter, Analysis of the impact of local blood vessel density on errors 1987). The paired intraosseous veins were the venae diploicae frontalis, temporalis posterior and occipitalis (Netter, 1987). The segmentation of To quantify the influence of the local blood vessel density on errors, a the former vein also included the entry and exit parts of the canales multi-scale rank correlation analysis was performed. This analysis was diploici (Benninghoff, 1993; Netter, 1987). designed to answer the question: blood vessels at which spatial scale With the high-σ-model, non-negligible errors were also found close around a source are relevant for the observed errors? To this end, the er- to the major brain arteries (anterior, lateral and posterior arteries) and rors observed at all source positions were correlated with the local their branches (Figs. 5(a) & 7(a)). For the intermediate-σ-model, blood vessel density at these positions, both for the forward and inverse some non-negligible errors were still found close to emissary or error measures, using Spearman's rho (Best and Roberts, 1975). The intraosseous vessels, but errors mainly clustered around major and local blood vessel density was obtained from spherical kernels around minor arteries (Figs. 5(b) & 7(b)). Finally, for the low-σ-model, non- each position, their diameters ranging between 0 mm and 100 mm negligible errors were no longer found close to emissary or intraosseous (multiples of the model resolution, 0.6 mm). Local blood vessel density vessels. Instead, errors now clustered strongly around major and minor was expressed as the ratio of blood vessel elements within the kernel to arteries (Figs. 5(c) & 7(c)). all elements within the kernel. Local blood vessel density was chosen as With both source models (dipolar and extended sources), the over- measure because of its invariance against blood vessel size (discussed in all strongest and most widespread errors were observed for the region Impact of source size section). of the carotid arteries. An example of the EEG topography differences

71 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 199

Fig. 4. Effect of the different head models on forward-calculated EEG topographies. (a) Location and orientation of the selected example dipoles in sagittal and coronal views (anterior point of view) indicated by cyan cones. Red and yellow: intracranial and intraosseous vessels, respectively. (b–h) Forward calculated EEG maps resulting from the dipoles shown in (a) and obtained with the no-blood-vessel-model (b; with overlaid electrode layout), with the high-σ-model (c), intermediate-σ-model (d), low-σ-model (e), no-CSF-model (f), dura-as-CSF- model (g), and dura-as-bone-model (h). RDM errors of the EEG maps relative to the no-blood-vessel model are indicated in the upper left corner above each EEG map. In each column, the model used to select the example dipole is highlighted by a light-gray box. In each case, the dipole producing the 100th strongest RDM error with the indicated model was selected.

for a dipole in this region and with the different models investigated is Impact of local blood vessel density on errors shown in Fig. 4, first column. Other areas with dense vasculature and pronounced errors included the anterior cingulate, the insula, and the The spatial error distributions as shown in Figs. 5–8 indicated a close medial temporal lobe (Figs. 4–8). spatial relation of local vessel density and error magnitudes for the

72 200 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208

Fig. 5. Spatial distribution of non-negligible errors induced by ignoring blood vessels: RDM errors of dipolar sources. Color and size of spheres represent RDM error at source positions. Transparent gray and yellow: brain and skull blood vessels, respectively Note the non-negligibly affected sources along small vessels (e.g., black box). As draining veins, such as the sagittal sinus, were not included in our model, there are no corresponding errors. (a) Results obtained with the high-σ-model, (b) the intermediate-σ-model, and (c) the low-σ-model, all in cor- onal and sagittal views. Black arrows: errors due to skull foramina and intraosseous vessels. intermediate- and low-σ-vessel-models, while the spatial distribution high values and broad spatial distribution of modeling errors are in of errors in the high-σ model appeared to be dictated by the position accordance with the literature (Ramon et al., 2006; Haueisen et al., of vessels penetrating the skull. To quantify these relations, we 1997; Slutzky et al., 2010; Ramon et al., 2004; van den Broek et al., performed a correlation analysis across multiple spatial scales. This con- 1998; Vorwerk et al., 2014; Bangera et al., 2010; Bénar and Gotman, firmed the visual impression of a strong relationship between local 2002; Lanfer et al., 2012b; Rice et al., 2013; Vanrumste et al., 2000; blood vessel density and error measures (cf. Fig. 9) for both low- and Wendel et al., 2008). intermediate-σ-models. For these models, correlations became Forward-calculated EEG results reflecting errors made by ignoring maximal with kernels of 20- to 30-mm diameters for forward and the dura (replaced by compact bone and CSF, respectively) are summa- inverse errors, respectively, indicating a critical spatial scale with the rized in Table 3 and shown in Fig. 4 (g) & (h). Ramon et al. (2014) and highest relevance of local blood vessel density to VCHM modeling (if Ramon (2012) have reported lower forward errors (0.057 mean RDM) the low-to-medium conductivity assumption is correct). Expectedly, er- when replacing the dura with CSF using dipolar sources. To the best of rors obtained in the high-σ-scenario did not show a strong correlation our knowledge, no investigation considering replacing the dura with of errors with local blood vessel density. compact bone exists for comparison, although such segmentation errors may occur. Effect of CSF and dura on modeling errors Computational requirements of submillimeter head modeling To put vessel-related errors in relation to other model errors, we ex- amined errors due to ignoring the CSF and dura. Forward-calculated The main criteria for the computational feasibility of forward and EEG maps reflecting errors made by ignoring the CSF (results for dipoles inverse EEG modeling are the computation time and the amount of with the 100th strongest RDM) are shown in Fig. 4(f). The changes in to- memory needed. With the current implementation (cf. FEM methods pographies induced by replacing CSF by gray matter were, as expected, section), computing one row of the transfer matrix (Wolters et al., pronounced (Table 3). Overall, forward and inverse errors showed 2004), corresponding to one EEG electrode, took approx. 24 min. Com- similar distributions. Non-negligible (N=5 mm or N=0.1 RDM) errors putation of the whole transfer matrix (a matrix with approx. were found throughout the source spaces, with clusters of higher 329 × 17 Mio. entries) for all 329 electrodes thus lasted 133.5 ± 3.8 h values, often on gyral crowns. Similar results have been reported by (mean ± std). After having calculated the transfer matrix (only once Lanfer and colleagues using dipolar sources (Lanfer et al., 2012b). The per model and sensor-configuration), one forward simulation could

73 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 201

Fig. 6. Spatial distribution of non-negligible errors induced by ignoring blood vessels: localization errors of dipolar sources. Cone bases are at the true source localization, cone tip is at the erroneous localization due to ignoring blood vessels. As seen for the forward errors, note the non-negligibly affected sources along small vessels (e.g., black box). Other conventions as in Fig. 5. then be performed in just approx. 120 ms per dipole. For all 2,229,036 of the type of source model (dipole, extended). The maximal inverse dipoles, the forward simulation thus lasted 74.5 ± 0.6 h. Times are errors, however, were considerably larger with the extended source given for a solver residual error in the range of 10−11 on a 2.8 GHz model than with the dipoles (discussed in Impact of source size CPU and may vary according to the geometrical complexity of the section). There also were more strongly affected inverse localizations models. No more than 30.5 GB of RAM were required for any operation. (as indicated by the large red cones in Fig. 8) in the high- than in the intermediate- and low-σ simulations of extended sources. The conduc- Discussion tivity of blood vessels, which we varied in our simulations over two or- ders of magnitude, appeared to only marginally influence the strength In the present study, we investigated the role of a detailed recon- of the dipolar errors, while the extended source errors were stronger struction of blood vessels in a submillimeter VCHM. This was made pos- for both low- and intermediate-σ-model. The percentage of non- sible by the use of anatomical submillimeter 7 T MRI data. Before such negligibly affected sources (RDM N= 0.1, localization error N= data became available, specific diffusion weighted sequences and con- 5 mm), however, showed much stronger variations. More than twice trast agents had to be used to create angiograms. Presumably for this as many sources were non-negligibly affected in the intermediate- reason, the effect of blood vessels on forward and inverse modeling and low-σ-models than in the high-σ-model (Table 3). This can be ex- has, up to now, never been investigated in detail. In the following, we plained by the high deviation of the intermediate and low conductivities discuss the results of the different conductivity scenarios and the from those of the surrounding brain tissue, which was not the case in modeling errors induced by ignoring the blood vessels located within the high conductivity scenario. the skull. Furthermore, we compare our simulation results to the litera- The error measure results summarized in Figs. 5–8 showed two ture and make suggestions on how to improve computational speed. Fi- distinct spatial error patterns: (i) Errors clustering around cerebral nally, we discuss limitations and perspectives of our work. arteries and (ii) errors clustering in the vicinity of skull foramina and intraosseous vessels. The latter error type was mainly present Errors with different blood vessel conductivity in the results obtained with the high-σ-model, while the former type errors were present in all 3 cases (all σ-models), but much Our findings, as summarized in Table 3, showed similar mean and stronger in results obtained with both intermediate- and low-σ- percentile errors irrespective of the conductivity σ (high, intermediate, models, reflected in the different percentages of affected sources as low) assumed for the blood vessel compartment, and also irrespective discussed above.

74 202 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208

Fig. 7. Spatial distribution of non-negligible errors induced by ignoring blood vessels: RDM errors of extended sources. Compared with the results obtained with the dipolar sources, there were fewer non-negligibly affected sources along small blood vessels (e.g., black box), while errors in vessel-rich areas were not diminished. Conventions as in Fig. 5.

Error clusters around arteries were widely distributed, affected the non-negligible errors (RDM N= 0.1 and localization errors N=5mm medial temporal lobe and followed the paths of the three major brain (Lanfer et al., 2012a)), although in a highly localized manner. arteries: the anterior cerebral artery, the middle cerebral artery and Results from spatial multi-scaled correlation of errors with local the posterior cerebral artery. As the arteries branched into smaller ves- blood vessel density (Fig. 9) also pointed to the different mechanisms sels, the errors became smaller until they vanished completely, which underlying the error generation in the high-σ-vessel-model compared happened earlier (at larger vessel diameters) for extended than for di- with the low- and intermediate-σ models. Blood vessel densities at polar sources (Figs. 5–8). The cingulate and insular cortices were strong- the scale of 20 to 30 mm, i.e. still mostly within the skull, correlated ly affected because of their dense vasculature. Because the draining best with forward and inverse errors of both low- and intermediate- veins and superficial cortical vessels were not included in the model σ-models. Forward and inverse errors related to the high-σ-model (cf. Limitations and further perspectives section), the outer surface of were, however, not strongly correlated with local blood vessel density, the cortex was less affected, with errors mainly at the frontal pole and but rather appeared dominated by errors due to vessels piercing the at the intersection of parietal, occipital and temporal cortices (TPO skull (Figs. 5–8), highlighting the different error mechanisms with dif- area). Including these missing vessels can be expected to further ferent vessel conductivities and a need for experimental clarification increase the number of affected areas and could also induce interesting of this issue (see Conclusions & outlook section). edge and tunneling effects as some of them pass through the CSF (with high conductivity) and some through the dura (with low conductivity). Blood-vessel-related errors in relation to previously described modeling Errors clustering in the vicinity of blood vessel skull foramina and errors intraosseous vessels (black arrows in Fig. 5 (a) & (b)) were most pro- nounced in the region in the vicinity of the carotid canal. Errors here To relate our findings to previously investigated modeling errors, we may affect source reconstruction in the medial and basal temporal compared our results obtained with the no-CSF-model, the dura-as- lobe, which is of interest in the context of mesial-temporal epilepsy bone/CSF-models and reports by two recent publications (Lanfer et al., (Waberski et al., 2000; Jung et al., 2009; Merlet et al., 1996; Assaf and 2012a; Güllmar et al., 2010) in which detailed error measures such as Ebersole, 1997; Merlet et al., 1998; Fernández-Torre et al., 1999a; RDM and localization error were given. Fernández-Torre et al., 1999b; Aydin et al., 2015; Aydin et al., 2014). The remaining blood vessel skull foramina and intraosseous veins CSF, dura and skull were in most cases too small (Lanfer et al., 2012a) to induce strong Ignoring the CSF caused similar maximal errors as ignoring vessels and widespread errors, despite being located between sources and (Table 3), but a larger mean error and a higher proportion of affected electrodes (Lanfer et al., 2012b). Nevertheless, most of these produced sources. The critical positioning of the CSF between sources and

75 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 203

Fig. 8. Spatial distribution of non-negligible errors induced by ignoring blood vessels: localization errors of extended sources. As in the case of forward errors (Fig. 7), non-negligibly affected sources along small blood vessels were reduced (e.g., black box). In vessel-rich regions, localization errors were magnified (large red cones, cf. Fig. 6). Conventions as in Fig. 6.

electrodes together with its large extend is the main reason why not in- et al., 1998; Vorwerk et al., 2014). Lanfer et al. (2012a) published a thor- cluding it creates such strong errors (Lanfer et al., 2012b), as confirmed ough investigation of the influence of skull segmentation inaccuracies by our results and in line with a large number of previous modeling on EEG forward and inverse problems, including effects due to skull studies (Ramon et al., 2006; Haueisen et al., 1997; Slutzky et al., 2010; holes, under- or overestimating skull thickness, or neglecting skull Ramon et al., 2004; van den Broek et al., 1998; Vorwerk et al., 2014; sinuses (cf. Table 3). Among these errors, those caused by ignoring a Bangera et al., 2010; Bénar and Gotman, 2002; Lanfer et al., 2012b; skull hole with a 10-mm diameter were most similarly to the errors Rice et al., 2013; Vanrumste et al., 2000; Wendel et al., 2008) and recent that we observed in relation to cerebral blood vessels. Lanfer and experimental findings (Rice et al., 2013). colleagues recommend that skull hole larger than 2 mm should be Replacing the dura by compact bone or CSF caused maximal model included in EEG head models. errors quite similar to those due to blood vessels (Table 3) but again with a larger spatial extent, probably for similar reasons as discussed Anisotropy for the case of the CSF above. Our present results confirm that the Another widely discussed source of errors in head modeling are an- dura plays a major role VCHM accuracy (Slutzky et al., 2010; Ramon isotropic conductivities. Several authors (Anwander et al., 2002; et al., 2014; Ramon, 2012) and that the inclusion of the dura is nearly Vorwerk et al., 2014; Bangera et al., 2010; Haueisen et al., 2002; as important as that of the CSF. Güllmar et al., 2010; Wolters et al., 2006; Hallez et al., 2005, 2008, In summary, on the whole-brain scale, CSF and dura are more impor- 2009; Rullmann et al., 2009; Wolters, 2003) have described the influ- tant for VCHM accuracy than blood vessels. On the other hand, local ence of white matter anisotropy in this context. The study by Güllmar errors due to ignoring blood vessels were on par with those due to ig- et al. (2010) is especially detailed and is therefore used here to compare noring CSF or dura (Table 3; Fig. 5) indicating that for critical regions our results with respect to the forward error measures. Güllmar and col- with dense vasculature and/or close to vessels piercing the skull, source leagues used a different inverse approach than Lanfer et al. (2012a) and localization directed at these areas may profit from including blood we did and to the best of our knowledge no study of anisotropy with a vessels as much as from modeling the CSF or dura. comparable inverse error metric exists. Inaccurate modeling of skull geometry has also been repeatedly The 95th and 50th percentiles of the RDM values, closest to ours, ob- reported to be a common source of model errors (Dannhauer et al., tained by Güllmar and colleagues with anisotropic models are listed in 2011; Ramon et al., 2004; Chauveau et al., 2004; Lanfer et al., 2012a; Table 3. When comparing the RDM values, it becomes apparent that Anwander et al., 2002; Ary et al., 1981; Cuffin, 1993; van den Broek the effect of including blood vessels is comparable to the effects due to

76 204 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208

Table 3 RDM and goal function scan localization error. Overview of all test models of this paper, together with selected models of Lanfer et al. (2012a) and Güllmar et al. (2010) (details in Blood- vessel-related errors in relation to previously described modeling errors section).

RDM Localization error

Model Max (unitless) Mean (unitless) N=0.1 Max (mm) Mean (mm) N=5 mm

This paper Dipolar sources High-σ 1.354 0.018 0.851% 23.546 0.148 0.676% Intermediate-σ 1.622 0.017 2.100% 29.686 0.242 1.322% Low-σ 1.651 0.017 2.316% 29.686 0.258 1.436% No-CSF 1.567 0.148 61.331% 35.211 3.498 27.605% Dura-as-bone 1.91 0.107 34.244% 49.623 2.715 21.859% Dura-as-CSF 1.66 0.093 32.872% 35.638 2.748 23.026% Extended sources High-σ 1.199 0.018 0.72% 37.355 0.152 0.495% Intermediate-σ 1.661 0.017 1.617% 61.948 0.265 0.944% Low-σ 1.718 0.017 1.83% 61.948 0.278 1.017% No-CSF 1.974 0.173 68.365% 68.442 3.398 19.045% Dura-as-bone 1.753 0.114 39.722% 57.161 2.204 13.507% Dura-as-CSF 1.107 0.092 33.019% 59.494 2.539 15.226%

Lanfer et al. (2012a) Segmentation defects 10 mm skull hole (1c) 0.889 0.016 1.905% 9.314 1.073 0.343% 4 mm constant skull & scalp (6a) 1.399 0.12 49.403% 27.1431 5.738 45.220% 6 mm constant skull & scalp (6b) 1.399 0.091 29.236% 28.227 3.748 23.824%

Model 95th percentile 50th percentile 5th percentile

This paper Dipolar sources High-σ 0.042 0.013 0.006 Intermediate-σ 0.06 0.008 0.003 Low-σ 0.063 0.008 0.003 No-CSF 0.342 0.123 0.037 Dura-as-bone 0.272 0.077 0.033 Dura-as-CSF 0.186 0.084 0.037 Extended sources High-σ 0.039 0.013 0.006 Intermediate-σ 0.061 0.008 0.003 Low-σ 0.063 0.008 0.003 No-CSF 0.407 0.142 0.040 Dura-as-bone 0.288 0.084 0.034 Dura-as-CSF 0.171 0.085 0.039

Güllmar et al. (2010) Anisotropic transversal:lateral ratios 1:2 0.064 0.018 0.004 1:10 0.265 0.071 0.016 1:100 0.643 0.191 0.050

a 1:2 transversal to longitudinal anisotropy ratio, which may be a realis- Impact of source size tic value as suggested by a number of recent studies (Bangera et al., 2010; Güllmar et al., 2010; Hallez et al., 2008; Wolters, 2003). For exam- We compared modeling results with dipolar (point-like) and ex- ple, Bangera et al. (2010) compared simulations of anisotropic models tended (surface of approx. 6 cm2) source models, respectively. Results with, among others, ratios between 1:2 to 1:10 with in-vivo intracortical obtained with these source sizes both support our general conclusions electrical stimulation measurements in epilepsy patients. They could regarding the importance of blood vessels in volume conductor head conclusively show that the 1:10 ratio fitted worst to the data for all modeling of EEG. However, there were also more subtle differences in four measured patients. On average, the best fitting ratio was 1:2. the error patterns, providing interesting insights on how source model Thus, ignoring blood vessels may cause similar forward errors than ig- size and VCHM structures interact and shape forward and inverse noring white matter anisotropy, at least with a presumably realistic solutions. transversal to longitudinal anisotropy ratio. With all other parameters kept constant, one might expect that a It is, however, important to keep in mind that our forward and structure would have maximal local effect onto forward and inverse er- inverse errors were probably underestimated as the majority of rors onto sources with a matching spatial extend, thus interpreting the superficial cortical vessels as well as the veins could not be included volume conductor as a spatial filter according to the principle of the in our model (cf. Limitations and further perspectives section). matched filter theorem (Rosenfeld and Kak, 1982). For example, in Furthermore, because of the use of identical source grids for forward our simulations, this would mean that dipolar sources, which have and inverse modeling, our localization errors are conservative (see close to no spatial extent, would be expected to have maximal effect Methods section). We can, therefore, conclude that, regardless of in the vicinity of small structures, like small blood vessels. Larger ex- the conductivity and of the source model used, blood vessels cause, tended sources would be expected to have maximal effect when com- on a local scale, errors that are comparable with errors produced by bined with larger structures, like large vessels, or other large-scale ignoring anisotropies, unrealistic modeling of the skull, and ignoring spatial smooth structures as the CSF or dura compartment. This is in- the CSF or the dura. deed what we observed from the percentage of non-negligible forward

77 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 205

five days. However, this computation step only needs to be performed once per model and sensor-configuration. Afterwards, forward simula- tion can be performed in just about a hundred of milliseconds per dipole. For our high source space resolution with more than 2 Mio. nodes, the computation of the leadfield for all dipoles still took 3 days. Without calculating a transfer matrix beforehand, one forward simula- tion for a model with about 17 Mio. unknowns would have lasted ap- proximately half an hour, which would have resulted in an excessive computational amount of more than 70 years. The transfer matrix tech- nique (Wolters et al., 2004) was hence crucial for the computational feasibility of our study. In the future, computation times may still significantly be reduced: for example, a lower IC(0)-CG solver accuracy might be sufficient for nearly all applications (Lew et al., 2009) which would be interesting to evaluate. The setup of the transfer matrix could be sped up by more than ten-fold when using the faster Algebraic MultiGrid preconditioned Conjugate Gradient (AMG-CG) FEM solver (Lew et al., 2009; Wolters et al., 2002; Stüben, 2001), at the cost of higher memory usage in the current implementation in SimBio-NeuroFEM. Parallelization on dis- tributed memory machines (Wolters et al., 2002; Krechel and Stüben, 2001) could still significantly reduce both computation time and mem- ory load. Most importantly, in routine source analysis scenarios, usually no more than 30,000 source space nodes are used, which would reduce the forward modeling computation time from 3 days down to about an hour. With such optimization, together with increased hardware performance, we anticipate that sub-mm FEM head modeling may become amendable for routine applications in science and neurological diagnostics.

Limitations and further perspectives

Several limitations have to be considered when interpreting the pre- sented results. First, our results are based on only one subject, and blood vessels show inter-individual variability (Benninghoff, 1993; Boyd, 1930; van der Zwan and Hillen, 1991; Tatu et al., 2012). Yet, the general Fig. 9. Rank correlation between error measures and blood vessel density at multiple layout of the cerebral vasculature is quite similar across individuals, spatial scales. The diameter of the spherical kernels used to determine the local blood both with respect to the major vessels and the location of brain regions vessel density was varied between 0 and 100 mm. Note the calculation of local blood with a dense vasculature, such as the insular region (Benninghoff, 1993; vessel density included vessels of all sizes; thus high values may indicate both, the presence of large vessels, or local clusters of many small vessels. (a) Results for forward Netter, 1987). Hence, as the strongest errors were located in these re- errors, and (b) inverse errors. gions, we expect that vessel-related errors will be present at similar levels and locations in other subjects as well. Second, the accuracy of the presented model could still be improved. errors (Table 3). A similar effect was also observed in the spatial distri- As mentioned before, few superficial cortical and dura vessels and no bution of errors throughout the volume conductor. As highlighted by draining sinuses (Nowinski et al., 2011) were included in the model be- the black boxes in Figs. 5–8,non-negligibleerrorsduetodipolarsources cause of their lower CNR. Incorporating these vessels is expected to even aligned along small blood vessels and mostly disappeared when further increase the proportion of the potentially-affected brain regions, switching to extended sources. In contrast, non-negligible errors of the particularly in the cortex, which would be highly relevant for source re- large sources close to large blood vessels were enhanced. Our findings construction. We expect that, due to blood-volume conservation, in- point towards complex interactions between spatial properties of cluding missing veins into our model would substantially increase the source and volume conductor models, which have received little atten- volume occupied by blood vessels. Such extended models could use sus- tion so far but may be practically important, as not all brain activation ceptibility weighted imaging data at 7 T, for segmenting veins. Also co- may be well approximated by dipolar sources and may rather involve registration of a 7 T blood vessel atlas (Nowinski et al., 2011) with our a wide range of different spatial scales (Ball et al., 2012). model could possibly enable us to better evaluate the true extent of blood vessels to be included in an enhanced model. Likewise, not all Computational requirements of submillimeter head modeling blood vessel foramina and intraosseous veins could be segmented in our current model, resulting in a likely underestimation of the resulting We showed that FEM modeling based on submillimeter 7 T MRI data modeling errors. We segmented 4 out of 9 and 3 out of 4 previously de- with more than 17 Mio. voxels is possible with current workstations scribed foramina containing blood vessels (Boyd, 1930; Benninghoff, and using Open-Source software (cf. FEM methods section). Improving 1994) and intraosseous veins (Benninghoff, 1993; Netter, 1987), the speed and memory usage of FEM computations is an important goal respectively. The foramina mastoide, condyloide, vesalius, caecum and in FEM research (Lew et al., 2009; Wolters et al., 2004; Nuno et al., 1997; squamosale as well as the venae diploicae temporalis anterior could not Wolters et al., 2002). With the chosen solver technique and parameter- be segmented. This might be due to the interindividual variability of izations and the current implementation in SimBio-NeuroFEM, comput- diploe veins (Benninghoff, 1993) and foramina size and location ing one row of a transfer matrix (Wolters et al., 2004) in a model with (Boyd, 1930). about 17 Mio. nodes took about half an hour, resulting in an overall Moreover, there are several areas where the current segmentation computation time for the full 329 electrodes transfer matrix of about could still be improved. For example, hyperintensities in the temporal

78 206 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 lobe and local susceptibility artifacts above the lamina cribosa of the which led us to model a wide range of conductivity values in the present ethmoid sinuses created small segmentation errors. The spongy bone, study. here modeled as intraosseous and emissary veins, could be further im- Beyond EEG, we can envision multiple applications which could proved. The choroid plexus was modeled with gray matter conductivity benefit from modeling blood vessels, also at submillimeter resolution. for lack of tissue specific values, but due to the deep location of the plex- For example, submillimeter head modeling could be especially well us we expect small model errors. Other areas with possible segmenta- suited for modeling of transcranial magnetic/direct current/alternating tion improvements are due to the lower CNR in the ventral part of the current stimulation to optimize the current flow in targeted brain imaging volumes (below cortex levels) and affected facial bones, buccal areas (Wagner et al., 2014). Other applications like traumatology and air, muscle and the spinal cord (the last two were completely left out of fNIRS could profit even more from the precise modeling of blood the model). Manual segmentation by neuroradiologists (current gold vessels. Furthermore, fMRI acquired at 7 T could make use of the high standard) could probably have recovered most of the missing tissues, blood vessel contrast in anatomical data to mask BOLD effects arising but is impractical for whole head segmentation with a submillimeter from superficial cortical vessels which are often misinterpreted as corti- resolution. Advances in high-field imaging, MR sequence development cal activity. and creating automated segmentation software optimized for 7 T MRI data should level these limitations in the near future. Acknowledgments Finally, the use of homogeneous, standard conductivity values also represents a limitation, since the values can be expected to be inhomo- The authors thank Christine Pickett, Olga Iljina and Dr. Joanne Eysell geneous in the living brain and will vary from standard values acquired for their comments on and proofreading of the manuscript. Further- ex-vivo. Including anisotropic conductivities in the model would be a more, we would like to thank the reviewers for their comments which first step to address this issue. The increase in computational load significantly improved the presented work. This work was supported induced by anisotropic conductivities might be a limiting factor for by the German Federal Ministry of Education and Research grants 7 T-based head modeling. Because only a minority of the blood vessels 16SV5834 NASS and 01GQ1510 OptiStim and DFG grant EXC 1086 included in our model was within the white matter compartment, we BrainLinks-BrainTools to the University of Freiburg. Furthermore, this expect no major insights for the questions addressed in the present study was partly supported by the priority program SPP1665 of the study from modeling white matter anisotropy. Recent advances in elec- German Research Foundation (project WO1425/5-1) and by the Nation- trical impedance tomography (EIT) and more specifically in magnetic al Institute of General Medical Sciences of the National Institutes of resonance EIT (Zhang et al., 2008; Woo and Seo, 2008; Meng et al., Health under grant number P41 GM103545-17. 2013; Degirmenci and Eyuboglu, 2013; Kim et al., 2008) suggest that using individualized anisotropic and inhomogeneous conductivities Appendix A. Supplementary data for head modeling may be possible in the future, opening up exciting new possibilities in volume conductor head modeling. Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.neuroimage.2015.12.041. Conclusions & outlook References For applications directed at regions with little vasculature we would Anwander, A., Wolters, C.H., Dümpelmann, M., Knösche, T., 2002. Influence of realistic suggest that, if the skull is modeled correctly and CSF, dura and anisot- skull and white matter anisotropy on the inverse problem in EEG/MEG-source local- ropy are present in a VCHM, the modeling of blood vessels is a possible ization. Proc. 13th Int. Conf. Biomagn., pp. 679–681. next step towards an even lower model error that may or may not be Ary, J.P., Klein, S.A., Fender, D.H., 1981. Location of sources of evoked scalp potentials: cor- rections for skull and scalp thicknesses. Biomed. Eng. IEEE Trans. Online 447–452. necessary, depending on the accuracy requirements of a study. For ap- Assaf, B.A., Ebersole, J.S., 1997. Continuous source imaging of scalp ictal rhythms in plications where high forward and/or inverse solution accuracies are temporal lobe epilepsy. Epilepsia 38, 1114–1123. necessary and which address areas close to the interhemispheric fissure, Aydin, Ü., et al., 2014. Combining EEG and MEG for the reconstruction of epileptic activity the temporal pole and the insula, the inclusion of blood vessels may be using a calibrated realistic volume conductor model. PLoS One 9, e93154. Aydin, Ü., et al., 2015. Combined EEG/MEG can outperform single modality EEG or MEG highly relevant and as important as considering the CSF, dura, and white source reconstruction in presurgical epilepsy diagnosis. PLoS One. matter anisotropy. Ball, T., et al., 2012. Variability of fMRI-response patterns at different spatial observation Finding the most suitable modeling techniques for blood vessels re- scales. Hum. Brain Mapp. 33, 1155–1171. Bangera, N.B., et al., 2010. Experimental validation of the influence of white matter anisot- quires further attention. Depending on which conductivities are as- ropy on the intracranial EEG forward solution. J. Comput. Neurosci. 29, 371–387. sumed for vessels, we found different mechanism of error generation. Bayford, R., Tizzard, A., 2012. Bioimpedance imaging: an overview of potential clinical ap- As researchers (Butt et al., 1990; Crone and Olesen, 1982) have mea- plications. Analyst 137, 4635–4643. Bénar, C.G., Gotman, J., 2002. Modeling of post-surgical brain and skull defects in the EEG sured the electrical resistance of the brain endothelium (although in inverse problem with the boundary element method. Clin. Neurophysiol. 113, 48–56. rat and frog), two paths towards a solution are currently investigated: Benninghoff, A., 1993. Anatomie. Makroskopische Anatomie, Embryologie und Histologie (i) implementing separate compartments for blood vessel endothelium des Menschen. Bd. 2. Reproduktionsorgane, endokrine Drüsen, Nervensystem, Sinnesorgane, Haut. In: D., Drenckhahn, W., Zenker (Eds.), Urban & Schwarzenberg. and lumen, requiring a volume mesh with local resolution approaching Benninghoff, A., 1994. Anatomie. Makroskopische Anatomie, Embryologie und Histologie des the single μm, making the development of new FEM technologies neces- Menschen. Bd. 1. Zellen- und Gewebelehre, Entwicklungsbiologie, Bewegungsapparat, sary; (ii) modeling the resistive properties of the endothelium as electri- Herz-Kreislauf-System, Immunsystem, Atem- und Verdauungsapparat. In: D., Drenckhahn, W., Zenker (Eds.), Urban & Schwarzenberg. cal boundary conditions. Developing algorithms required for both Best, D.J., Roberts, D.E., 1975. Algorithm AS 89: the upper tail probabilities of Spearman's approaches and making resistances compatible with software is the rho. Appl. Stat. 24, 377. subject of our ongoing research. Once established, these methods will Boyd, G.I., 1930. The emissary foramina of the cranium in man and the anthropoids. permit to investigate the effect of the blood-CSF (choroid plexus) and J. Anat. 65, 108–121. Brazier, M.A.B., 1949. The electrical fields at the surface of the head during sleep. arachnoid barriers, which are also a combination of highly isolating Electroencephalogr. Clin. Neurophysiol. 1, 195–204. tight junctions and conductive fluids. Further investigations will also Bruno, P., Vatta, F., Mininel, S., Inchingolo, P., 2003. Head model extension for the study of be needed to clarify the role of the apparent dependence of blood con- bioelectric phenomena. Biomed. Sci. Instrum. 39, 59–64. Buchner, H., et al., 1997. Inverse localization of electric dipole current sources in finite ductivity and anisotropy relative to flow velocity and vessel diameter element models of the human head. Electroencephalogr. Clin. Neurophysiol. 102, observed in impedance — plethysmography and cardiology (Wtorek 267–278. and Polinski, 2005; Sakamoto and Kanai, 1979). In this context, direct Butt, A.M., Jones, H.C., Abbott, N.J., 1990. Electrical resistance across the blood-brain barrier in anaesthetized rats: a developmental study. J. Physiol. 429, 47–62. measurements in animal models would also be highly useful to resolve Carter, T.J., et al., 2005. Application of soft tissue modelling to image-guided surgery. Med. the current lack of data on the exact conductivity of cerebral vessels, Eng. Phys. 27, 893–909.

79 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208 207

Chauveau, N., et al., 2004. Effects of skull thickness, anisotropy, and inhomogeneity on for- Lanfer, B., et al., 2012a. Influences of skull segmentation inaccuracies on EEG source anal- ward EEG/ERP computations using a spherical three-dimensional resistor mesh ysis. Neuroimage 62, 418–431. model. Hum. Brain Mapp. 21, 86–97. Lau, S., Flemming, L., Haueisen, J., 2014. Magnetoencephalography signals are influenced Cooper, R., Winter, A., Crow, H., Walter, W.G., 1965. Comparison of subcortical, cortical by skull defects. Clin. Neurophysiol. 125, 1653–1662. and scalp activity using chronically indwelling electrodes in man. Electroencephalogr. Lew, S., Wolters, C.H., Dierkes, T., Röer, C., MacLeod, R.S., 2009. Accuracy and run-time Clin. Neurophysiol. 18, 217–228. comparison for different potential approaches and iterative solvers in finite element Crone, C., Olesen, S.P., 1982. Electrical resistance of brain microvascular endothelium. method based EEG source analysis. Appl. Numer. Math. Trans. IMACS 59, 1970–1988. Brain Res. 241, 49–55. Li, J., Wang, K., Zhu, S., He, B., 2007. Effects of holes on EEG forward solutions using a re- Cuffin, B.N., 1993. Effects of local variations in skull and scalp thickness on EEGs and alistic geometry head model. J. Neural Eng. 4, 197–204. MEGs. IEEE Trans. Biomed. Eng. 40, 42–48. Lüsebrink, F., Wollrab, A., Speck, O., 2013. Cortical thickness determination of the human Daneman, R., 2012. The blood–brain barrier in health and disease. Ann. Neurol. 72, 648–672. brain using high resolution 3 T and 7 T MRI data. Neuroimage 70,122–131. Dannhauer, M., Lanfer, B., Wolters, C.H., Knösche, T.R., 2011. Modeling of the human skull Maderwald, S., et al., 2008. To TOF or not to TOF: strategies for non-contrast-enhanced in- in EEG source analysis. Hum. Brain Mapp. 32, 1383–1399. tracranial MRA at 7 T. Magn. Reson. Mater. Phys. Biol. Med. 21, 159–167. Datta, A., Dmochowski, J.P., Guleyupoglu, B., Bikson, M., Fregni, F., 2013. Cranial electro- Manniesing, R., Viergever, M.A., Niessen, W.J., 2006. Vessel enhancing diffusion: a scale therapy stimulation and transcranial pulsed current stimulation: a computer based space representation of vessel structures. Med. Image Anal. 10, 815–825. high-resolution modeling study. NeuroImage 65, 280–287. Manola, L., Roelofsen, B.H., Holsheimer, J., Marani, E., Geelen, J., 2005. Modelling motor Degirmenci, E., Eyuboglu, B.M., 2013. Practical realization of magnetic resonance conduc- cortex stimulation for chronic pain control: electrical potential field, activating func- tivity tensor imaging (MRCTI). IEEE Trans. Med. Imaging 32, 601–608. tions and responses of simple nerve fibre models. Med. Biol. Eng. Comput. 43, Derix, J., et al., 2014. Visualization of the amygdalo-hippocampal border and its structural 335–343. variability by 7 T and 3 T magnetic resonance imaging. Hum. Brain Mapp. 35, Meijs, J.W.H., Peters, M.J., 1987. The EEG and MEG, Using a Model of Eccentric Spheres to 4316–4329. Describe the Head. IEEE Trans. Biomed. Eng. BME-34, 913–920. Fernández-Corazza, M., Beltrachini, L., von Ellenrieder, N., Muravchik, C.H., 2013. Analysis Meijs, J.W.H., Weier, O.W., Peters, M.J., van Oosterom, A., 1989. On the numerical accuracy of parametric estimation of head tissue conductivities using electrical impedance to- of the boundary element method (EEG application). IEEE Trans. Biomed. Eng. 36, mography. Biomed. Signal Process. Control 8, 830–837. 1038–1049. Fernández-Torre, J.L., Alarcón, G., Binnie, C.D., Polkey, C.E., 1999b. Comparison of sphenoi- Meng, Z.J., et al., 2013. Numerical simulations of MREIT conductivity imaging for brain dal, foramen ovale and anterior temporal placements for detecting interictal epilepti- tumor detection. Comput. Math. Methods Med. 2013. form discharges in presurgical assessment for temporal lobe epilepsy. Clin. Merlet, I., Garcia-Larrea, L., Grégoire, M.C., Lavenne, F., Mauguière, F., 1996. Source Neurophysiol. 110, 895–904. propagation of interictal spikes in temporal lobe epilepsy Correlations between Fernández-Torre, J.L., et al., 1999a. Generation of scalp discharges in temporal lobe epilep- spike dipole modelling and [18F]fluorodeoxyglucose PET data. Brain 119, 377–392. sy as suggested by intraoperative electrocorticographic recordings. J. Neurol. Merlet, I., et al., 1998. Topographical reliability of mesio-temporal sources of interictal Neurosurg. Psychiatry 67, 51–58. spikes in temporal lobe epilepsy. Electroencephalogr. Clin. Neurophysiol. 107, Frangi, A.F., Niessen, W.J., Vincken, K.L., Viergever, M.A., 1998. In: Wells, W.M., Colchester, 206–212. A., Delp, S. (Eds.), Med. Image Comput. Comput.-Assist. Interv. — MICCAI'98 1496. Michel, C.M., Murray, M.M., 2012. Towards the utilization of EEG as a brain imaging tool. Springer, Berlin Heidelberg, pp. 130–137. Neuroimage 61, 371–385. Frank, E., 1952. Electric potential produced by two point current sources in a homoge- Michel, C.M., et al., 2004. EEG source imaging. Clin. Neurophysiol. 115, 2195–2222. neous conducting sphere. J. Appl. Phys. 23, 1225–1228. Miller, K., et al., 2010. Modelling brain deformations for computer-integrated neurosur- Geisler, C.D., Gerstein, G.L., 1961. The surface EEG in relation to its sources. gery. Int. J. Numer. Methods Biomed. Eng. 26, 117–138. Electroencephalogr. Clin. Neurophysiol. 13, 927–934. Mosher, J.C., Lewis, P.S., Leahy, R.M., 1992. Multiple dipole modeling and localization from Giacometti, P., Perdue, K.L., Diamond, S.G., 2014. Algorithm to find high density EEG scalp spatio-temporal MEG data. IEEE Trans. Biomed. Eng. 39, 541–557. coordinates and analysis of their correspondence to structural and functional regions Netter, F., 1987. Farbatlanten der Medizin. Neuroanatomie und Physiologie Bd. 5: of the brain. J. Neurosci. Methods 229, 84–96. http://dx.doi.org/10.1016/j.jneumeth. Nervensystem 1 Neuroanatomie und Physiologie. Georg Thieme. 2014.04.020. Nowinski, W.L., et al., 2011. Three-dimensional reference and stereotactic atlas of human Grech, R., et al., 2008. Review on solving the inverse problem in EEG source analysis. cerebrovasculature from 7 Tesla. Neuroimage 55, 986–998. J. Neuroeng. Rehabil. 5, 25. Nuno, L., Balbastre, J.V., Castane, H., 1997. Analysis of general lossy inhomogeneous and Güllmar, D., Haueisen, J., Reichenbach, J.R., 2010. Influence of anisotropic electrical con- anisotropic waveguides by the finite-element method (FEM) using edge elements. ductivity in white matter tissue on the EEG/MEG forward and inverse solution. A Microw. Theory Tech. IEEE Trans. Online 45, 446–449. high-resolution whole head simulation study. NeuroImage 51, 145–163. Oostenveld, R., Oostendorp, T.F., 2002. Validating the boundary element method for for- Hallez, H., Staelens, S., Lemahieu, I., 2009. Dipole estimation errors due to not incorporat- ward and inverse EEG computations in the presence of a hole in the skull. Hum. ing anisotropic conductivities in realistic head models for EEG source analysis. Phys. Brain Mapp. 17, 179–192. Med. Biol. 54, 6079. Oostenveld, R., Praamstra, P., 2001. The five percent electrode system for high-resolution Hallez, H., Vanrumste, B., Hese, P.V., Delputte, S., Lemahieu, I., 2008. Dipole estimation er- EEG and ERP measurements. Clin. Neurophysiol. 112, 713–719. rors due to differences in modeling anisotropic conductivities in realistic head models Oostenveld, R., Fries, P., Maris, E., Schoffelen, J.-M., 2011. FieldTrip: open source software for EEG source analysis. Phys. Med. Biol. 53, 1877. for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput. Hallez, H., et al., 2005. A finite difference method with reciprocity used to incorporate an- Intell. Neurosci. 2011, 1–9. isotropy in electroencephalogram dipole source localization. Phys. Med. Biol. 50, Opitz, A., Windhoff, M., Heidemann, R.M., Turner, R., Thielscher, A., 2011. How the brain 3787–3806. tissue shapes the electric field induced by transcranial magnetic stimulation. Hartmann, U., Berti, G., Kruggel, F., 2010. SimBio-Vgrid. at http://www.rheinahrcampus. Neuroimage 58, 849–859. de/~medsim/vgrid/. Panzer, M.B., Myers, B.S., Capehart, B.P., Bass, C.R., 2012. Development of a finite element Haueisen, J., Ramon, C., Czapski, P., Eiselt, M., 1995. On the influence of volume currents model for blast brain injury and the effects of CSF cavitation. Ann. Biomed. Eng. 40, and extended sources on neuromagnetic fields: a simulation study. Ann. Biomed. 1530–1544. Eng. 23, 728–739. Pascual-Marqui, R.D., 1999. Review of methods for solving the EEG inverse problem. Int. Haueisen, J., Ramon, C., Eiselt, M., Brauer, H., Nowak, H., 1997. Influence of tissue resistiv- J. Bioelectromagnetics 1,75–86. ities on neuromagnetic fields and electric potentials studied with a finite element Ramon, C., 2012. Effect of dura layer on scalp EEG simulations. Int. J. Bioelectromagnetics model of the head. IEEE Trans. Biomed. Eng. 44, 727–735. 14, 27–28. Haueisen, J., et al., 2002. The influence of brain tissue anisotropy on human EEG and MEG. Ramon, C., Garguilo, P., Fridgeirsson, E.A., Haueisen, J., 2014. Changes in scalp potentials NeuroImage 15, 159–166. and spatial smoothing effects of inclusion of dura layer in human head models for Heasman, B.C., et al., 2002. A hole in the skull distorts substantially the distribution of EEG simulations. Front. Neuroeng. 7, 32. extracranial electrical fields in an in vitro model. J. Clin. Neurophysiol. 19, 163. Ramon, C., Schimpf, P.H., Haueisen, J., 2006. Influence of head models on EEG simulations Heers, M., et al., 2012. MEG-based identification of the epileptogenic zone in occult and inverse source localizations. Biomed. Eng. Online 5, 10. peri-insular epilepsy. Seizure 21,128–133. Ramon, C., Schimpf, P., Haueisen, J., Holmes, M., Ishimaru, A., 2004. Role of soft bone, CSF Hosek, R.S., Sances, A., Jodat, R.W., Larson, S.J., 1978. The contributions of intracerebral cur- and gray matter in EEG simulations. Brain Topogr. 16, 245–248. rents to the EEG and evoked potentials. IEEE Trans. Biomed. Eng. BME-25, 405–413. Ramon, C., et al., 2009. Similarities between simulated spatial spectra of scalp EEG, MEG Jung, K.-Y., et al., 2009. Spatiotemporospectral characteristics of scalp ictal EEG in mesial and structural MRI. Brain Topogr. 22, 191–196. temporal lobe epilepsy with hippocampal sclerosis. Brain Res. 1287, 206–219. Rampp, S., Stefan, H., 2007. Magnetoencephalography in presurgical epilepsy diagnosis. Kaipio, J., Somersalo, E., 2007. Statistical inverse problems: discretization, model Expert Rev. Med. Devices 4, 335–347. reduction and inverse crimes. J. Comput. Appl. Math. 198, 493–504. Rice, J.K., Rorden, C., Little, J.S., Parra, L.C., 2013. Subject position affects EEG magnitudes. Kim, H.J., et al., 2008. In vivo electrical conductivity imaging of a canine brain using a 3 T Neuroimage 64, 476–484. MREIT system. Physiol. Meas. 29, 1145. Rosenfeld, A., Kak, A.C., 1982. Digital Picture Processing. Academic Press. Krechel, A., Stüben, K., 2001. Parallel algebraic multigrid based on subdomain blocking. Rullmann, M., et al., 2009. EEG source analysis of epileptiform activity using a 1 mm an- Parallel Comput. 27, 1009–1031. isotropic hexahedra finite element head model. Neuroimage 44, 399–410. Kroon, D.-J., 2009. Hessian based frangi vesselness filter - file exchange - MATLAB Central. Sadleir, R.J., Vannorsdall, T.D., Schretlen, D.J., Gordon, B., 2010. Transcranial direct current at http://www.mathworks.com/matlabcentral/fileexchange/24409-hessian-based- stimulation (tDCS) in a realistic head model. Neuroimage 51, 1310–1318. frangi-vesselness-filter. Sakamoto, K., Kanai, H., 1979. Electrical characteristics of flowing blood. IEEE Trans. Lanfer, B., Paul-Jordanov, I., Scherg, M., Wolters, C.H., 2012b. Influence of interior cerebro- Biomed. Eng. BME-26, 686–695. spinal fluid compartments on EEG source analysis. Biomed. Tech. (Berl) http://dx.doi. Schneider, M.R., 1972. A multistage process for computing virtual dipolar sources of EEG org/10.1515/bmt-2012-4020. discharges from surface information. IEEE Trans. Biomed. Eng. BME-19, 1–12.

80 208 L.D.J. Fiederer et al. / NeuroImage 128 (2016) 193–208

SimBio Development Group, 2012. SimBio — a generic environment for bio-numerical Vorwerk, J., et al., 2014. A guideline for head volume conductor modeling in EEG and simulations. at https://www.mrt.uni-jena.de/simbio/index.php/Main_Page. MEG. Neuroimage 100, 590–607. Slutzky, M.W., et al., 2010. Optimal spacing of surface electrode arrays for brain machine Waberski, T.D., et al., 2000. Source reconstruction of mesial-temporal epileptiform activ- interface applications. J. Neural Eng. 7, 26004. ity: comparison of inverse techniques. Epilepsia 41, 1574–1583. Sparkes, M., Valentin, A., Alarcón, G., 2009. Mechanisms involved in the conduction of Wagner, S., et al., 2014. Investigation of tDCS volume conduction effects in a highly anterior temporal epileptiform discharges to the scalp. Clin. Neurophysiol. 120, realistic head model. J. Neural Eng. 11, 016002. 2063–2070. Wendel, K., Narra, N.G., Hannula, M., Kauppinen, P., Malmivuo, J., 2008. The influence of Srinivasan, R., Tucker, D., Murias, M., 1998. Estimating the spatial Nyquist of the human CSF on EEG sensitivity distributions of multilayered head models. IEEE Trans. Biomed. EEG. Behav. Res. Methods 30, 8–19. Eng. 55, 1454–1456. Stinstra, J.G., Hopenfeld, B., MacLeod, R.S., 2005b. On the passive cardiac conductivity. Wendel, K., et al., 2009. EEG/MEG source imaging: methods, challenges, and open issues. Ann. Biomed. Eng. 33, 1743–1751. Comput. Intell. Neurosci. 2009. Stinstra, J.G., Shome, S., Hopenfeld, B., MacLeod, R.S., 2005a. Modelling passive cardiac Wilson, F.N., Bayley, R.H., 1950. The electric field of an eccentric dipole in a homogeneous conductivity during ischaemia. Med. Biol. Eng. Comput. 43, 776–782. spherical conducting medium. Circulation 1, 84–92. Stüben, K., 2001. A review of algebraic multigrid. J. Comput. Appl. Math. 128, 281–309. Wolters, C.H., 2003. Influence of tissue conductivity inhomogeneity and anisotropy to Tatu, L., Moulin, T., Vuillier, F., Bogousslavsky, J., 2012. Arterial territories of the human EEG/MEG based source localization in the human brain. brain. Front. Neurol. Neurosci. 30, 99–110. Wolters, C.H., Anwander, A., Berti, G., Hartmann, U., 2007. Geometry-Adapted Hexahedral Thevenet, M., Bertrand, O., Perrin, F., Pernier, J., 1992. Finite element method for a realistic Meshes Improve Accuracy of Finite-Element-Method-Based EEG Source Analysis. head model of electrical brain activities. 1992 14th Annu. Int. Conf. IEEE Eng. Med. IEEE Trans. Biomed. Eng. 54, 1446–1453. Biol. Soc. 5, pp. 2024–2025 Wolters, C.H., Grasedyck, L., Hackbusch, W., 2004. Efficient computation of lead field bases Van de Moortele, P.-F., et al., 2009. T1 weighted brain images at 7 Tesla unbiased for pro- and influence matrix for the FEM-based EEG and MEG inverse problem. Inverse Prob. ton density, T2⁎ contrast and RF coil receive B1 sensitivity with simultaneous vessel 20, 1099–1116. visualization. Neuroimage 46, 432–446. Wolters, C.H., Kuhn, M., Anwander, A., Reitzinger, S., 2002. A parallel algebraic multigrid van den Broek, S.P., Reinders, F., Donderwinkel, M., Peters, M.J., 1998. Volume conduction solver for finite element method based source localization in the human brain. effects in EEG and MEG. Electroencephalogr. Clin. Neurophysiol. 106, 522–534. Comput. Vis. Sci. 5, 165–177. van der Zwan, A., Hillen, B., 1991. Review of the variability of the territories of the major Wolters, C.H., et al., 2006. Influence of tissue conductivity anisotropy on EEG/MEG field cerebral arteries. Stroke 22, 1078–1084. and return current computation in a realistic head model: a simulation and Vanrumste, B., et al., 2000. Dipole location errors in electroencephalogram source analysis visualization study using high-resolution finite element modeling. Neuroimage 30, due to volume conductor model errors. Med. Biol. Eng. Comput. 38, 528–534. 813–826. Vatta, F., Bruno, P., Mininel, S., Inchingolo, P., 2005. EEG simulation accuracy: reference Woo, E.J., Seo, J.K., 2008. Magnetic resonance electrical impedance tomography (MREIT) choice and head models extension. Int. J. Bioelectromagnetics 7, 154–157. for high-resolution conductivity imaging. Physiol. Meas. 29, R1. Vonach, M., et al., 2012. A method for rapid production of subject specific finite Wtorek, J., Polinski, A., 2005. The contribution of blood-flow-induced conductivity element meshes for electrical impedance tomography of the human head. Physiol. changes to measured impedance. IEEE Trans. Biomed. Eng. 52, 41–49. Meas. 33, 801. Yang, J.(.J.)., Dai, J., Zhuang, Z., 2009. In: Duffy, V.G. (Ed.), Digit. Hum. Model.Springer, Ber- Voo, L., Kumaresan, S., Pintar, F.A., Yoganandan, N., Sances Jr, A., 1996. Finite-element lin Heidelberg, pp. 661–670. models of the human head. Med. Biol. Eng. Comput. 34, 375–381. Zhang, X., Yan, D., Zhu, S., He, B., 2008. Noninvasive imaging of head-brain conductivity Vorwerk, J., Clerc, M., Burger, M., Wolters, C.H., 2012. Comparison of boundary element profiles. IEEE Eng. Med. Biol. Mag. 27, 78–83. and finite element approaches to the EEG forward problem. Proc 46 DGBMT Annu. Conf. 57, pp. 795–798.

81 82 Published in Journal of Neural Engineering October 2017

VIII. MAPPING THE FINE STRUCTURE OF CORTICAL ACTIVITY WITH DIFFERENT MICRO-ECOG ELECTRODE ARRAY GEOMETRIES

83 IOP

Journal of Neural Engineering

Journal of Neural Engineering

J. Neural Eng. J. Neural Eng. 14 (2017) 056004 (23pp) https://doi.org/10.1088/1741-2552/aa785e

14 Mapping the fne structure of cortical

2017 activity with different micro-ECoG electrode

© 2017 IOP Publishing Ltd array geometries

JNEIEZ Xi Wang1,2,3,4, C Alexis Gkogkidis1,2,3,4, Olga Iljina1,2,4,5,6, 1,2,4,5 7 8 9,10 056004 Lukas D J Fiederer , Christian Henle , Irina Mader , Jan Kaminsky , Thomas Stieglitz3,4, Mortimer Gierthmuehlen2,4 and Tonio Ball1,2,4,11

1 X Wang et al Department of Neurosurgery, Epilepsy Center, Translational Neurotechnology Lab, Medical Center— University of Freiburg, Faculty of Medicine, University of Freiburg, AG Ball, Engelbergerstr. 21 3.0 EG, 79106 Freiburg, Germany 2 Department of Neurosurgery, Medical Center—University of Freiburg, Faculty of Medicine, University of Freiburg, Breisacher Str. 64, 79106 Freiburg, Germany 3 Laboratory for Biomedical Microtechnology, Department of Microsystems Engineering (IMTEK), Printed in the UK University of Freiburg, Georges-Koehler-Allee 102, 79110 Freiburg, Germany 4 BrainLinks-BrainTools Cluster of Excellence, University of Freiburg, Georges-Koehler-Allee 80, 79110 Freiburg, Germany JNE 5 Department of Neurobiology and Biophysics, Faculty of Biology, University of Freiburg, Schaenzlestr. 1, 79104 Freiburg, Germany 6 10.1088/1741-2552/aa785e GRK 1624 ‘Frequency effects in language’, University of Freiburg, Belfortstraße 18, 79098 Freiburg, Germany 7 CorTec GmbH, Georges-Koehler-Allee 010, 79110 Freiburg, Germany 8 Department of Neuroradiology, Medical Center—University of Freiburg, Faculty of Medicine, Paper University of Freiburg, Breisacher Str. 64, 79106 Freiburg, Germany 9 Department of Neurosurgery, St. Gertrauden Krankenhaus, Paretzer Straße 12, 10713 Berlin, Germany 1741-2552 E-mail: [email protected]

Received 21 February 2017, revised 16 May 2017 Accepted for publication 9 June 2017 5 Published 16 August 2017

Abstract Objective. Innovations in micro-electrocorticography (µECoG) electrode array manufacturing now allow for intricate designs with smaller contact diameters and/or pitch (i.e. inter-contact distance) down to the sub-mm range. The aims of the present study were: (i) to investigate whether frequency ranges up to 400 Hz can be reproducibly observed in µECoG recordings and (ii) to examine how differences in topographical substructure between these frequency bands and electrode array geometries can be quantifed. We also investigated, for the frst time, the infuence of blood vessels on signal properties and assessed the infuence of cortical vasculature on topographic mapping. Approach. The present study employed two µECoG electrode arrays with different contact diameters and inter-contact distances, which were used to characterize neural activity from the somatosensory cortex of minipigs in a broad frequency range up to 400 Hz. The analysed neural data were recorded in acute experiments under anaesthesia during peripheral electrical stimulation. Main results. We observed that µECoG recordings reliably revealed multi-focal cortical somatosensory response patterns, in

10 This author’s current address is different from the address where the work was carried out. 11 Author to whom any correspondence should be addressed. Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

1741-2552/17/056004+23$33.00 1 © 2017 IOP Publishing Ltd Printed in the UK

84 J. Neural Eng. 14 (2017) 056004 X Wang et al which response peaks were often less than 1 cm apart and would thus not have been resolvable with conventional ECoG. The response patterns differed by stimulation site and intensity, they were distinct for different frequency bands, and the results of functional mapping proved independent of cortical vascular. Our analysis of different frequency bands exhibited differences in the number of activation peaks in topographical substructures. Notably, signal strength and signal-to-noise ratios differed between the two electrode arrays, possibly due to their different sensitivity for variations in spatial patterns and signal strengths. Signifcance. Our fndings that the geometry of µECoG electrode arrays can strongly infuence their recording performance can help to make informed decisions that maybe important in number of clinical contexts, including high-resolution brain mapping, advanced epilepsy diagnostics or brain–machine interfacing.

Keywords: µECoG array size, topographic mapping, somatosensory cortex, cortical vascular, minipig (Some fgures may appear in colour only in the online journal)

List of abbreviations they play an important role in pre-neurosurgical evaluation of pharmaco-resistant epilepsy (Ojemann et al 1989, Engel Abbreviation Complete term 1996) and in general neuroscience research. ECoG has proven useful to study such dynamic neuronal processes as move- AEP auditory-evoked potential ment execution (Toro et al 1994, Crone et al 1998, Ball et al BBB blood–brain barrier 2009), speech perception and production (Ojemann et al BMI brain–machine interfacing 1989, Crone et al 2001a, 2001b, Derix et al 2014) or natural- BW body weight istic social interaction (Derix et al 2012). The conventional CAD computer-aided design ‘macro’-ECoG electrode arrays used in previous studies and ECG electrocardiogram in current epilepsy diagnostics (Engel et al 1996, 2007) have FDR false discovery rate contacts with a diameter of several mm and an inter-contact FR fast ripple distance on the order of 1 cm. In recent decades, however, HFO high frequency oscillation there has been growing interest in increasing the spatial reso- IMTEK Institute of Microsystems Engineering, lution. To this aim, novel electrode arrays with smaller con- University of Freiburg, Freiburg, Germany tacts and higher electrode densities have been developed and iqr interquartile range tested with different experimental paradigms and animal spe- LFP local feld potential cies (Hollenberg et al 2006, Kim et al 2007, Kitzmiller et al MEMS micro-electromechanical systems 2007, Blakely et al 2008, Hosp et al 2008, Slutzky et al 2008, MRI magnetic resonance imaging 2010, 2011, Kellis et al 2009, 2016, Leuthardt et al 2009, MUA multi-unit activity Wang et al 2009, Khodagholy et al 2011, 2015, Viventi et al PSP post-synaptic potential 2011, Fukushima et al 2012, Wang et al 2016, Flint et al 2017, relSP relative spectral power Trumpis et al 2017). These electrode arrays span a range of relSPSnr signal-to-noise ratio of the relative spectral spatial scales of approximately two orders of magnitude, both power with respect to electrode contact size and spacing. Electrode SEP somatosensory evoked potential sizes range from 1 to 2 mm (Rubehn et al 2009, Vinjamuri SNR signal-to-noise ratio et al 2009, Wang et al 2009, 2016, Wilks et al 2009) down to SSLM signifcant spatial local maxima a few µm, e.g. 4 µm in Kellis et al (2016), 10 and 20 µm in SUA single-unit activity Khodagholy et al (2011, 2015), 40 µm in Blanco et al (2010), TE time of echo or 75 µm in Leuthardt et al (2009). Electrode spacing ranges TR time of repetition from 3 to 4 mm (Blakely et al 2008, Vinjamuri et al 2009, TTL transistor-transistor logic Wang et al 2009, 2016) down to 30–150 µm (Kitzmiller et al VEP visual-evoked potential 2007, Khodagholy et al 2011, 2015; see table A1 for details). 3D three-dimensional While the terms used in the literature differ, all recording (µ)ECoG (micro-)electrocorticography techniques with geometries smaller than conventional clinical electrode arrays may collectively be defned as ‘micro-’ECoG 1. Introduction (µECoG). In the present study we follow this defnition (alter- natively, the term µECoG could be reserved from electrodes Electrocorticographic (ECoG) recordings can provide detailed that are substantially smaller geometries, down to the µm spatio-temporal information on cortical population activity scale, and the intermediate range could be distinguished as, (Di and Barth 1993, Crone 2000, Ojemann et al 2013), and e.g., meso-ECoG).

2

85 J. Neural Eng. 14 (2017) 056004 X Wang et al

There are several potential advantages of µECoG. As the 1999, Barth 2003, Leuthardt et al 2009), and little is known about overall size of electrode arrays is a main risk factor for implant- its spatio-temporal properties. Neuronal population activity in ation-related complications (Hamer et al 2002, Wong et al high-gamma frequencies between 250 and 500 Hz, also termed 2009), complications can be expected to occur less frequently ‘fast ripples’ (FRs; Bragin et al 1999a, 1999b, Curio 2000, Staba during µECoG than macro-ECoG implantations (Morrell and et al 2002, Crone et al 2006), have been observed very locally. Epilepsy Study Group 2011). µECoG electrode arrays have The spatial extent of brain areas generating FRs did not exceed also been used for spatially detailed functional mapping in 1 mm3 in the rat hippocampus (Bragin et al 2002), and µECoG animal models and in human subjects. For instance, the cor- arrays may be particularly useful to detect them. The capacity of tical underpinnings of auditory-evoked potentials (AEPs) and µECoG to capture such high-frequency components of the neu- somatosensory-evoked potentials (SEPs) have been elucidated ronal signal, however, is currently controversial: While Worrell in rats (Barth and Di 1991), sheep (Gierthmuehlen et al 2014) et al (2008) showed that smaller electrode contacts matching and in human patients (Rembado et al 2017). Event-related the size of neuronal activity generators can detect FRs in the potentials in µECoG have been employed to study sleep spin- human amygdalo-hippocampal region, Chatillon et al (2011) dles, visual processing, and electrographic seizures in cats observed no infuence of electrode size on FR detection in the (Viventi et al 2011). Event-related potential studies employing rat hippocampus. Furthermore, Rouse et al (2016) examined µECoG have also been conducted in humans to detect cortical closed-loop BMI control with different inter-contact distances. sites for phoneme- (Blakely et al 2008) and word-specifc Using signals in high frequencies (75–105 Hz), and they found (Kellis et al 2010) processing, as well as to localize cortical that the performance was improved with smaller electrode arrays motor functions via movement-related potentials or changes in that obtained. Depending on the decoding strategy, however, the the spectral magnitude (Kellis et al 2009, Leuthardt et al 2009, beneft of spatial resolution (inter-contact distance in the afore- Vinjamuri et al 2009, Wang et al 2009, Rouse et al 2013, 2016, mentioned study) may differ between the signal properties of Hotson et al 2015, Wang et al 2016, Flint et al 2017). The neuronal recording and the implemented strategy of BMI control spatially and temporally localized neuronal effects observed (Rouse et al 2016). Thus, whether smaller electrode contacts and/ in these studies suggest that µECoG electrode arrays have a or smaller inter-contact distances are more effcient in detecting high capability to differentiate signals within small cortical high-frequency cortical activity is yet not entirely clear, and one regions. In addition to basic neuroscience research, such char- aim of the present study was to address this question. acteristics can be useful for a number of clinical applications, A second and related question we address in this study is such as high-detail functional mapping, presurgical epilepsy whether increasing the spatial detail of recordings will lead to diag nostics and brain–machine interfacing (BMI), which all a better spatial resolution of the functional organization of the require source/control signals from small cortical regions. cerebral cortex. This question is relevant with respect to the Compared to other modalities used for functional mapping, main motivations for using µECoG electrode arrays, namely, such as those based on hemodynamic methods or conventional (i) to elucidate the characteristics of neuronal population ECoG, µECoG research is not yet as widely used. Only several activity and the functional organization of the cerebral cortex dozens of µECoG studies have been conducted, (see Appendix) with high spatial detail and (ii) to increase the amount of mean- and many open questions exist regarding the properties of this ingful information available for clinical . For technique and its capacity to resolve topographic details. For example, using a high-resolution 360-channel electrode array example, invasive diagnostics of epilepsy requires high spatial with a 500 µm electrode size, Viventi et al (2011) were able precision to achieve seizure alleviation without compromising to visualize previously unknown recurrent spiral waves during functional areas. Pre-neurosurgical monitoring with ECoG seizure propagation in the feline neocortex. Most previous aims to delineate eloquent brain areas by neurostimulation on µECoG studies, however, have focused on cortical response the one hand and to determine the seizure onset zone based features at individual electrode contacts. Detailed topograph- on recordings of brain activity on the other hand. These data ical maps of very high-frequency neuronal effects have rarely are used to decide (i) whether surgery is possible given the been reported (Blakely et al 2008, Hosp et al 2008, Wang et al estimated risks of functional impairment, and (ii) whether sur- 2009, Wang et al 2016) and require further investigation. gery in the area identifed based on (i) will suffce to achieve The third open question addressed in this study concerns seizure freedom. The amount of spatial detail is important for the infuence of cortical blood vessels on µECoG record- these decisions, and increasing the spatial resolution of clinical ings. Intracranial recordings are obtained from cortical tissue, procedures by using µECoG is expected to enhance the preci- which is perfused by vasculature. Blood vessels are sepa- sion of pre-neurosurgical diagnostics (Kim et al 2007, Stead rated from the brain by the blood–brain barrier (BBB), which et al 2010, Kellis et al 2016), but open questions remain on the is, among other functions, important for the precise control impact of different µECoG geometries on the recorded data. of the ionic environment of neural tissue (Daneman 2012). One open question concerns the role of µECOG electrode Endothelial cells, which are the main structure component in array geometry (i.e. contact diameter and inter-contact distance) the BBB, enable a particularly high electrical resistance of the in detection of high-frequency signal components. Most µECoG vessel walls in the brain (Crone and Olesen 1982, Fiederer studies have investigated frequency bands up to ca. 250 Hz et al 2016). Since electrode contacts in subdural recordings (Blakely et al 2008, Wang et al 2009, Khodagholy et al 2011, often lie on blood vessels, i.e. between the source of neural Fukushima et al 2012). Cortical activity in much higher frequen- activity and the recording site, they may impede or otherwise cies (>250 Hz) has only rarely been reported (Jones and Barth infuence electrophysiological measurements. Only a couple

3

86 J. Neural Eng. 14 (2017) 056004 X Wang et al of studies so far have addressed the impact of blood vessels on neuronal recordings (Miller et al 2009, Bleichner et al 2011). Miller et al (2009) observed the weakest signals from macro-ECoG contacts located above cortical vasculature and excluded such contacts from further analyses. Bleichner et al (2011) compared macro-ECoG recordings directly from the cortical surface with recordings obtained on blood vessels, observing an attenuation of the absolute spectral power at elec- trode contacts located above the vessels. This effect was most pronounced in the 30–70 Hz frequency range. These studies have been conducted with ‘macro-’ECoG, and, to our knowl- edge, no previous study has described the impact of blood ves- sels on µECoG recordings. Since the aforementioned study by Bleichner et al (2011) investigated the infuence on absolute power values only up to the 70–115 Hz frequency range, it is also unclear whether and to what extent activity in higher frequency bands is affected. The present study addressed these questions by employing two custom 4 × 12-contact µECoG electrode arrays with dif- ferent electrode contact sizes and contact-to-contact distances. We recorded neural responses elicited by peripheral electrical stimulation in the somatosensory cortex in a minipig animal model. The resulting SEP data obtained from the same cor- tical patch at a lower and higher spatial resolutions were trans- Figure 1. µECoG electrode array layout and positioning on the formed into spatial maps of frequency bands up to 400 Hz. This cortex. (a) Layouts of the two types of platinum-silicone µECoG mapping procedure show, among other fndings, that ‘zooming electrode arrays used in the present study, with 3.5 mm and 1.68 mm in’ on cortical activation patterns indeed provides additional inter-contact distances, respectively. (b) 3D reconstruction of spatial detail in terms of the number of signifcant activation the position of the larger (green) and the smaller (blue) µECoG electrode arrays on the minipigs’ left somatosensory cortex (i.e. the peaks. Overall, our results indicate that the geometry of the area approximately inside the dotted black line, see section 2). Note µECoG electrode array can have a considerable infuence on that the two electrode arrays were placed on the cortex sequentially; different signal features, and that it therefore needs to be taken they are visualized on top of each other for display purposes only. into consideration for clinical and research applications.. were placed on the exposed surface of the somatosensory cortex parallel to the central sulcus (Craner and Ray 1991). 2. Methods

2.1. Animals and implantation procedure 2.2. µECoG electrode arrays

Four Goettingen minipigs of 20–26 kg body weight (BW) were Two novel µECoG electrode arrays in a hexagonal arrangement intramuscularly premedicated with midazolam (0.5 mg/kgBW) (fgure 1(a)), both with 48 (4 × 12) contacts but of different sizes, and ketamine (20 mg/kgBW) and intravenously anesthetized were sequentially implanted subdurally to record SEPs. The elec- with propofol (2–4 mg/kgBW). Following endotracheal intu- trodes were all designed and manufactured by the Laboratory for 1 bation, 12–15 breaths min− were provided by a volume-con- Biomedical Micro-technology at the Institute of Microsystems trolled ventilator (Servo 900C, Siemens Elema, Solna/Sweden) Engineering (IMTEK), University of Freiburg, Germany (Henle at a 10–15 ml/kgBW tidal volume, 5 mbar positive end-expi- et al 2009). The novelty of the electrode arrays is based in the ratory pressures, and normalized oxygen and carbon dioxide high-purity platinum foil embedded in medical grade silicone tension and pH values. Anaesthesia was maintained through rubber (Schuettler et al 2005), which is well accepted by the body an ear-vein with propofol (15–18 mg/kgBW/h), fentanyl (2–3 in long-term implantations manufactured using an automatic and µg/kgBW/h) and pancuronium (0.2–0.4 mg/kgBW/h). Fluid fexible laser-structured fabrication process. requirements were substituted with ringer solution (10 mg/ The larger electrode array (implanted frst, fgure 1(a), left) kgBW/h). The electrocardiogram (ECG), body-temperature had a size of 37.8 mm × 15.4 mm, a thickness of 0.31 mm, and oxygen saturation were monitored continuously. It was pos- an electrode contact diameter of 1.81 mm, and a centre-to- sible to maintain this anaesthesia for up to 12 h. The study was centre inter-contact distance of 3.5 mm. The smaller elec- approved by the Regierungspräsidium Baden–Württemberg trode array (implanted second, fgure 1(a), right) had a size of and the Animal Committee of the University of Freiburg. The 18 mm × 7.5 mm, a thickness of 0.23 mm, an electrode contact ‘Principles of laboratory animal care’ (NIH publication no. diameter of 0.87 mm, and a inter-contact distance of 1.68 mm. 86- 53 23, revised 1996) were followed for all experiments. Electrode contacts and connecting tracks were made of platinum The left somatosensory cortex was approached as previously (12.5 µm thick layer of platinum) in a silicone substrate. The described (Gierthmuehlen et al 2011), and the electrode arrays metal tracks were designed to be meander-shaped to facilitate

4

87 J. Neural Eng. 14 (2017) 056004 X Wang et al

Figure 2. Somatosensory-evoked potentials (SEPs) recorded with µECoG. (a) and (c) Topographic maps of SEP amplitude changes upon contralateral stimulation (1.0 mA stimulation intensity, larger µECoG electrode array) at two different stimulation sites. The stimulation sites corresponding to panels (a) and (c) are shown in (b). The maps show the trial- and time-averaged potential change in a time window from 14 to 23 ms after stimulation onset. The electrode contacts with signifcant positive and negative potential changes are marked by larger flled black and white circles, respectively (sign test, FDR-corrected at q < 0.01). The white and black dots in the centre of electrodes indicate the contacts with the maximum positive and negative amplitude changes, respectively. Among the remaining electrode contacts, which did not reach signifcance after FDR correction, those with p ⩽ 0.3 (sign test) are marked by larger flled gray circles, and those with p > 0.3 (sign test) are marked by the smaller flled black circles. (d) and (e) Topographic maps at the same stimulation site and within the same time window as in (c) but from a different minipig stimulated at 1.0 mA (d) and 8.0 mA (e). The signals were recorded with the larger µECoG electrode array. (f) SEP amplitudes were proportional to the stimulation intensity. The black dots between the horizontal black error bars indicate the maximum positive amplitude values at different stimulation intensities from the same minipig at the same stimulation site within the same time window as in (d) and (e). The error bars indicate the corresponding standard errors across trials. The SEPs were signifcant (∗: sign test, FDR-corrected at q < 0.01); the black line is the polynomial ft to the data (see section 2).

bending and stretching of the electrode array (Schuettler et al of SEPs (Craner and Ray 1991, fgure 1(b)). The minipigs’ 2009). Wires (bundled in silicone tubing) are welded to the noses were marked with a regular grid of 10 dots (approxi- metal tracks (Schuettler et al 2008). Silicone adhesive was used mately 2.5 mm diameter, 7 mm inter-contact distance) around to electrically seal the welding spots against body liquids. All the right nostrils in order to insert the monopolar stimulation 48 electrode contacts were used to record cortical responses. electrodes at equal distances (Gierthmuehlen et al 2011). Two extra metal wires with needle electrodes were directly con- First, the larger electrode array (electrode contact diam- nected to the amplifer: one was inserted into the contra-lateral eter = 1.81 mm, inter-contact distance = 3.5 mm) was placed (right) hemisphere as the reference, and the other was inserted on the contralateral somatosensory cortex. Current pulses at the posterior edge of the wound as electrical grounding. As were applied to two stimulation sites on opposing sides of the an electrical shielding, an aluminium-cover was placed over the nostril (fgure 2(b)). The pulses had durations of 100 µs, and whole wound and connected to the common ground. inter-stimulus intervals of 495 ms and were delivered in 3 min sessions, yielding approximately 360 trials. To determine the minimum stimulation intensity at which clear SEPs could be 2.3. Recording, stimulation and online analysis elicited, consecutive sessions with 0.5 mA, 1.0 mA, 2.0 mA, The µECoG electrode arrays were connected to four 16-channel 4.0 mA and 8.0 mA were conducted at two stimulation amplifers (gUSB amp, g.tec, Schiedlberg, Austria) via a sites, one after another. The data of each recording session switchbox. We used 12 channels on each amplifer for equal was transferred to a workstation computer and was imme- distribution and one channel for the TTL trigger pulse to syn- diately analysed. This online analysis consisted of common chronize the stimulation and recordings. For data acquisition, average re-referencing, software high-pass fltering (4th order we used the BCI2000 system (Schalk et al 2004). The signals Butterworth, cut-off frequency 1 Hz) to remove offsets that from each channel were digitized at a sampling rate of 4800 impeded online visualization, and trial averaging. The online Hz, amplifed (×3000 gain) and bandpass fltered between 0.1 analysis performed during the experiment revealed that with Hz and 2000 Hz (8th order Butterworth bandpass flter). the larger electrode array, clear SEPs were obtained with Stimulation was performed with single isolated constant stimulation intensities between 1.0 and 8.0 mA, while 0.5 current pulses generated using an ML180 stimulator. This mA was too low to elicit clear measurable SEPs. We therefore stimulator was connected to the PowerLab 8/30 controller chose the 1.0 and 8.0 mA intensities for the following position (both AD-Instruments, Spechbach, Germany) with a BNC test (also conducted with the larger electrode array), where cable and activated by a TTL pulse. As the rostrum of the we stimulated all of the 10 marked stimulation sites around minipig is represented on a relatively large area of the cortex, the nostril. The pulses were again delivered with durations of we chose the appropriate area of the minipigs’ noses for stim- 100 µs, inter-stimulus intervals of 495 ms, and sessions of ulation and the nasal somatosensory area for the recording 3 min, yielding approximately 360 trials for each combination

5

88 J. Neural Eng. 14 (2017) 056004 X Wang et al of the two stimulation intensities and the 10 stimulation sites. and averaged to a single SEP trace for each electrode contact and Online analysis of the SEP data obtained from this position each individual session. (iii) Each trial of these data was further test was performed as described previously. The position of divided into 16 non-overlapping time windows of approximately the contact on the nasal somatosensory cortex that showed 9 ms, and the mean SEP ampl itudes for each of these time win- SEPs with the largest amplitudes was marked with the spot dows were calculated. Then, median SEP amplitudes across trials of a laser pointer. The larger electrode array was then care- were calculated separately for each time window, electrode con- fully removed and the smaller electrode array was placed tact and session. Baseline correction was performed by subtract- such that its centre was aligned to the laser point. The same ing the median SEP values over the frst two time windows from stimulation protocol as for the large electrode array was again the SEP amplitude in each of the following time windows. Then, run for the smaller electrode array (electrode contact diam- a sign test was employed to detect signifcant SEP amplitude eter = 0.87 mm, inter-contact distance = 1.68 mm). All the changes, and the obtained p-values were corrected for multiple tests were performed within 5–6 h following the surgery, and comparisons across 48 contacts and 16 time windows using the the minipigs were sacrifced after the experiments. false discovery rate (FDR) correction approach (Benjamini and Yekutieli 2001) at a q-level of 0.01. These steps were performed separately for each recording session. Signifcant contacts which 2.4. MRI and electrode visualization recorded the maximum positive SEP amplitude changes will be To visualize the electrode contacts on the cortex (fgure 1(b)), a referred to as ‘positive maxima contacts’ hereafter. As shown 3D brain model was created by segmenting a post-mortem iso- in fgure 2(f), we used a polynomial ft to the data to visualize tropic (0.4 mm × 0.4 mm × 0.4 mm) T1-weighted magn etic res- the relation between stimulation intensities and the amplitudes onance image (MRI) of the minipig’s brain obtained on a Bruker of positive maxima contacts in each time window. Among the 3T Machine, Bruker Corporation, MA, Siemens, Germany. The contacts that showed no signifcant effects after FDR correction, image was obtained with the following param eters: maximum some had large p-values > 0.3, clearly showing no effects. The gradient strength 40 mT m−1, TR = 1900 ms, TE = 900 ms, remaining contacts that were neither clearly signifcant nor very 32-channel receiver coil, section thickness = 0.4 mm, feld of large (p < 0.3, but above signifcance level) fell into an interme- view 100 × 100 mm, interpolated to a 256 × 256 matrix. As the diate range of values. The distinction between these three cases brain was fxed in formalin before the MRI was performed, the (signifcant after FDR-corrected at q < 0.01, clearly no effect with tissue contrasts were changed and the segmentation of the cor- p > 0.3, and intermediate cases) was important for the evaluation tical gray matter was performed by performing the following of spatial response patterns (see below). (iv) Topographic maps of steps. First, the brain was extracted from the background using the SEP amplitude changes were computed for all time windows the BET tool (Smith 2002) provided in the FSL-toolbox (Smith and sessions using a linear interpolation method for smoother et al 2004, Woolrich et al 2009). The extracted brain volume still visualization. Different symbols were used to visualize the results had many segmentation faults, likely due to (i) the high signal of the statistical analyses in these maps (see captions of fgure 2 amplitude of the background formalin and; (ii) the fact that the for details). Additionally, as an alternative data set for the spec- algorithm was originally designed to extract brain tissue from tral analysis described in section 2.5.2, data within a longer period whole head scans, preferentially using T1 and T2 data simul- of −133 ms to 495 ms relative to stimulation onset were analysed taneously (Smith 2002). Then, segmentation errors were cor- and processed in the same way as described above. These data rected using custom MATLAB scripts (version R2012b, The will be referred to as ‘long-trial data’ hereafter. MathWorks Inc., Natick, MA) with a semi-automatic gray-value threshold algorithm based on 26-direction 3D food flling, 2.5.2. Analysis of relative spectral power changes. The spec- taking into account the faulty extraction to impose local con- tral analysis was performed using a time-resolved fast Fourier straints/fexibilities (Zhang et al 2001). Lastly, the (partially transform (FFT) for all trials; the data were pre-whitened to resected) dura mater was modelled using the mesh_shrinkwrap ensure that all frequency components contributed equally for algorithm implementation for MATLAB by Darren Weber the next analysis steps). Two versions with different time-fre- (freely available at http://eeg.sourceforge.net/ as part of the quency resolution were computed. First, short-trial data were BioelectromagnetismMatlab Toolbox). analysed with a sliding window of 20.8 ms and a window step of 2.0 ms in order to have a suffcient number of time- and fre- quency bins given the limited trial length, resulting in 51 fre- 2.5. µECoG data analysis quency bins (0–2400 Hz in steps of 48 Hz). These data will 2.5.1. SEP analysis. The SEP data were analysed using custom be referred to as ‘low-frequency-resolution data’ (48 Hz fre- MATLAB programs in the following way: (i) A 1 Hz high-pass quency resolution). Logarithmic relative power changes were flter (4th order Butterworth, cut-off frequency 1 Hz) was applied computed using the mean over the frst 10 time bins as a base- to the raw data to eliminate the low-frequency drifts due to hard- line, by which all absolute power values were divided. We anal- ware noise, and the data were re-referenced to a common average ysed four frequency bands in this case: 24–72 Hz (high-gamma reference across all functioning electrode contacts. (ii) Data in a band), 72–168 Hz (very high-gamma band 1), 168–264 Hz time period of −41 ms to −125 ms relative to stimulation onset (very high-gamma band 2) and 264–408 Hz (very high-gamma were cut out for each trial for all recording sessions. These data band 3). Second, the long-trial data were analysed with a slid- will be referred to as ‘short-trial data’ in the rest of the manuscript. ing window of 100 ms and a window step of 27.5 ms in order Approximately 360 trials for each recording session were acquired to obtain a higher frequency resolution compared with the

6

89 J. Neural Eng. 14 (2017) 056004 X Wang et al

Figure 3. Somatosensory-evoked relative spectral power changes recorded with µECoG. The results were from the very high-gamma band 2 (168–264 Hz) at different stimulation sites and intensities. The relative spectral power changes are shown for the same data sessions as in fgure 2. All of the conventions are described as in fgure 2.

Figure 4. Comparison of recording properties using µECoG electrode arrays with different geometries. (a) Topographic maps of relative spectral power changes recorded with the larger electrode array (left) and the smaller electrode array (right), both showing very high- gamma band 2 (168–264 Hz) responses in a time window of 5–14 ms after stimulation onset and at 1.0 mA. The same stimulation site and the same minipig were used for these recordings. Topographic maps in (b) show the relative spectral power changes in very high- gamma band 3 (264–408 Hz) from another minipig recorded in the same time window as in (a) with the larger electrode array (left) and the smaller electrode array (right), at 8.0 mA stimulation intensity. Black boxes in the left panel in (a) and (b): approximate position of the smaller electrode array relative to the position of the larger electrode array on the cortex. The conventions for signifcance are as described in fgure 2. (c) Statistical comparison of relative spectral power changes and SNR for the positive responses across all sessions (sign test, FDR-corrected, q < 0.01). Blue and cyan bars: relative spectral power (relSP) responses recorded with the larger and smaller electrode arrays, respectively. Red and magenta bars: SNR of relative spectral power responses (relSPSnr) recorded with the larger and smaller electrode arrays. ∗∗ and ∗: signifcant differences between the results obtained with larger and smaller electrode array (sign test, FDR- corrected) at q < 0.0001 and q < 0.01, respectively.

‘low-frequency-resolution data’, yielding 241 frequency bins The 16 time windows for the low-frequency-resolution (0–2400 Hz in steps of 10 Hz). These data will be referred to as data and 20 for the high-frequency-resolution data were com- ‘high-frequency-resolution data’ (10 Hz frequency resolution). puted as for the SEP data and visualized as interpolated topo- In this case, the logarithmic relative spectral power changes graphic maps. The same statistical procedure as used for the were calculated as described above but using the frst two time SEP data was applied after baseline correction. The coding of bins as a baseline period. The analysed frequency bands were electrode contacts for signifcance and the polynomial func- as follows: 0–15 Hz (alpha band), 15–25 Hz (beta band), 25–45 tion for visualizing the relationship between the relative spec- Hz (gamma band), 45–85 Hz (high-gamma band), 85–155 Hz tral power changes and the stimulation intensities (as shown (very high-gamma band 1), 155–255 Hz (very high-gamma in fgure 3(f)) were implemented in the same way as for the band 2) and 255–405 Hz (very high-gamma band 3). SEP data.

7

90 J. Neural Eng. 14 (2017) 056004 X Wang et al

Figure 5. Spatial patterns of µECoG responses. Topographic maps of SEP amplitude changes (a) and relative power changes (b)– (e) in different gamma frequency bands recorded with the larger electrode array. Rows 1 and 3 show neural effects from the same recording session in two different time windows after stimulation onset, 14–23 ms (row 1) and 23–33 ms (row 3). Row 2 shows neural effects recorded from another session during a time window of 14–23 ms after stimulation onset. Conventions for signifcance are as described in fgures 2 and 4).The topographic maps of gamma responses in each row of (b)–(e) show examples for the different numbers of SSLM (here indicated by the magenta numbers for illustration purposes). Note that the contacts in the middle of the black ellipses in row 2 show different responses at different statistical signifcance levels in different frequency bands despite being obtained in the same stimulation sessions.

2.5.3. Analysis of µECoG signal quality depending on frequency band, and session, for the larger and the smaller electrode array geometry. To assess the impact of the elec- electrode arrays, respectively. Pie charts, as shown in fgure 6, trode array geometry on the quality of µECoG recordings, were used to illustrate the distribution of specifc numbers of we compared both the relative spectral power changes and positive signifcant peaks across all sessions in selected fre- the signal-to-noise ratio (SNR) at positive maxima contacts quency bands and time windows. for the larger and the smaller electrode arrays (fgure 4). To this end, we pair-matched all recording sessions obtained in 2.5.5. Analysis of µECoG signal quality depending on each minipig using the same stimulation parameters for the electrode contact position relative to cortical vasculature. larger and the smaller arrays, resulting in a total of 85 stimu- We initially tried to determine the exact locations of indi- lation sessions with identical stimulation conditions for each vidual electrode contacts relative to cortical vasculature electrode array. For all of these sessions, the medians of the manually like in previous study (Miller et al 2009, Bleich- relative spectral power changes in each time window and the ner et al 2011). In our experience, however, this method frequency band across all trials were computed as described proved to be tedious and too subjective, and we abandoned above. In the SNR analysis, we defned the noise as the inter- this method in favour of custom semi-automatic approach quartile range (iqr) as a measure of the variability (noise) that can be described as follows. First, we acquired intra- across trials of the relative spectral power changes in each ses- operative photographs of the somatosensory cortex with and sion. SNRs for the positive maxima contacts were calculated without the implanted electrode arrays. The photographs as the trial-averaged (median) spectral power values divided with and without electrodes were then co-registered based by the noise level in each frequency band and time window. on the positions of cortical vessels using the MATLAB pro- We then applied a sign test (q < 0.01, FDR-corrected) for gram Image Registration. The colour images (fgure 1(b)) each frequency band and time window to test the positive were then converted to gray-scale photos (Adobe Photo- maxima contacts of the larger and smaller electrode arrays for shop CS3, Adobe Systems Incorporated, CA). The cortex differences in relative spectral power changes and SNRs. in these images had a light gray colour and the blood ves- sels were dark gray (fgures 7(a) and (b), and background 2.5.4. Comparison of response map complexity depending image). The electrode contact centres were determined in on electrode array geometry. To compare the spatial com- the photographs compared with the implanted electrode plexity of responses between larger and smaller electrode contact positions on the cortex. The mean gray values of arrays, we detected the number of positive signifcant peaks the area underneath each contact were computed based on (referred to as signifcant spatial local maxima (SSLM)) in the photographs of the cortex without implanted electrode each topographic map. For this analysis, matched sessions arrays. The mean gray values were high if an electrode con- were used as in section 2.5.1. Positive signifcant peaks were tact was positioned on the cortex and low if it lay above identifed as responses with positive signifcant relative spec- a vessel. Accordingly, we defned contacts as ‘cortex con- tral power changes higher than those of all of their direct tacts’ if the mean gray values were above the 75th percentile neighbours (q < 0.01, FDR-corrected; as marked in fgures 5 of all mean gray values across contacts, and contacts with and 6). For this analysis, contacts were required to have at mean gray values below the 25th percentile were defned least three neighbouring contacts with a distance equal to the as ‘vessel contacts’. Contacts between the 25th and 75th inter-contact distance of the respective electrode array. This percentiles were excluded from this analysis because the was not the case for some contacts at the edge of the electrode assignment to either cortex or blood vessels was not always arrays, precluding an assessment of the neuronal responses unambiguous in this group. The median values of relative in the adjacent area (not covered by electrode contacts). The spectral power changes for the ‘cortex contacts’ and ‘ves- number of SSLMs was calculated for each time window, sel contacts’ groups were calculated for each frequency

8

91 J. Neural Eng. 14 (2017) 056004 X Wang et al

Figure 6. Topographic complexity of µECoG responses observed with different electrode array geometries. (a) Top row: topographic maps of SEPs recorded with the larger (left) and the smaller (right) electrode arrays (5–14 ms time window, 1.0 mA stimulation intensity). The black box in the left panel indicates the position of the smaller electrode array relative to the larger electrode array. The pie charts colour- code the number of distinct peaks found across recording sessions, with the whole pie chart representing all 188 recording sessions from all minipigs. (b) The same fgure components as in (a) but in a different minipig and in a different time window (14–23 ms). (c) The number of SSLM averaged across sessions for the larger and smaller electrode arrays. The symbol colours indicate the time windows relative to the onset of stimulation. Symbol shape: triangles indicate highly signifcant differences between the number of SSLM obtained with the larger and smaller electrode arrays (p < 0.005, Wilcoxon rank sum test), squares: 0.005 < p < 0.05, circles: non-signifcant differences ( p ⩾ 0.05). The symbols above and below the dotted diagonal indicate cases with more SSLM with the larger and smaller electrode array, respectively. (d)–(o): the same results for different gamma frequency components; all of the conventions are as described in (a)–(c). The topographic maps in each column of the fgure show the neuronal effects that were observed in different frequency bands simultaneously during the same stimulation session. The larger electrode array yielded more SSLM in the SEP time domain (c), especially in the later time windows. With increasing frequency, the number of SSLM observed with the smaller electrode array increased compared with the larger electrode array, particularly in the earlier time windows. band, each time window and each session separately. To 3. Results determine whether there was a signifcant response dif- ference between the contacts on the vessels and those on 3.1. µECoG responses dependent on stimulation site the cortex, a statistical comparison of the median relative and intensity spectral power changes was performed between these two Using the novel µECoG electrode arrays, we were able to reli- contact groups across all sessions for each frequency band ably record spatially fne-grained cortical activity patterns. and each time window (Wilcoxon rank sum test, q < 0.01, Figures 2 and 3 depict representative topographic maps of SEPs, FDR-corrected).

9

92 J. Neural Eng. 14 (2017) 056004 X Wang et al

Figure 7. The effects of cortical blood vessels on µECoG signal. (a) Topographic map of relative spectral power changes (very high- gamma band 3, i.e. 168–264 Hz, time window 14–23 ms after stimulation onset, smaller electrode array, 1.0 mA stimulation intensity). Conventions for signifcance are described as in fgures 2 and 4. Yellow square: area that segregates the two spatial distinctive foci of cortical activation. (b) Photo during the experiment showing the smaller µECoG electrode array on the cortex of the same minipig as in (a). The position of the yellow square shows the same electrodes as in (a). Note that both contacts appear to be in direct contact with the cortical surface, not with a major blood vessel. To verify cortical contacts, we used additional photographs of the cortical surface without the electrode array in place (c), (d). The left most panels in (c) and (d) show photos of the larger (c) and the smaller (d) µECoG electrode arrays placed on the cortex of one minipig during the stimulation experiment. The red dots indicate the centres of each contact, and the yellow rectangle area in (c) shows the approximate location of the smaller electrode array (shown in (d) relative to the larger array. The second columns in (c) and (d) show which contacts were assigned to the cortex (cyan) and which to the vessels (magenta) based on the mean gray values from the high contrast black–white background photograph under each contact (see section 2). The white circles indicate electrodes at which no clear differentiation was possible. The remaining two panels in (c) and (d) show the topographic maps of relative spectral power changes in high-gamma band (24–72 Hz, third column of (c) and (d)) and very high-gamma band 2 (168–264 Hz, fourth column of (c)), as well as very high-gamma band 3 (264–408 Hz, fourth column of (d)). These maps were generated using data from the time window 14–23 ms after stimulation onset at individual contacts of the larger (c) and smaller (d) electrode arrays. Contacts with signifcant changes (sign test, FDR-corrected, q < 0.01) are marked with flled circles. The yellow solid lines in (c) and (d) show that some areas with signifcant µECoG responses are spatially separated by areas without signifcant effects and this separation cannot be explained by blood vessels. Thus, the observed spatial structure appears to be a genuine property of the cortical activation pattern rather than an artefact caused by blood vessels. (e) A statistical comparison of the relative spectral power changes recorded on the cortex (cyan) versus vessel (magenta). The median values and their respective standard error over all 101 sessions are shown for all frequency bands. ∗∗: signifcant differences between µECoG recordings from vessels and cortex (Wilcoxon rank sum test, FDR-corrected, q < 0.01). the corresponding logarithmic relative spectral power changes (fgures 2(e) and 3(e)), demonstrating that the SEP amplitude for the same minipig, and the stimulation parameters obtained (fgures 2(d) and (e)), the relative power changes (fgures 3(d) with the larger electrode array. These data indicate that SEP and and (e)) and the number of contacts with signifcant effects spectral power changes exhibited a strongly overlapping topog- were modulated by stimulation intensity. Additionally, for raphy (fgures 2(a), (c), (d), (e) and 3 (a), (c), (d), (e)). SEPs, when the number of electrode contacts with signifcant Both site-specifc and intensity-related topographically effects increased with higher intensity, the signifcant effects focalized SEPs and gamma responses were reproducibly were observable in two distinct areas and did not spread to observed. For instance, changes of the stimulation sites caused all of the other contacts in between. Analysis of the relation a spatial shift of the main response (fgures 2(a), (c) and 3(a), between stimulation intensity and the amplitude of the neu- (c)), while all other stimulation parameters were kept constant. ronal responses for both SEPs (fgure 2(f)) and relative power Changes of stimulation intensity also infuenced the spatial changes (fgure 3(f)) indicated that the response amplitude extent of stimulation-related neuronal effects. Figures 2(d), increased with the intensity of stimulation, as expected. High (e) and 3(d), (e) show the topographical maps with stimula- spatial specifcity of neuronal effects and their dependence on tion intensities at 1.0 mA (fgures 2(d) and 3(d)) and 8.0 mA stimulation intensity were also observed with the smaller elec- trode array.

10

93 J. Neural Eng. 14 (2017) 056004 X Wang et al 3.2. Infuence of electrode array geometry on the amount spatial patterns a genuine property of cortical activation, or of spatial detail and the quality of neuronal recordings are they possibly artefacts caused by cortical vasculature? For example, the µECoG response map in fgure 7(a) shows two dis- In the present study, two different µECoG electrode arrays tinct regions with signifcant responses that are separated by a were utilized, with 3.5 mm and 1.68 mm inter-contact distances valley with no signifcant responses (yellow box in fgure 7(a)). and 1.81 mm and 0.87 mm contact diameters, respectively. ‘ ’ Was the cortex in this region truly silent, or was a blood vessel Neuronal responses elicited by peripheral electrical stimula- running between the electrode contacts and the cortex, possibly tion and recorded with the smaller electrode array revealed preventing responses from being detected? We used intraopera- more topographical details than those obtained with the larger tive photographs, both with the electrode arrays in place (fgure electrode array in the same cortical region ( fgures 4(a) and 7(b), left-most panels in fgures 7(c) and (d)) and of the same (b)). In the examples in fgures 4(a) and (b), activity in the very region of exposed cortex without electrode arrays (fgures 7(c) high-gamma bands 2 and 3 recorded with the larger electrode and (d), remaining panels) to address this question. array (fgures 4(a) and (b), left panel) revealed a single, spa- In many examples, comparison of response and vascular tially extended response pattern. Using the smaller electrode patterns showed spatial response structures that could not be array placed on the same area, however, several spatially seg- explained by vascular effects (fgures 7(a) (d)). Furthermore, regated peaks could be clearly distinguished (fgures 4(a) and – we found that the location of electrode contacts, i.e. on the (b), right panel). These peaks were separated by electrode con- cortex or on blood vessels, generally did not signifcantly tacts with non-signifcant responses (marked by larger flled- impact the neuronal responses (fgure 7(e)).The only excep- gray circles and smaller flled-black circles, see section 2). tion was the level of beta band (15 25 Hz) responses, which To investigate whether the geometry of the electrode arrays – were signifcantly higher at electrode contacts lying directly affected signal quality, we compared the SNR and the max- on the cortex than on the vessels. These results suggest that at imum amplitudes of spectral power changes observed with the least the level of gamma band activity is not distorted by the larger and the smaller electrode arrays. This analysis shows presence of cortical blood vessels. that the maximum amplitudes (see section 2), especially for very high-gamma responses, were signifcantly higher (sign test, FDR-corrected) in recordings with the smaller electrode 4. Discussion array, while the SNR was better in recordings with the larger electrode array (fgure 4(c)). 4.1. Suitability of µECoG electrode arrays for cortical recordings 3.3. Topographic complexity of µECoG responses observed with different electrode array geometries The spatially and functionally specifc patterns of cortical activity in the present study demonstrate that µECoG is well As shown in section 3.2, the spatial structures of µECoG suited to obtain high-quality neuronal recordings of cortical response maps likely refect different cortical activation patterns activity. Our recordings had high signal-to-noise ratios and in different spatial scales. Given this fnding, an examination of were sensitive to spatially and temporally very local changes the role of µECoG electrode array geometry on the topographic in the neural signal. The high amount of spatial detail that can complexity of the µECoG responses was the next reasonable be seen in our functional maps may be attributed to the innova- analysis step. To this end, we used the number of SSLM in the tive procedures implemented to fabricate the electrode arrays. µECoG response maps as an index of topographic complexity. Customized non-penetrating fexible µECoG electrode arrays To obtain a comprehensive overview of topographical com- from IMTEK, University of Freiburg, were used to record brain plexity, the number of SSLM was determined for both electrode activity in the somatosensory cortex of the minipig. Compared array types and each session individually for averaged frequency to other techniques that use polyimide and manufacturing with bands and time windows. These values were used as the basis MEMS technology (Rubehn et al 2009), the electrode fabrica- for the results presented in this section. For example, in the very tion technology applied in this study permits a shorter design- high frequency maps shown in fgure 5(e), the number of SSLM to-proto type time. This bene ft is gained because the actual ranged from 1 to 3. The larger electrode array yielded more geometry is designed as a computer-aided design (CAD) fle and SSLM in the SEP response maps (fgures 6(a)–(c)), at least in transferred to a laser that ablates silicone and metal in a highly the later time windows. However, with increasing frequency of automated process (Henle et al 2012). In contrast to most com- the analysed µECoG signal component, the number of SSLM mercially available silicone rubber-based electrode arrays, which observed with the small electrode array increased compared are predominantly handmade and therefore restricted in feature with the larger electrode array (fgures 6(d)–(o)). Particularly in sizes to the millimeter range (Stieglitz et al 2009), this process the earlier time windows, this difference between the larger and allows for electrode arrays in the sub-millimeter range. In this smaller electrode array was highly signifcant (rank sum test, work, we show that the designed electrode arrays have great p < 0.005, triangles in fgures 6 (f), (i), (l) and (o)). potential for reliable and robust recording of neuronal assembly activity on millimeter (larger electrode array, contact Ø 1.81 mm, inter-contact distance 3.5 mm) and sub-millimeter (smaller elec- 3.4. Infuence of cortical vasculature on the µECoG signal trode array, contact Ø 0.87 mm, centre-to-centre inter-contact dis- The results in sections 3.2 and 3.3 show that µECoG can cap- tance 1.68 mm) scales up to very high frequency bands. Medical ture cortical responses with fne-grained spatial detail. Are these grade silicone rubber and high-purity platinum foil are known to

11

94 J. Neural Eng. 14 (2017) 056004 X Wang et al be stable and well accepted by the body in long-term implant- the results presented here tend to support the fndings of Worrell ations (several decades), allowing for electrode arrays designed et al (2008) and thus suggest an impact of electrode contact for animal studies to be used in human clinical trials with min- size on cortical recordings, the optimal electrode contact size imal legislative hurdles. In addition, electrode arrays covering a for minipigs and humans were only estimated, and still remain smaller cortical area may reduce clinical surgical risks and com- unclear. Even the question arises, whether there are several plications, such as subclinical seizures or trauma, which can arise optimal electrode contact sizes, depending on the area and the due to decreased pressure over the cortical surface (Wong et al activity pattern that is supposed to be investigated, rather than 2009). These features of the designed µECoG electrode arrays, a universal size that can be applied to arbitrary areas and pat- namely, enhanced signal detection, material properties and terns. For instance, HFO generators (especially FR generators) reduced clinical risk, make them promising candidates for long- were found to be less than 1 mm3 (Bragin et al 2000, 2002). If term implantation in human clinical applications. the electrode contact size exceeds the area of interest, a larger electrode contact might record events in a larger area, i.e. more neuronal populations, leading to a blurred recording and to a 4.2. The role of electrode array geometry in µECoG reduction in total recorded signal strength. Such behaviour was signal quality observed in the study by Worrell et al (2008) as well as in the The frst primary question addressed in this study concerns the present study. However, for non-HFO activity or other types of role of electrode array size (contact diameter and inter-contact general activity, electrode contact size might be chosen differ- distance) in neuronal µECoG recordings and, in particular, its ently in order to achieve an optimal spatial sampling. impact on the detection of high-frequency signal components. In a BMI study with monkeys performing closed-loop Optimal contact diameters for subdural electrode arrays BMI control tasks, Rouse and his colleague (Rouse et al were suggested to be 0.7 mm for rats and 1.25 mm for humans 2016) suggested optimal inter-contact distances to be 3 mm based on fnite element modelling and spatial spectral analysis for independent epidural recordings. For humans, this dis- (Freeman et al 2000, Slutzky et al 2008). An experimental tance might be larger due to the thicker dura mater (Bundy study by Worrell et al (2008) compared contact areas between et al 2014, Rouse et al 2016). However, a better signal quality 2 2 9.4 mm (Ø 1.73 mm) and 0.0012 mm (Ø 40 µm, micro-wires) can be obtained if the µECoG electrode array is implanted in humans, with the former being larger than the stated optimal subdurally (Bundy et al 2014). Furthermore, the infuence diameter for humans and the latter much smaller. This previous of inter-contact distance was found to be different between study found that high frequency oscillations (HFOs) were BMI control tasks and normal neuronal recordings (Rouse more frequently recorded using micro-wires, suggesting that et al 2016). Specifcally, for normal neuronal recordings, the small contacts are more suitable to detect the sub- millimeter signal from each contact site represents the activity directly generators of HFOs (Worrell et al 2008). However, another under the contact and is minimally distributed between the study by (Chatillon et al 2011) compared three different con- electrodes. However, in the BMI control task, the signal might 2 2 tact areas between 0.849 mm (Ø 0.52 mm) and 0.018 mm (Ø be distributed depending on the utilized decoding strategy. 0.08 mm) in rats. Both of these were smaller than the proposed Thus, the practical limit for normal neuronal recordings using optimal diameter for rats, although the results suggested that µECoG is likely smaller than 3 mm. Ours is the frst study to there was no difference in HFO detection. The same study also probe the impact of inter-contact distance in normal neuronal suggested that there should be no difference in HFO detec- recording. It is clear that more spatial details can be obtained tion in human recordings using contacts with sizes between with electrode arrays that have smaller inter-contact distances 2 2 1.698 mm (Ø 1.04 mm) and 0.036 mm (Ø 0.16 mm), smaller (1.68 mm; topographic mappings in fgures 4 and 7). We can than the proposed optimal diameter for humans. Based on also conclude based on the present results that the smaller the results of the present study, given the comparability of the electrode array does not provide the same data as a macro elec- minipig’s gyrencephalic brain to the human brain in terms of trode array; rather, each contact can record neuronal activity general anatomy, growth and development (Jelsing et al 2006), independently. Thus, it is worth seeking the practical limit for the optimal electrode contact diameter for the minipig might normal neuronal recording if the technical conditions permit. be between 0.7 and 1.25 mm (the proposed diameters for rats A related question arising from electrode contact miniaturi- and humans), but presumably closer to 1.25 mm. The fndings zation is how the impedance increase associated with electrode presented here support this range of proposed optimal elec- contact miniaturization impacts neural recordings and whether trode contact diameter given that the smaller electrode array the described reduction in SNR might be due to these imped- with a diameter of 0.87 mm is nearer to the proposed optimal ance changes. Usually, the ratio between the electrode contact diameter for humans than the larger electrode array with a impedance and applied amplifer impedance (rather than the diameter of 1.81 mm. Moreover, the electrode array with the contact impedance alone) is the main factor for assessing the smaller diameter (0.87 mm) recorded signifcantly higher rela- quality of neuronal recordings. If this ratio is <1%, the elec- tive spectral power changes (especially in the very high-gamma trode contact size appears to have no effect on impedance- 3 frequency band, up to 408 Hz, fgure 4(c)) in the cortex area related signal attenuation (Chatillon et al 2011). In previous with the main neuronal responses. This result is similar to that reports as well as in the present study, the observed ratio was observed in Worrell’s study (Worrell et al 2008). Although, we consistently near this limit (electrode contact impedance <5 were able to detect differences in recording performance using kΩ and amplifer input impedance >10 GΩ in the present different electrode contact sizes (fgures 4(a), (b) and 7), and study), excluding the possibility of neuronal signal distortion

12

95 J. Neural Eng. 14 (2017) 056004 X Wang et al due to impedance. Thus, cortical volume, i.e. the neuronal pop- for the visual cortex in cats (Siegel and König 2003) rats (Oke ulations beneath the electrode contact, was the main factor for et al 2010) and humans (Tallon-Baudry et al 2005, Chaumon the observed reduction in signal strength. However, a differ- et al 2009). Such overlapping patterns have been proposed to ence between smaller and larger contact impedances exists in indicate different aspects of information processing. the sense that smaller contacts usually have higher impedances Whether the increased complexity observed in this and the (Stieglitz 2009). This could be the reason why signals recorded aforementioned studies refects the actual activity in the under- with larger contacts had a better SNR as shown in fgure 4(c). lying tissue and how it can be reconciled with fndings from studies investigating single- and multi-unit activity (SUA and MUA) remains unclear. µECoG signals refect the contrib- 4.3. Assessment of topographic complexity captured utions of electrical feld variations (postsynaptic potentials, with µECoG electrode arrays of different geometries PSPs) arising at different cortical layers. The larger and more The second main objective of the present study was to inves- numerous pyramidal cells have somata residing primarily tigate not only whether µECoG electrode arrays can record in cortical layer V and dendritic arbores in more superfcial spatially segregated activity peaks (such as in the studies by layers, i.e. closer to the µECoG contacts on the brain surface Hosp et al (2008), Wang et al (2009) and Wang et al 2016) but (Zanos 2009). This arrangement was demonstrated by Barth also whether different electrode array geometries affect the and Di (Barth and Di 1990) in the rat auditory cortex using a representation of topographical complexity in terms of both micro- electrode array of cortex-penetrating depth electrodes the number of SSLM and their respective frequency bands. to record local feld potentials (LFPs). In this study, the frst As mentioned in the Introduction, topographic maps pre- component of the auditory evoked potential arose from layer sented in previous studies often contained only one activated II and the second component from layer IV and V. In addi- area, representing the main response. However, topographic tion, the rates of excitatory and inhibitory PSPs generated maps with two or multiple segregated, respondent areas were at the dendrites of pyramidal cells differ from the respective rarely reported. These results raise the question of whether the cell’s fring rate, which can differ even between neighbouring observation of a single activation peak refects the actual activa- pyramidal cells. A previous study suggested that coexisting tion pattern or rather results from the summation of larger (or low- and high- gamma oscillations may originate from dif- multiple) neuronal populations. Such sources would be distin- ferent cortical layers due to the cross-layer arborisation of guishable when using electrode arrays with smaller inter-con- both excitatory and inhibitory axons, especially for responses tact distance and electrode contact diameters, respectively. For in the higher frequency bands (Watts and Thomson 2005). instance, using microelectrodes with a sub-millimeter scale, Siegel and König (2003) proposed that the high-gamma cor- Hosp et al (2008) showed that the spatial representation of the tical oscillations refect an emergent property of local net- SEP P1-N1 amplitude evoked in the hindlimb area of the rat works rather than activity driven by fast thalamic oscillations, was separated into two activity foci, a medial and a lateral feld, as has been previously proposed and that these oscillations in three of fve rats. Spectral power maps of fnger movement- are little affected by inputs from other networks. There are related activity in humans were obtained from electrode arrays also studies investigating the somatosensory cortex of mon- with 4 mm inter-contact distances. These data showed multiple keys showing that 600 Hz oscillations are modulated by soma- segregated areas around micro-scale activation peaks (Wang tosensory stimuli that refect the timing of cortical spike bursts et al 2009, 2016), similar to the outcomes reported for a larger (Baker et al 2003) and that multi-unit activity usually refects electrode array (3.5 mm inter-contact distance) in this study. the spiking outputs of neurons in an area. Furthermore, the present results show that compared to the larger Taking all of this into account, it can be assumed that cortical electrode array, the smaller electrode array (1.68 mm inter-con- recordings refect a substantial temporal and spatial summa- tact distance) can robustly detect a higher number of separate tion of overlapping activity, both within and between neigh- activation areas in the same cortical location (fgure 6). We found bouring neuron populations, due to the different fring rates and a quantitative measure (SSLM) to address these topics by com- PSP rates. The responses at each electrode contact site would paring two electrode arrays with different spatial resolutions; therefore be the result of asynchronous contrib utions from sub- specifcally, these electrode arrays were used to record activity populations of cells in different lamina, and (very) high-gamma- elicited by unvarying stimuli in the same area of the cortex. frequency oscillations recorded in the somatosensory cortex We were able to show that especially in the high-frequency of the minipig may be due to emergent properties of local net- domain, the geometric properties of electrode arrays infuence works. This arrangement could generate the multiple activation the mapped topographic complexity (fgure 6), a difference that peaks observed in the topographic maps presented here. can be expressed in terms the number of SSLM quanti ty. The number of SSLM recorded with the smaller electrode array was 4.4. Impact of cortical blood vessels on µECoG signals higher than for the larger electrode array, especially in very high frequency bands. We also found very high-gamma-frequency The third main question addressed in this study was the role of oscillations (up to 408 Hz) coexisting with low- or classical cortical blood vessels in µECoG recordings and whether they gamma frequency oscillations in overlapping, partially overlap- infuence signal components. A previous study by Bleichner et al ping and spatially segregated cortical locations in the minipig’s (2011) investigated the effect of blood vessels in macro-ECoG somatosensory cortex. A similar coexistence of low- and high- recordings, showing that the absolute power spectral density in gamma-frequency oscillations was also observed and described the frequency band from 30 to 70 Hz was clearly attenuated

13

96 J. Neural Eng. 14 (2017) 056004 X Wang et al for electrode contacts located on blood vessels (Bleichner surgical procedure. Fentanyl is an intravenous opioid agent et al 2011). In the present study, we compared relative spectral that can suppress the activity of the central nervous system power changes. We assume that these values are more robust but has been proposed to be less reliable as an aesthetic and than measures of absolute power given that baseline variation- to have signifcant side effects (Philbin et al 1990, Streisand corrected values better refect differences between cortex and et al 1993). Therefore, fentanyl is frequently used in combina- blood vessel electrode contact groups in spectral powers of tion with other agents to maintain anaesthesia due to its pri- different frequency bands (Leuchter et al 1987, Leuchter and mary action as an inhibitor of nociception (Miller 2005). It Walter 1989). In addition, we used a different approach to has been reported that opioids have little effect on spinal and defne the electrode contact position compared to the method of subcortical SEP recordings (short-latency SEP) but result in a (Bleichner et al 2011) where electrode contact positions were slight reduction in the amplitude of cortical responses (long- defned based on the opinions of two independent observers. latency SEP) and longer latencies for late cortical responses To be more objective regarding electrode contact assignment (Bithal 2014). However, propofol was the primary aesthetic in (i.e. cortex or vessel), we used the 25 percentile and 75 per- the present study. Most studies investigating SEPs under GA centile of all electrode contact gray values as an assignment merely induced using propofol only suggest that it decreases threshold rather than the median. Electrode contacts with gray the amplitude of the cortical responses (Kumar et al 2000, values falling between this range were excluded from further Bithal 2014). However, in combination with opioids, propofol analysis (white circles in 2nd columns of fgures 7(a) and (b)). reduces the amplitude of evoked potentials less than other GA Although this objective approach reduces the amount of infor- agents (Schwartz et al 1997) and thus is the most common mation available for the analysis, this processing step ensured option for monitoring cortical SEPs (Calancie et al 1998). that we compared signals that certainly came from electrode However, the reported infuence of GA on the functional contacts either on the cortex or on vessels. Using this method, state of the cortex varies depending on the used agent, dose and we show that if all investigated parameters (i.e. data from both network studied. Even the subject’s age, race, or body temper- electrode arrays with high-frequency-resolution data sets) are ature may infuence the obtained results. Thus, the agent’s uti- included in the analysis, there are no signifcant differences lization or its infuence needs to be considered for each case between the relative spectral power changes in the gamma fre- individually. It has been suggested from studies on auditory or quency band as measured at the cortex and vessel electrode visual evoked potentials (AEPs, VEPs) that external stimulus- contacts, and there was a signifcant decrease for blood vessel evoked brain responses are suppressed in cortical regions of electrode contacts in the beta band (15–25 Hz). Such a decrease higher-order information processing under GA; however, such in beta frequencies could be related to the properties of the cor- suppression is not observed in the primary sensory cortices, tical layer(s) in which this activity originates. High-frequency which engage in low-order information processing (Kerssens oscillations may come from more superfcial cortical layers et al 2005, AEP/sevofurane; Dueck et al 2005, AEP/propofol; (Oke et al 2010). A contact lying on a vessel would lie farther Ramani et al 2007, VEP/sevofurane; Boveroux et al 2010, from the source of cortical activity, resulting in a greater rela- AEP, VEP/propofol). In contrast, studies on SEPs elicited by tive distance increase of generators are in superfcial cortical tactile or painful stimuli showed stronger suppression in the layers, possibly contributing to frequency-specifc differences primary somatosensory cortex (Antognini et al 1997, Kumar in the attenuation of responses. In fact, results might be more et al 2000, Bonhomme et al 2001, Bithal 2014). This effect complicated than what we presented here, and they should be that may be due to the inhibition of ascending sensory infor- addressed in more detail in further study, for example, is the mation (stimuli input) in the spinal cord, which is absent in contact size plays a role in such kind of analysis, or if we focus AEP or VEP recordings. Therefore, one can presume that on more specifc time/frequency range, the tendency will also if the somatosensory stimulation is applied at sites that are keep constant? However, at the very least, these data provide a innervated independently of the spinal cord, e.g. facial nerves, general hint that signifcant effects might not be infuenced by activations in lower-order cortical areas might persist, such as the contact location but rather by the source of the signal. That was proposed to be the case for AEPs and VEPs. is, if the source is under the vessels, a signifcant effect can be Furthermore, SEPs consist of both short- and long-latency detected even if the contacts lie above it. components. The results from human clinical applications have indicated that GA has a stronger suppressive effect on long-latency than short-latency SEPs (Kumar et al 2000, Uhrig 4.5. Infuence of anaesthesia on µECoG signals et al 2014). Huotari and his colleagues (Huotari et al 2004) As all of the results in this study were obtained under general found that short-latency SEPs evoked by painful stimuli at anaesthesia (GA), an important question is whether there are the median nerve showed no differences whether they were differences in the brain responses that are evoked by external recorded in awake patients’ or those in propofol-induced GA. sensory stimuli in the awake and GA state. Zhang and his colleagues (Zhang et al 2014) compared the In the present study, GA was induced by propofol and amplitude and latency differences of short- and long-latency maintained with propofol, fentanyl and pancuronium SEPs elicited by whisker defection in rats. The responses were (section 2.1). Pancuronium is a muscle relaxant or neuromus- recorded in various cortical areas in animals under GA induced cular blocking agent, the latter of which has been reported to by different propofol doses. Short-latency SEPs exhibited no have no effect on SEPs (Sloan 1998, Bithal 2014) and is usu- dose-dependent differences, whereas long-latency SEPs were ally applied after the induction of anaesthesia to facilitate the delayed and showed decreased amplitudes. This prolonged

14

97 J. Neural Eng. 14 (2017) 056004 X Wang et al latency effect of late components were also found in other contacts only has a signifcant impact on beta-band power studies using different animal models and GA agents. For but not on gamma-band power. To further clarify the roles of example, this effect was observed in cats for VEPs using pento- spacing, size and spatial complexity in different frequency barbital (Robson 1967), in rats for SEPs and halothane (Chapin bands, suitable studies for investigating the smooth trans ition et al 1981), and in rats for VEPs with desfurane (Hudetz et al between SUA/MUA activity and µECoG signals must be con- 2009). The reason why short-latency components are unaltered ducted, such as the one recently published by Khodagholy under GA is unclear, although it has been proposed that these et al (2015). In this previous analysis, a micro- electrode array responses emerge from the primary sensory cortices and that was employed to record LFP signals as well as single spikes feed-forward projections might be more insusceptible to GA, from the surface of the brain in epilepsy patients. Of special such as for the abovementioned lower-order processing areas. interest in this context is (i) how we can investigate the specifc However, the previous studies mentioned here were pri- electrode array geometry that is suitable for different activity marily focused on evoked brain responses, amplitudes or patterns and (ii) how micro- electrode arrays will advance our latency alterations in the time domain. Only few studies have understanding of the complexity of spatial cortical activity described alterations of responses between the awake and GA and of the multiple separated areas around activation peaks. state in the spectral domain. Most of these studies investigated Lastly, new strategies must be developed to translate the out- the results only with respect to an overall power alteration in comes of these basic questions to advanced BMIs, high-res- the form of power-law plots that included quasi-continuous olution cortical mapping and other therapeutic neurological frequencies (Zhang et al 2014, Insanally et al 2016). There was and neuropsychological clinical applications. A previous study one study from Mhuircheartaigh’s group (Mhuircheartaigh reports that infection rates as an implantation-related compli- et al 2013) in which the results were presented in terms of cation in patients with conventional ECoG recordings were different frequency bands. These data showed that there was increased with more than 100 electrode contacts, more than a slow-activity (0.5–1.5 Hz) saturation after a specifc dose ten cables, more than one cable exit site, and more than 14 d of propofol in humans. In this study, however, the evoked of implant ation (Wiggins et al 1999). Thus, also with µECoG responses were elicited by various stimulus modalities. implantation risks may be higher if a larger number of µECoG Taking the fndings from all of these important studies into arrays need to be implanted at the same time. Alongside the account, we conclude that propofol-induced GA in combination surgical procedure per se, also other factors such as µECoG with opioids suppresses SEPs primarily due to the interruption array geometries or the length of the implantation period of high-order information integration. Moreover, this effect pri- may be relevant. We recognize that the amount of electrode marily manifests as reduced amplitude and prolonged latency grids, electrode geometries and the design of the implant are of the long-latency component. We further conclude that SEPs certainly factors that should not be neglected when planning evoked by spine-independent facial electrical stimulation should µECoG procedures for clinical applications in humans. The be less affected by GA. This was the case for AEPs and VEPs, exact choices related to these factors will depend on the goal two modalities that are processed independently of the spinal of the project, and informed solutions need to be designed with cord. The results in the present study were primarily obtained caution to achieve an optimal balance in the trade-off between from early SEP responses recorded from the lower-order pri- implantation risk and the desired spatial coverage. mary somatosensory cortex and were elicited by electrical stim- ulation at the nostrils of minipigs. Moreover, these recordings Acknowledgments were made when the animals were primarily under propofol- induced and -maintained GA. We therefore propose that the SEP CH is affliated with the new spin-off company CorTec GmbH results presented here should be largely uninfuenced by the GA. that was funded by a BMBF-grant of the Federal Republic Unfortunately, with respect to our spectral domain analyses, we of Germany and that developed one of the µECoG electrodes cannot reach such a frm conclusion. Nonetheless, we are con- used in this study. MG, TS, and TB are scientifc partners of fdent that the results presented here provide important insights CorTec GmbH and do not have any fnancial interest in or in the role of electrode array geometry. We note, however, that affliation with CorTec GmbH. The remaining authors (XW, alterations between the awake and GA state in the spectral AG, OI, LF, IM, and JK) did and do not have any affliation domain still need to be investigated more thoroughly. with CorTec, and they do not have any other confict of inter- est to declare. This work was funded by Deutsche Forschungsgemeinschaft, 5. Conclusions DFG, (grant EXC1086 BrainLinks-BrainTools) and the German Federal Ministry of Education and Research, BMBF We showed that for the somatosensory cortex of the minipig, (grant 01GQ1510 OptiStim and 13GW0053D MotorBic). the spatial information extent increases when using electrode They were not involved in study design, collection, and anal- arrays with fner spatial resolution in terms of both inter-con- ysis, interpretation, writing or decision-making. tact distance and electrode contact diameter. This increase in The authors thank Katharina Foerster and Professor Joerg information is observed especially in the low, high and very Haberstroh for perioperative management in the experiments high-gamma band and can be quantifed in terms of the number and Wolfgang Meier for interconnecting the µECoG electrodes. SSLM. In addition, smaller contacts might perform better in detecting high-frequency signal components. We also showed that the presence of blood vessels subjacent to electrode array Appendix

15

98 J. Neural Eng. 14 (2017) 056004 X Wang et al )

& ection f Continued ( # # Vibrissa Vibrissa complex sonototopy sonototopy sonototopy sonototopy sonototopy sonototopy sonototopy sonototopy somatotopy somatotopy auditory-click Auditory-click Auditory-click Auditory-click Auditory-click and auditory-click and auditory-click and auditory-click of an averaged IIS of an averaged Whisker-de Topographical map Topographical Vibrissa somatotopy somatotopy Vibrissa Vibrissa somatotopy Vibrissa Vibrissa somatotopy somatotopy Vibrissa Vibrissa somatotopy Vibrissa Vibrissa somatotopy somatotopy Vibrissa # # # # # # # # # # High frq. focalization map Topographic no foc. disc. no foc. disc. no foc. disc. no foc. disc. Up to 40 Hz, Up to 40 Hz, Up to 200 Hz, Up to 600 Hz, . f # # in vessel vessel Blood Results ## maps latency latency 2 peaks signs of peaks in Multiple peaks but peaks but responses activation for longer- topographic No multiple multi-column ### ### ### ### # ### ### ### ### ### ### # ### test) Human for human application (cytotoxicity (cytotoxicity F F F F T T T T T T T T & & & & T T T T # # # # # # # # # Instruments) Instruments) Instruments) Non-commercial T Suitable (Rhodes Medical (Rhodes Medical (Rhodes Medical Pt Pt Commercial Pt Commercial Pt Commercial Pt Pt Ag Au Non-commercial T Ag Ag ball steel steel Stainless Stainless Pt within electrode Ag-AgCl biomedical grade PDMS 8 8 8 4 16 8 8 8 8 8 8 8 8

× × × × × × × × × × × × × #

8 8 8 4 8 8 8 8 8 8 8 8 16 ECoG studies in recent decades. ECoG studies Electrode µ 2 ) ) ) ) ) ) 2 2 2 2 2 2 ) ) ) 2 2 2 m µ 5.25 mm 5.25 4 mm 4 7.5 mm 7.5 3.5 mm 3.5 mm 3.5 3.5 mm 3.5 4 mm 4 mm 4 mm 3.5 3.5 mm 3.5 #

× × × × × × × × × × 0.5 mm 0.5 0.5 mm 0.5 0.5 mm 0.5 mm 0.5 0.5 mm 0.5 0.5 mm 0.5 mm 0.5 mm 0.5 0.5 mm 0.5 0.5 mm 0.5 mm 400 Spacing

(4 (4 (4 (covered area)(covered Size Material Type Analysis (7.5 (3.5 (3.5 (3.5 (3.5 (3.5 5.25 of Overview m m m m m m m m m m m m m µ µ µ µ µ µ µ µ µ µ µ µ µ Ǿ # square 100 100 100 100 100 100 100 100 100 150 100 200 100 Table A1.

&

ection- f SEPs VEPs in SCx activity MAEPs MAEPs MAEPs MAEPs MAEPs, activity potentials electrical/ SEPs, ASEPs SEPs, spontaneous related SEPs AEPs, vibrissa AEPs, vibrissae spontaneous activity related SEPs, VEPs related SEPs, MSEPs, spontaneous mechanical stimulus- Task

& & Acute Acute Epileptiform Acute Acute AEPs, vibrissa Acute SEPs Vibrissa Acute chronic chronic Acute Acute Epid. Acute Whisker-de Epip. Acute Epip. Acute MAEPs, vibrissa Subd. Acute (subd.) (subd.) (subd.) (subd.) (subd.) (subd.) (subd.) SCx & PV ACx near midline barrels (SCx) Rats Right PT Epip. Acute SEPs Vibrissa 1 pig Left OCCx a 4 rats Right whisker 4 rats Right PT Epic. 4 rats Right SCx. Epic. 7 rats Right PT Epic. 5 rats7 ratsPT Right Epic. Right PT Epic. 4 rats Right 4 rats Right PT Epic. 4 rats Right PT Epic. 11 rats Right S1S2, 15 rats Right clNCx Epip. Acute 16 catsACx Left Subd. ) )

) ) ) )

al

) ) ) ) ) ) ) ) al

et 1993 1991 2006 2007 1980 2006 et ( ( ( ( 2000 1990 1992 1999 1995 1994 1995 1991 Kaga Kaga Barth Barth Jones Barth ( ( ( ( ( ( ( ( and Di and Di Di and and Di Di and al al al al Author Subjects Location Di

Benison and Barth and Barth and Barth Kitzmiller Franowicz Franowicz Jones Hollenberg Hollenberg et et et et MacDonald Barth ( Barth (

16

99 J. Neural Eng. 14 (2017) 056004 X Wang et al ) motor & Continued ( # # # # # # movements movements movements distribution Percentage of Phonemotopy Motor cortical Motor cortical forelimb evoked forelimb evoked forelimb evoked forelimb evoked somatotopy with somatotopy somatotopy with somatotopy Limb-stimulation power distribution power Movement-related Movement-related somatotopy gamma band power band power gamma cortical somatotopy cortical somatotopy with forelimb evoked with forelimb evoked # # # # # High frq. no foc disc. no foc. disc focalization map Topographic no foc. disc. no foc. disc. no foc. disc. no foc. disc. high gamma high gamma Up to 80 Hz, Up to 170 Hz, Up to 180 Hz, Up to 180 Hz, Up to 566 Hz, Up to 100 Hz, Up to 150 Hz, show more focal show increase in ampl. . f # in vessel vessel Blood Results ## ## ## ## ## ## map maps peaks in Multiple 2 peaks in topographic P1-N1 ampl. ### ### ### ### ### test) Human for human for human for human for human application (cytotoxicity (cytotoxicity F F F F F human For F human For T Potentially T Potentially T Potentially & & & & T T T T # Systems, Systems, Systems, Germany) Germany) Germany) Reutlingen, Reutlingen, Reutlingen, Racine, WI) Racine, Commercial Commercial Commercial Commercial Commercial Commercial Instruments, (PMT Corp.) Tech Medical Tech Technologies) Technologies) (Multichannel (Multichannel (Multichannel (Tucker Davis Davis (Tucker Davis (Tucker Non-commercial F human For Non-commercial T Suitable Non-commercial T Non-commercial ) # # Pt Commercial(Ad PI PI PI grade grade PDMS Pt/Pt-Ir Pt/Pt-Ir Tinitrite Tinitrite Tinitrite Tinitrite silicone, silicone, Medical- Medical- silicone / Pt within Pt within Pt within within PI within PI PI kapton Au within

×

4

32 32

4 4 8

× × × Continued × × 8 (ref.)) 8 (ref.))

× × × # #

16/4

1 1 4 2 8 ( + +

4/4 ×

161 161 72 contacts within Ti 72 contacts 72 contacts 1 Electrode (64 (64 m ) 2 µ m µ m m m m m Table A1. µ µ µ µ µ m/700 4 mm 4 #

µ m medio- 1 mm 1 2 mm 32 contacts 1 mm 3 mm µ × 750 750 150 Spacing

m/400 µ 500 (4 medio-lateral, medio-lateral, (covered area)(covered Size Material Type Analysis lateral, 750 antero-posterior antero-posterior antero-posterior m 640 m 640 m m 640 m m/ m/ m µ µ µ µ µ µ µ µ Ǿ # # # # square 75 100 100 150 100 350 100 100 0.125 mm 0.5 mm 0.125 mm 0.5 mm # ECoG motor-related motor-related motor-related VEPs ECoG activity µ & & potentials potentials movement- movement- related SEPs ECoG, EMG related potentials and motor learning Stimulation mapping Stimulation mapping SSEPs SSEPs Task

& & & & chronic chronic chronic chronic Acute Acute Acute Acute (3 of 27)

& & epid. epid. Epid. Chronic stimulation- Electrical Epid. Chronic Movement-related Epid. Chronic with Decoding Epid. Acute brain Spontaneous Subd. Chronic Movement-related Subd. Acute Subd. Subd. # M1 M1 Right cMCx SMCx SMCx SMCx Left IFG Subd. Chronic Pronunciation-related contrlateral unilaterally contralaterally Rats Right Rats Right 1 pig Left OCCx Subd. Acute a a 5 rats Right 3 rats Right 2 male 16 rats Left cMCx Epid. 27 rats MCx Epid. human human patient human patient patients Monkeys SMCx Monkeys 1 female 1 female a a

) ) ) ) ) ) )

al al

al al

) ) ) )

et et et et 2008 2010 2011 2007 2008 2007 2008 (2009) ( ( ( ( ( ( ( 2008 2007 2009 2009 ( ( ( ( al al al al al al al al Author Subjects Location Slutzky Slutzky Slutzky Slutzky Slutzky Slutzky Blakely Blakely

Kim Kim Leuthardt Hosp Kitzmiller Kellis Kellis et et et et et et et et Molina-Luna Molina-Luna

17

100 J. Neural Eng. 14 (2017) 056004 X Wang et al ) seizure F) Continued & ( & # # # # # # # # # # # map (T components of sleep topographic maps Finger-movement- related somatotopy related somatotopy Visual somatotopy, somatotopy, Visual Decoding accuracy Decoding accuracy # # # # # # 120 Hz) – smaller foc, disc. High frq. no foc. disc focalization map Topographic no foc. disc. no foc. disc. no foc. disc. no foc. disc. no foc. disc. (60 Up to 25 Hz, Coherence in Up to 120 Hz, Up to 500 Hz, Up to 120 Hz, Up to 150 Hz, Up to 160 Hz, high-frq. band Up to 40 Hz, no . f # in vessel vessel Blood Results nger- ## ## ## ## ## ## f map maps peaks in Multiple in movement- movement- topographic related ampl. ### ### ### ### ### ### ### Human for human for human Potentially For human For For human For For humanFor 2 peaks For human For For human For application F F F F F F F F & & & & & & & & T T T T T T T T Racine, WI) Racine, Racine, WI) Racine, Racine, WI) Racine, Commercial Instruments, Instruments, Instruments, (PMT Corp.) Tech Medical Tech Tech Medical Tech Tech Medical Tech Commercial(Ad Commercial(Ad Non-commercial T Non-commercial T Non-commercial Non-commercial Non-commercial T Potentially Non-commercial Non-commercial T Non-commercial ) # # # PI PI Pt-Ir Commercial(Ad Au-Pt Au-Pt within within within PDMS/ PMMA Au with within PI within PI parylene C parylene C and silicon parylene C PEDOT:PSS PEDOT:PSS PEDOT:PSS PEDOT:PSS 2

2

5 4)

16 +

+ Continued

× ×

×

6 16/12 ( 8 (4 (32 contacts (ref.) (ground)) 37 contacts 32 contacts Au, 16 contacts 360 contacts Pt within 252 contacts within Pt 256 contacts Au, Pt, Electrode m 2 ) µ 2 m m m Table A1. µ µ µ 1 mm 6 mm 6 60 mm 60

– m/400 1 mm 1 2 mm 1 mm 32 contacts Au-Pt 3 mm 28 contacts Pt-Ir Non-commercial T 3 mm 28 contacts Pt-Ir Non-commercial T × × µ 500 Spacing

0.5 (6 35 (covered area)(covered Size Material Type Analysis 300 m m m m m m m µ µ µ µ µ µ µ Ǿ m square m square 60 m square 30 1 mm 1 mm 3 mm square square 40 µ µ µ 1.5 mm 4 mm 16 contacts 1.5 mm 4 mm 16 contacts 300 500 300 150 300 300 50 20 10

spontaneous spontaneous SEPs VEPs VEPs VEPs & control control control activity activity activity LFP-modulated Stimulus- ECoG, BM ECoG, BMI ECoG, BMI & µ µ µ epileptic activity spiking activity evoked LFPs electrographic seizures Task

& & AcuteVEPs, Sleep spindles, Acute chronic chronic Acute Acute Chronic Used for stimulationmm/ 1.5

& # Epid Chronic Movement-related epid. Epid. Chronic Movement-related Epid. Subd. Chronic Motor cortical Subd. Chronic Subd. Chronic brain Spontaneous Subd. Chronic Motor cortical Subd. Chronic Subd. PMd PMd ## # & & SMCx in rats; Left MCx Left MCx M1 M1 anterior to anterior to Left SMCx Subd. Chronic Movement-related arm area of Right hemi. of left hemi. central sulcus central sulcus Rat Cats VCx Rats Right SCx Subd. Acute a a a and 2 12 rats VCx Epid. 1 male 13 rats patient human human patient human patient human 10 catsVCx Left patients patients Monkeys Right 1 female 1 female 9 human 1 monkey parts Large a 3 monkeys Proximal 3 monkeys

) ) ) ) )

al al al al al

al

al

al

al ) ) ) ) ) ) ) ) )

et et et et et et et et et 2011 2009 2011 2009 2015 ( ( ( ( ( 2011 2013 2009 2011 2010 2009 2010 2016 2015 ( ( ( ( ( ( ( ( ( al al al al al Author Subjects Location Rubehn

Kim Toda Toda Vinjamuri Vinjamuri Wang Wang Wilks Wilks Rouse Rouse Thongpang et et et et et Blanco Hotson Viventi Viventi Khodagholy Khodagholy

18

101 J. Neural Eng. 14 (2017) 056004 X Wang et al # # map epicortical;

polydimethyl polydimethyl

=

Tonotopic map Tonotopic =

combined Nose-stimulation- diff. topographical diff. related somatotopy discussion;

High-gamma power power High-gamma

= =

motor cortex; motor cortex;

=

High frq. coherence coherence frequencies rapidly with with rapidly Higher frq., focalization map Topographic no foc. disc. no foc. disc. no foc. disc. declined more more declined Up to 300 Hz, Up to 150 Hz, Up to 800 Hz, micro-electrocorticography; micro-electrocorticography; more focalized

distance in higher higher in distance =

polyimide; PDMS polyimide;

. somatosensory cortex; somatosensory cortex; =

f focalization; disc.

# = in

vessel vessel Blood vessels Results = ECoG on blood blood on

premotor dorsal area; Epic. µ Independent Independent

=

occipital cortex; MCx occipital cortex;

iridium; PI iridium;

## ## = maps

=

peaks in multiple Multiple 2 peaks in in peaks 2 2 or more power diff. power peaks, both topographic power maps power in ampl. and topographical topographical auditory evoked potentials; ASEPs potentials; auditory evoked

amplitude; foc. =

=

### # For For Human titanium; Ir titanium; humans humans

electrocorticography; electrocorticography; visual cortex; PMd visual cortex; For human For application

=

= =

parieto-temporal areas; SCx

F F F F = F For Department of Microsystems Engineering, University of Department of Microsystems Engineering, University & & & &

T T T T =

gold; Ti gold;

=

parieto-ventral areas; OCCx parieto-ventral

= interictal spikes; ampl. interictal spikes;

=

(Integra (Integra (IMTEK) Commercial Commercial parietal cortex; VCx parietal cortex; (Integra Inc.) (Integra LifeSciences,

Plainsboro, NJ) Non-commercial = platinum; Au platinum; auditory cortex; PT auditory cortex;

= =

) # Pt Commercial Au Non-commercial somatosensory evoked potentials; AEPs potentials; somatosensory evoked

grade silicon silicone visual evoked potentials; ECoG visual evoked Medical- Pt-tipped

=

rubber, Pt/Ir rubber, =

chlorine; Pt chlorine;

8 8 8 8

Continued 10 (96 = × × × ×

used) frequency; ACx frequency; 8 8 8 4 × (

= 10 Electrode

initial negative responses; IIS initial negative 2

) = silver; Cl silver; 2 primary motor cortex; PCx primary motor cortex;

secondary somatosensory areas; PV

= = = Table A1.

3.25 mm 3.25 4 mm 4

uence; frq. × × 1.5 mm 1.5 Spacing f

(4 in (covered area)(covered Size Material Type Analysis

1.68 mm/3.5 mm 1.68 3.25 =

. f m m µ Ǿ µ 4 2 mm 4 mm 1.5 mm 4 mm 200 eld potentials; Ag potentials; eld 1.81 mm 0.87 mm/ 0.87 f local

ECoG, ECoG, inferior frontal gyrus; M1 initial positive responses; N1 initial positive µ

=

= =

middle-latency auditory evoked potentials; (S)SEP auditory evoked middle-latency

poly(3,4-ethylenedioxythiophene) doped with poly(styrene sulfonate; IMTEK poly(3,4-ethylenedioxythiophene)

SEPs SEPs = AEPs control

primary somatosensory areas; S2 =

not referred to in the respective study. not referred to in the respective ECoG, BMI BMI control

µ = middle-latency somatosensory evoked potentials; VEPs potentials; somatosensory evoked middle-latency

=

= ed in the respective study. in study. ed in the respective #

Task f Movement-related Movement-related diameter; Chronic

epidural; MAEPs frequency domain; P1 frequency

hemisphere; S1 =

= =

sensorimotor cortex; IFG sensorimotor cortex;

= Ǿ

&

=

brain–machine-interface; LFPs brain–machine-interface; subd. Subd. Acute

Epid. =

right difference; difference;

& M1 M1 Subd. Chronic Movement-related IFG = Right subdural; Epid.

time domain; F

temporal lobe = =

poly(methylmethacrylate); PEDOT:PSS poly(methylmethacrylate);

=

3 rats ACx Epid. Chronic patients patients patients 5 human 6 human 5 human 4 minipigs Left SCx Epid. Acute

)

caudolateral neocortex; hemi. caudolateral neocortex; caudal motor cortex; SMCx caudal motor cortex;

electromyography; BMI electromyography; al al epipial; Subd. corporation; diff. al

) ) )

= = et et

= et = 2017 =

( 2017 2016 2016 study ( ( ( al Author Subjects Location Current

Trumpis Trumpis Flint Wang Wang Kellis Kellis et Mean number of the subjects involved in is not speci Mean number of the subjects involved EMG siloxane; PMMA T Germany; Freiburg, corp. a Epip. potentials; MSEPs auditory and somatosensory evoked clNCx cMCx

19

102 J. Neural Eng. 14 (2017) 056004 X Wang et al References Leuthardt E C 2014 Characterization of the effects of the human dura on macro- and micro-electrocorticographic recordings J. Neural Eng. 11 016006 Antognini J F, Buonocore M H, Disbrow E A and Carstens E 1997 Calancie B, Harris W, Broton J G, Alexeeva N and Green B A 1998 Isofurane anesthesia blunts cerebral responses to noxious and ‘Threshold-level’ multipulse transcranial electrical stimulation innocuous stimuli: a fMRI study Life Sci. 61 PL 349–54 of motor cortex for intraoperative monitoring of spinal motor Baker S N, Gabriel C and Lemon R N 2003 EEG oscillations at 600 tracts: description of method and comparison to somatosensory Hz are macroscopic markers for cortical spike bursts J. Physiol. evoked potential monitoring J. Neurosurg. 88 457–70 550 529–34 Chapin J K, Waterhouse B D and Woodward D J 1981 Differences Ball T, Schulze-Bonhage A, Aertsen A and Mehring C 2009 in cutaneous sensory response properties of single Differential representation of arm movement direction in somatosensory cortical neurons in awake and halothane relation to cortical anatomy and function J. Neural Eng. anesthetized rats Brain Res. Bull. 6 63–70 6 016006 Chatillon C E, Zelmann R, Bortel A, Avoli M and Gotman J Barth D S 2003 Submillisecond synchronization of fast electrical 2011 Contact size does not affect high frequency oscillation oscillations in neocortex J. Neurosci. 23 2502–10 detection in intracerebral EEG recordings in a rat epilepsy Barth D S and Di S 1990 Three-dimensional analysis of auditory- model Clin. Neurophysiol. 122 1701–5 evoked potentials in rat neocortex J. Neurophysiol. 64 1527–36 Chaumon M, Schwartz D and Tallon-Baudry C 2009 Unconscious Barth D S and Di S 1991 The functional anatomy of middle latency learning versus visual perception: dissociable roles for gamma auditory evoked potentials Brain Res. 565 109–15 oscillations revealed in MEG J. Cogn. Neurosci. 21 2287–99 Barth D S and Di S 1992 Topographical analysis of epileptiform Craner S L and Ray R H 1991 Somatosensory cortex of the neonatal potentials in rat somatosensory cortex: the interictal to ictal pig: I. Topographic organization of the primary somatosensory transition Brain Res. 591 33–43 cortex (SI) J. Comp. Neurol. 306 24–38 Benison A M, Rector D M and Barth D S 2007 Hemispheric Crone C and Olesen S P 1982 Electrical resistance of brain mapping of secondary somatosensory cortex in the rat microvascular endothelium Brain Res. 241 49–55 J. Neurophysiol. 97 200–7 Crone N E 2000 Functional mapping with ECoG spectral analysis Benjamini Y and Yekutieli D 2001 The control of the false Adv. Neurol. 84 343–51 discovery rate in multiple testing under dependency Ann. Stat. Crone N E, Boatman D, Gordon B and Hao L 2001a Induced 29 1165–88 electrocorticographic gamma activity during auditory Bithal P 2014 Anaesthetic considerations for evoked potentials perception Brazier Award-winning article Clin. Neurophysiol. monitoring J. Neuroanaesth. Crit. Care 1 2–12 112 565–82 Blakely T, Miller K J, Rao R P, Holmes M D and Ojemann J G 2008 Crone N E, Hao L, Hart J Jr, Boatman D, Lesser R P, Irizarry R and Localization and classifcation of phonemes using high spatial Gordon B 2001b Electrocorticographic gamma activity during resolution electrocorticography (ECoG) grids Conf. Proc. IEEE word production in spoken and sign language Neurology Engineering in Medicine and Biology Society pp 4964–7 57 2045–53 Blanco J A, Stead M, Krieger A, Viventi J, Marsh W R, Lee K H, Crone N E, Miglioretti D L, Gordon B and Lesser R P 1998 Worrell G A and Litt B 2010 Unsupervised classifcation of Functional mapping of human sensorimotor cortex with high-frequency oscillations in human neocortical epilepsy and electrocorticographic spectral analysis. II. Event-related control patients J. Neurophysiol. 104 2900–12 synchronization in the gamma band Brain 121 2301–15 Bleichner M G, Vansteensel M J, Huiskamp G M, Hermes D, Crone N E, Sinai A and Korzeniewska A 2006 High-frequency Aarnoutse E J, Ferrier C H and Ramsey N F 2011 The effects gamma oscillations and human brain mapping with of blood vessels on electrocorticography J. Neural Eng. electrocorticography Prog. Brain Res. 159 275–95 8 044002 Curio G 2000 Linking 600 Hz ‘spikelike’ EEG/MEG wavelets Bonhomme V, Fiset P, Meuret P, Backman S, Plourde G, Paus T, (‘sigma-bursts’) to cellular substrates: concepts and caveats Bushnell M C and Evans A C 2001 Propofol anesthesia J. Clin. Neurophysiol. 17 377–96 and cerebral blood fow changes elicited by vibrotactile Daneman R 2012 The blood-brain barrier in health and disease stimulation: a positron emission tomography study Ann. Neurol. 72 648–72 J. Neurophysiol. 85 1299–308 Derix J, Iljina O, Schulze-Bonhage A, Aertsen A and Ball T Boveroux P et al 2010 Breakdown of within- and between- 2012 ‘Doctor’ or ‘darling’? Decoding the communication network resting state functional magnetic resonance imaging partner from ECoG of the anterior temporal lobe during non- connectivity during propofol-induced loss of consciousness experimental, real-life social interaction Front. Hum. Neurosci. Anesthesiology 113 1038–53 6 251 Bragin A, Engel J Jr, Wilson C L, Fried I and Buzsaki G 1999b Derix J, Iljina O, Weiske J, Schulze-Bonhage A, Aertsen A and High-frequency oscillations in human brain Hippocampus Ball T 2014 From speech to thought: the neuronal basis of 9 137–42 cognitive units in non-experimental, real-life communication Bragin A, Engel J Jr, Wilson C L, Fried I and Mathern G W 1999a investigated using ECoG Front. Hum. Neurosci. 8 383 Hippocampal and entorhinal cortex high-frequency oscillations Di S and Barth D S 1991 Topographic analysis of feld potentials in (100–500 Hz) in human epileptic brain and in kainic acid— rat vibrissa/barrel cortex Brain Res. 546 106–12 treated rats with chronic seizures Epilepsia 40 127–37 Di S and Barth D S 1993 Binaural versus monaural auditory evoked Bragin A, Mody I, Wilson C L and Engel J Jr 2002 Local potentials in rat neocortex Brain Res. 630 303–14 generation of fast ripples in epileptic brain J. Neurosci. Di S, Brett B and Barth D S 1994 Polysensory evoked potentials 22 2012–21 in rat parietotemporal cortex: combined auditory and Bragin A, Wilson C L and Engel J Jr 2000 Chronic epileptogenesis somatosensory responses Brain Res. 642 267–80 requires development of a network of pathologically Dueck M H et al 2005 Propofol attenuates responses of the auditory interconnected neuron clusters: a hypothesis Epilepsia cortex to acoustic stimulation in a dose-dependent manner: a 41 S144–52 FMRI study Acta. Anaesthesiol. Scand. 49 784–91 Bundy D T, Zellmer E, Gaona C M, Sharma M, Szrama N, Engel J Jr 1996 Surgery for seizures New Engl. J. Med. Hacker C, Freudenburg Z V, Daitch A, Moran D W and 334 647–53

20

103 J. Neural Eng. 14 (2017) 056004 X Wang et al Engel J Jr, Pedley T A, Aicardi J, Moshé S L, Dichter M A, Jones M S and Barth D S 1999 Spatiotemporal organization of fast Trimble M R and Perucca E 2007 Epilepsy: A Comprehensive (<200 Hz) electrical oscillations in rat vibrissa/barrel cortex Textbook 2nd edn (Philadelphia, PA: Lippincott Williams & J. Neurophysiol. 82 1599–609 Wilkins) Jones M S, MacDonald K D, Choi B, Dudek F E and Barth D S Fiederer L et al 2016 The role of blood vessels in high-resolution 2000 Intracellular correlates of fast (<200 Hz) electrical volume conductor head modeling of EEG Neuroimage oscillations in rat somatosensory cortex J. Neurophysiol. 128 193–208 84 1505–18 Flint R D, Rosenow J M, Tate M C and Slutzky M W 2017 Kaga K, Hink R F, Shinoda Y and Suzuki J 1980 Evidence for a Continuous decoding of human grasp kinematics using primary cortical origin of a middle latency auditory evoked epidural and subdural signals J. Neural Eng. 14 016005 potential in cats Electroencephalogr. Clin. Neurophysiol. Franowicz M N and Barth D S 1995 Comparison of evoked 50 254–66 potentials and high-frequency (gamma-band) oscillating Kellis S, Miller K, Thomson K, Brown R, House P and potentials in rat auditory cortex J. Neurophysiol. 74 96–112 Greger B 2010 Decoding spoken words using local feld Freeman W J, Rogers L J, Holmes M D and Silbergeld D L potentials recorded from the cortical surface J. Neural Eng. 2000 Spatial spectral analysis of human electrocorticograms 7 056007 including the alpha and gamma bands J. Neurosci. Methods Kellis S, Sorensen L, Darvas F, Sayres C, O’Neill K, Brown R B, 95 111–21 House P, Ojemann J and Greger B 2016 Multi-scale analysis Fukushima M, Saunders R C, Leopold D A, Mishkin M and of neural activity in humans: Implications for micro-scale Averbeck B B 2012 Spontaneous high-gamma band activity electrocorticography Clin. Neurophysiol. 127 591–601 refects functional organization of auditory cortex in the awake Kellis S S, House P A, Thomson K E, Brown R and Greger B macaque Neuron 74 899–910 2009 Human neocortical electrical activity recorded Gierthmuehlen M, Ball T, Henle C, Wang X, Rickert J, Raab M, on nonpenetrating microwire arrays: applicability for Freiman T, Stieglitz T and Kaminsky J 2011 Evaluation neuroprostheses Neurosurg. Focus. 27 E9 of µECoG electrode arrays in the minipig: experimental Kerssens C, Hamann S and Peltier S, Hu X P, Byas-Smith M G procedure and neurosurgical approach J. Neurosci. Methods and Sebel P S 2005 Attenuated brain response to auditory word 202 77–86 stimulation with sevofurane: a functional magnetic resonance Gierthmuehlen M et al 2014 Mapping of sheep sensory cortex imaging study in humans Anesthesiology 103 11–9 with a novel microelectrocorticography grid J. Comp. Neurol. Khodagholy D, Doublet T, Gurfnkel M, Quilichini P, Ismailova E, 522 3590–608 Leleux P, Herve T, Sanaur S, Bernard C and Malliaras G G Hamer H M et al 2002 Complications of invasive video-EEG 2011 Highly conformable conducting polymer electrodes for monitoring with subdural grid electrodes Neurology in vivo recordings Adv. Mater. 23 H268–72 58 97–103 Khodagholy D, Gelinas J N, Thesen T, Doyle W, Devinsky O, Henle C, Meier W, Schuettler M, Boretius T and Stieglitz T Malliaras G G and Buzsaki G 2015 NeuroGrid: recording 2009 Electrical characterization of platinum, stainless action potentials from the surface of the brain Nat. Neurosci. steel and platinum/iridium as electrode materials for a new 18 310–5 neural interface World Congress on Medical Physics and Kim D H et al 2010 Dissolvable flms of silk fbroin for Biomedical Engineering ed O Dössel and W C Schlegel ultrathin conformal bio-integrated electronics Nat. Mater. (IFMBE Proc. vol 25/9) (Berlin: Springer) http://doi. 9 511–7 org/10.1007/978-3-642-03889-1_27 Kim J, Richner T J, Thongpang S, Sillay K A, Niemann D B, Henle C, Schuettler M, Rickert J and Stieglitz T 2012 Towards Ahmed A S, Krugner-Higby L A and Williams J C 2009 electrocorticographic electrodes for chronic use in BCI Flexible thin flm electrode arrays for minimally-invasive applications Towards Practical BCIs ed A Brendan et al neurological monitoring Conf. Proc. IEEE Engineering in (Berlin: Springer) pp 85–103 Medicine and Biology Society pp 5506–9 Hollenberg B A, Richards C D, Richards R, Bahr D F and Kim J, Wilson J A and Williams J C 2007 A cortical recording Rector D M 2006 A MEMS fabricated fexible electrode array platform utilizing µECoG electrode arrays Conf. Proc. for recording surface feld potentials J. Neurosci. Methods IEEE Engineering in Medicine and Biology Society 153 147–53 pp 5353–7 Hosp J A, Molina-Luna K, Hertler B, Atiemo C O, Stett A and Kitzmiller J, Beversdorf D and Hansford D 2006 Fabrication and Luft A R 2008 Thin-flm epidural microelectrode arrays for testing of microelectrodes for small-feld cortical surface somatosensory and motor cortex mapping in rat J. Neurosci. recordings Biomed. Microdevices 8 81–5 Methods 172 255–62 Kitzmiller J P, Hansford D J, Fortin L D, Obrietan K H, Hotson A G and Schneider D S 2015 Drosophila melanogaster Bergdall V K and Beversdorf D Q 2007 Micro-feld evoked Natural Variation Affects Growth Dynamics of Infecting potentials recorded from the porcine sub-dural cortical Listeria monocytogenes G3 5 2593–600 surface utilizing a microelectrode array J. Neurosci. Methods Hudetz A G, Vizuete J A and Imas O A 2009 Desfurane selectively 162 155–61 suppresses long-latency cortical neuronal response to fash in Kumar A, Bhattacharya A and Makhija N 2000 Evoked potential the rat Anesthesiology 111 231–9 monitoring in anaesthesia and analgesia Anaesthesia Huotari A M, Koskinen M, Suominen K, Alahuhta S, Remes R, 55 225–41 Hartikainen K M and Jäntti V 2004 Evoked EEG patterns Leuchter A F, Spar J E, Walter D O and Weiner H 1987 during burst suppression with propofol Br. J. Anaesth. 92 18–24 Electroencephalographic spectra and coherence in the Insanally M, Trumpis M, Wang C, Chiang C H, Woods V, Palopoli- diagnosis of Alzheimer’s-type and multi-infarct dementia. A Trojani K, Bossi S, Froemke R C and Viventi J 2016 A low- pilot study Arch. Gen. Psychiatry 44 993–8 cost, multiplexed µECoG system for high-density recordings in Leuchter A F and Walter D O 1989 Diagnosis and assessment of freely moving rodents J. Neural Eng. 13 026030 dementia using functional brain imaging Int. Psychogeriatr. Jelsing J, Hay-Schmidt A, Dyrby T, Hemmingsen R, Uylings H B 1 63–72 and Pakkenberg B 2006 The prefrontal cortex in the Göttingen MacDonald K D and Barth D S 1995 High frequency (gamma- minipig brain defned by neural projection criteria and band) oscillating potentials in rat somatosensory and auditory cytoarchitecture Brain Res. Bull. 70 322–36 cortex Brain Res. 694 1–12

21

104 J. Neural Eng. 14 (2017) 056004 X Wang et al

Miller K J, Sorensen L B, Ojemann J G and den Nijs M 2009 Siegel M and König P 2003 A functional gamma-band defned Power-law scaling in the brain surface electric potential by stimulus-dependent synchronization in area 18 of awake PLoS Comput. Biol. 5 e1000609 behaving cats J. Neurosci. 23 4251–60 Miller R D 2005 Miller’s Anesthesia 7th edn (New York: Elsevier) Sloan T B 1998 Anesthetic effects on electrophysiologic recordings Molina-Luna K, Buitrago M M, Hertler B, Schubring M, Haiss F, J. Clin. Neurophysiol. 15 217–26 Nisch W, Schulz J B and Luft A R 2007 Cortical stimulation Slutzky M W, Jordan L R, Krieg T, Chen M, Mogul D J and mapping using epidurally implanted thin-flm microelectrode Miller L E 2010 Optimal spacing of surface electrode arrays arrays J. Neurosci. Methods. 161 118–25 for brain-machine interface applications J. Neural Eng. Molina-Luna K, Hertler B, Buitrago M M and Luft A R 2008 Motor 7 26004 learning transiently changes cortical somatotopy Neuroimage Slutzky M W, Jordan L R, Lindberg E W, Lindsay K E and Miller L E 40 1748–54 2011 Decoding the rat forelimb movement direction from Morrell M J and RNS System in Epilepsy Study Group 2011 epidural and intracortical feld potentials J. Neural Eng. Responsive cortical stimulation for the treatment of medically 8 036013 intractable partial epilepsy Neurology 77 1295–304 Slutzky M W, Jordan L R and Miller L E 2008 Optimal spatial Mhuircheartaigh R N, Warnaby C, Rogers R, Jbabdi S and Tracey I resolution of epidural and subdural electrode arrays for brain- 2013 Slow-wave activity saturation and thalamocortical machine interface applications Conf. Proc. IEEE Engineering isolation during propofol anesthesia in humans Sci. Transl. in Medicine and Biology Society pp 3771–4 Med. 5 208ra148 Smith S M 2002 Fast robust automated brain extraction Hum. Brain Ojemann G, Ojemann J, Lettich E and Berger M 1989 Cortical Mapp. 17 143–55 language localization in left, dominant hemisphere Smith S M et al 2004 Advances in functional and structural MR J. Neurosurg. 71 316–26 image analysis and implementation as FSL Neuroimage Ojemann G A, Ramsey N F and Ojemann J 2013 Relation between 23 S208–19 functional magnetic resonance imaging (fMRI) and single Staba R J, Wilson C L, Bragin A, Fried I and Engel J Jr 2002 neuron, local feld potential (LFP) and electrocorticography Quantitative analysis of high-frequency oscillations (80–500 (ECoG) activity in human cortex Front. Hum. Neurosci. 7 34 Hz) recorded in human epileptic hippocampus and entorhinal Oke O O, Magony A, Anver H, Ward P D, Jiruska P, Jefferys J G cortex J. Neurophysiol. 88 1743–52 and Vreugdenhil M 2010 High-frequency gamma oscillations Stead M, Bower M, Brinkmann B H, Lee K, Marsh W R, coexist with low-frequency gamma oscillations in the rat visual Meyer F B, Litt B, Van Gompel J and Worrell G A 2010 cortex in vitro Eur. J. Neurosci. 31 1435–45 Microseizures and the spatiotemporal scales of human partial Philbin D M, Rosow C E, Schneider R C, Koski G and epilepsy Brain 133 2789–97 D’Ambra M N 1990 Fentanyl and sufentanil anesthesia Stieglitz T, Rubehn B, Henle C, Kisban S, Herwik S, Ruther P and revisited: how much is enough? Anesthesiology 73 5–11 Schuettler M 2009 Brain-computer interfaces: an overview Ramani R, Qiu M and Constable R T 2007 Sevofurane 0.25 of the hardware to record neural signals from the cortex MAC preferentially affects higher order association areas: a Prog. Brain Res. 175 297–315 functional magnetic resonance imaging study in volunteers Streisand J B, Bailey P L and LeMaire L 1993 Fentanyl-induced Anesth. Analg. 105 648–55 rigidity and unconsciousness in human volunteers. Incidence, Rembado I et al 2017 Independent component decomposition of duration, and plasma concentrations Anesthesiology 78 629–34 human somatosensory evoked potentials recorded by micro- Tallon-Baudry C, Bertrand O, Henaff M A, Isnard J and Fischer C electrocorticography Int. J. Neural Syst. 27 1650052 2005 Attention modulates gamma-band oscillations differently Robson J G 1967 The effects of anesthetic drugs on cortical units in the human lateral occipital cortex and fusiform gyrus Anesthesiology 28 144–54 Cereb. Cortex 15 654–62 Rouse A, Williams J, Wheeler J J and Moran D 2016 Spatial Toda H, Suzuki T, Sawahata H, Majima K, Kamitani Y and co-adaptation of cortical control columns in a micro-ECoG Hasegawa I 2011 Simultaneous recording of ECoG and brain-computer interface J. Neural Eng. 13 056018 intracortical neuronal activity using a fexible multichannel Rouse A G, Williams J J, Wheeler J J and Moran D W 2013 electrode-mesh in visual cortex Neuroimage 54 203–12 Cortical adaptation to a chronic micro-electrocorticographic Toro C, Cox C, Friehs G, Ojakangas C, Maxwell R, Gates J R, brain computer interface J. Neurosci. 33 1326–30 Gumnit R J and Ebner T J 1994 8–12 Hz rhythmic oscillations Rubehn B, Bosman C, Oostenveld R, Fries P and Stieglitz T 2009 A in human motor cortex during two-dimensional arm MEMS-based fexible multichannel ECoG-electrode array movements: evidence for representation of kinematic parameters J. Neural Eng. 6 036003 Electroencephalogr. Clin. Neurophysiol. 93 390–403 Schalk G, Mcfarl D J, Hinterberger T, Birbaumer N and Thongpang S et al 2011 A micro-electrocorticography platform and Wolpaw J R 2004 BCI2000: A general-purpose brain-computer deployment strategies for chronic BCI applications Clin. EEG interface (BCI) system IEEE Trans. Biomed. Eng. Neurosci. 42 259–65 51 1034–43 Trumpis M, Insanally M, Zou J, Elsharif A, Ghomashchi A, Sertac Schuettler M, Henle C, Ordonez J S Meier W, Guenther T and Artan N, Froemke R C and Viventi J 2017 A low-cost, scalable, Stieglitz T 2008 Interconnection technologies for laser- current-sensing digital headstage for high channel count patterned electrode arrays Conf. Proc. IEEE Engineering in µECoG J. Neural Eng. 14 026009 Medicine and Biology Society pp 3212–5 Uhrig L, Dehaene S and Jarraya B 2014 Cerebral mechanisms of Schuettler M, Pfau D, Ordonez J S, Henle C, Woias P and general anesthesia Ann. Fr. Anesth. Reanim. 33 72–82 Stieglitz T 2009 Stretchable tracks for laser-machined neural Vinjamuri R et al 2009 A fuzzy logic model for hand posture electrode arrays Conf. Proc. IEEE Engineering in Medicine control using human cortical activity recorded by micro-ECog and Biology Society pp 1612–5 electrodes Conf. Proc. IEEE Engineering in Medicine and Schuettler M, Stiess S, King B V and Suaning G J 2005 Fabrication Biology Society pp 4339–42 of implantable microelectrode arrays by laser cutting of Viventi J et al 2011 Flexible, foldable, actively multiplexed, high- silicone rubber and platinum foil J. Neural Eng. 2 S121–8 density electrode array for mapping brain activity in vivo Schwartz D M, Schwartz J A, Pratt R E Jr, Wierzbowski L R and Nat. Neurosci. 14 1599–605 Sestokas A K 1997 Infuence of nitrous oxide on posterior Wang P T et al 2016 Comparison of decoding resolution of standard tibial nerve cortical somatosensory evoked potentials J. Spinal and high-density electrocorticogram electrodes J. Neural Eng. Disord. 10 80–6 13 026016

22

105 J. Neural Eng. 14 (2017) 056004 X Wang et al Wang W et al 2009 Human motor cortical activity recorded with Woolrich M W, Jbabdi S, Patenaude B, Chappell M, Makni S, Micro-ECoG electrodes, during individual fnger movements Behrens T, Beckmann C, Jenkinson M and Smith S M 2009 Conf. Proc. IEEE Engineering in Medicine and Biology Society Bayesian analysis of neuroimaging data in FSL Neuroimage pp 586–9 45 S173–86 Watts J and Thomson A M 2005 Excitatory and inhibitory connections Worrell G A, Gardner A B, Stead S M, Hu S, Goerss S, show selectivity in the neocortex J. Physiol. 562 89–97 Cascino G J, Meyer F B, Marsh R and Litt B 2008 High- Wilks S J, Koivuniemi A S, Thongpang S, Williams J C and frequency oscillations in human temporal lobe: simultaneous Otto K J 2009 Evaluation of micro-electrocorticographic microwire and clinical macroelectrode recordings Brain electrodes for electrostimulation Conf. Proc. IEEE Engineering 131 928–37 in Medicine and Biology Society pp 5510–3 Zanos S 2009 Neural correlates of high-frequency intracortical and Wiggins G C, Elisevich K and Smith B J 1999 Morbidity and infection epicortical feld potentials J. Neurosci. 29 3673–5 in combined subdural grid and strip electrode investigation for Zhang Y, Brady M and Smith S 2001 Segmentation of brain MR intractable epilepsy Epilepsy Res. 37 73–80 images through a hidden Markov random feld model and Wong C H, Birkett J, Byth K, Dexter M, Somerville E, Gill D, the expectation-maximization algorithm IEEE Trans. Med. Chaseling R, Fearnside M and Bleasel A 2009 Risk factors Imaging 20 45–57 for complications during intracranial electrode recording in Zhang Y, Li Z, Dong H and Yu T 2014 Effects of general anesthesia presurgical evaluation of drug resistant partial epilepsy with propofol on thalamocortical sensory processing in rats Acta. Neurochir. 151 37–50 J. Pharmacol. Sci. 126 370–81

23

106 To be submitted to NeuroImage

IX. CORTICAL-DEPTH AND FREQUENCY-BAND DEPENDENCY OF THE SPATIAL EXTENT OF THE GENERATORS UNDERLYING µECOG RECORDINGS

Xi Wang1,2,3,4, Lukas D.J.Fiederer1,2,4,5, Christian Henle6, Irina Mader7, Jan Kaminsky8,*, Mortimer Gierthmuehlen2,4,Thomas Stieglitz3,4, Tonio Ball1,2,4,**

1 Translational Neurotechnology Lab, Epilepsy Centre, Medical Center – University of Freiburg, Faculty of Medicine, University of Freiburg, Breisacher Str. 64, 79106 Freiburg, Germany 2 Department of Neurosurgery, Medical Center – University of Freiburg, Faculty of Medicine, University of Freiburg, Breisacher Str. 64, 79106 Freiburg, Germany 3 Laboratory for Biomedical Micro-technology, Department of Microsystem Engineering (IMTEK), University of Freiburg, Georges-Koehler-Allee 103, 79110 Freiburg, Germany 4 BrainLinks-BrainTools Cluster of Excellence, University of Freiburg, Georges-Koehler- Allee 80, 79110 Freiburg, Germany 5 Department of Neurobiology and Biophysics, Faculty of Biology, University of Freiburg, Schaenzlestr. 1, 79104 Freiburg, Germany 6 CorTec GmbH, Georges-Koehler-Allee 010, 79110 Freiburg, Germany 7 Department of Neuroradiology, Medical Center – University of Freiburg, Faculty of Medicine, University of Freiburg, Breisacher Str. 64, 79106 Freiburg, Germany 8 Department of Neurosurgery, St. GertraudenKrankenhaus, ParetzerStraße 12, 10713 Berlin, Germany

* this author’s current address is different from the address where the work was carried out ** Corresponding author: PD Tonio Ball, AG Ball, Engelbergerstr.21 3.0 EG, 79106 Freiburg, [email protected], phone: +4976127093160

107 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

ABSTRACT Objective. Over the last two decades, studies of the volume conduction of neuronal population activity have mainly concentrated on conventional scalp EEG and rarely on intracranial EEG (iEEG). However, iEEG maintains the high-temporal resolution property and has a good balance between invasive risk and spatial resolution, especially for micro- electrocorticographical (μECoG) recordings as used in this study. This paper for the first time utilizes finite-element-method (FEM) to solve the forward problem based on μECoG recordings. Comparison of experimental and simulated data of somatosensory evoked potentials in different frequency range was investigated in the present study. Approach. Experimental data were obtained with custom μECoG electrode arrays applied over the minipigs’ acutely prepared somatosensory cortex while electrically stimulating their snout. Volume conductor models were constructed base on T1-weighted MRI data. The spatial resolution of the cortical mesh in the direct vicinity of the µECoG electrode array was refined in order to improve the accuracy of the FEM simulations. More than 300,000 dipole sources were simulated in the refined volume. Main results. Through quantitative comparison of the experimental and simulated signal’s spatial distribution, strength and profile, we create a better understanding of the neuronal current sources underlying the μECoG signal. We extend previous results on the spatial reach of LFP and confirm that the size of LFP generators increases with cortical depth. Furthermore, based on the LFP’s spatial reach we suggest that the optimal subdural brain-machine-interface (BMI) spatial resolution should be smaller than 2 mm, at least for the minipig animal model. Our results also indicate that lower-frequency activity may originate in deeper cortical layer and higher-frequency activity in more superficial cortical layers. Significance. The present study is the first to investigate the neuronal signal sources of μECoG recordings using volume conduction. Our observations on the spatial reach of LFP and the depth segregation of frequency generators strengthen the fundamental knowledge about the neuronal activity. Based on this knowledge, applications like high-resolution brain mapping, advanced epilepsy diagnostics and BMI can be further advanced. Keywords: FEM, µECoG, neuronal current source, Göttingen minipig

108 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings 1. INTRODUCTION Electrophysiological recording of the brain activity, such as electroencephalography (EEG), electrocorticography (ECoG), and a new branch becoming popular lately, the micro-scaled ECoG (µECoG) or more invasively depth recording in deeper brain structures for the local field potential (LFP), is an important and prominent method both in clinical practice and neuroscience research. Signals recorded through those methods are based on the electric currents inside the activated cortex area, so-called electrical source, which can be reconstructed though computational simulation with proper mathematical volume model. Reconstruction of the electric filed from source to electrode is termed as source localization, which helps the neuroscientists to learn more how the information are produced and propagated from source to electrode (Nunez 1981) and in result to explore the electrophysiological basis of neural coding, communication and information processing at different levels of neural organization. It is well-known that ‘forward problem’ is one of the key contents in source localization (Hallez et al 2007) which computes the signal that can be recorded at the electrode site due to a certain single or multiple source activation. For the forward problem, a proper head model plays an important role for the consequent results, which should consist of the specific head tissue with different electrical and anisotropic properties depending on the goal and method for each individual study (Wolters et al 2004). Numerical approach has been tested such as concentric shell models (Nunez and Srinivasan 2006), the boundary element methods (BEM, Hamalainen and Sarvas 1989, Fuchs et al 2002) and the finite element methods (FEM, Wolters 2007a). However, in realistic, due to the volume conduct effect, electrical activities distribute from their sources through inhomogeneous head tissues with various electrical properties (anisotropic conductivities). Thus only FEM allows a more realistic representation of the complicated head volume conductor (Wolters et al 2004, Wolters 2007a, Wagner et al 2016) and thus becomes a popular method combining simulation to study the relationship between electric sources in the brain and the resulting electrical potentials at the electrode, i.e., to solve the ‘forward problem’. With electrophysiological theory as concluded in previous review (Buzsáki et al 2012), it is well-understood that the original source of all kinds of extracellular electrophysiological signal (EEG, (µ)ECoG, LFP) can be explained as the summation of postsynaptic potentials (PSPs) from many simultaneously activated neurons measured at a given extracellular position. Depend on the extracellular location to measure the signal, different terms are defined, i.e., electrode nearby the activated neurons then we will get the LFP; electrode lying on the cortex (epidural or subdural) /scalp surface, the EEG / (µ)ECoG will be obtained. Distance between source and recording site decide the amount of neurons involved in and the spatial extent where the signal propagate in to the specified superimposing measuring location. However, the synaptic communication from presynaptic neuron to postsynaptic neuron is not only excitatory PSP which enhance the summation at the given location, but also the inhibitory PSP which diminish the final summation. Thus one cannot just say the more neuron population included in the higher potential will be superimposed at the given location. At least, it is true within a certain distance away from the recording site for the LFP and then the LFP keeps consistent out of this distance which was defined as ‘spatial reach’ to describe the size of the neuron population can be measured in Lindén’s study (Lindén et al 2011). Previous study reported that the spatial reach for LFP was a radius of 0.5 - 3 mm around the electrode contact of depth electrodes (Logothetis 2003). Lindén’s group using simulation with a simple biophysical model proved that, for LFP the spatial reach existed but varied depending on the neuron morphology, the synapse distribution and the correlation in synaptic activity. For the EEG / (µ)ECoG recording, signal attenuation during propagation in different brain tissues due to inhomogeneous geometry distribution and anisotropic

109 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings conductivities is another factor need to be considered. And we also wonder that is there also ‘spatial reach’ for the EEG / (µ)ECoG recording contact, or how large the neuron population can be recorded by the EEG / (µ)ECoG recording contact? This should be the first question we address in the present study. From the anatomical level, the EEG signal are thought to be originated from approximately1010 pyramidal cells in the gray matter of cortex because they are well-aliened in a similar orientation and thus the induced PSPs through them can be line up and superimposed a detectable potential at the given location (Nicholson and Llinas 1971), which are mainly found in the layer III, IV and V of cortex (Brodmann 1909). The (µ)ECoG signal should have the same anatomical source but avoid the influence from the skull and scalp during the propagation. Meanwhile it is well-known that EEG / (µ)EcoG activity shows oscillations in different frequencies which are associated with different roles of brain function and sometimes these oscillations from different frequencies co-exist to represent synchronized activity over a network of neurons. Thus a question rise up, that, is there a spatial distributions for the EEG / (µ)EcoG activity, especially in different frequencies? Studies before showed the co-existence of high-frequency gamma oscillations in the superficial layers and low- frequency gamma in the deeper layers in visual cortex of different animal model (rodent, Oko et al 2010; macaque, Maier et al 2010, Spaak et al 2014). Co-existence of low- and high- frequency gamma oscillations has been also found in our experimental study in minipig model (Wang et al 2017), however, whether this layer-specific rule will be kept the same in the somatosensory cortex like in visual cortex as reported or not, present article will address it as the second question, Simulation with FEM model for the conventional scalp EEG recording has been intensively investigated in the last two decades due to the increasing computational power and more application of mathematical analytical theory. However, FEM modelling for scalp EEG seem to no unique solution (Yan et al 1991) due to the following reason: First, the scalp EEG recording has a very high temporal resolution but relatively low spatial resolution, either the individual electrode contact size or the inter-electrode contact distance, which means only summed activities of a large number of neurons simultaneously activated can be recorded, thus the contributions of individual source will be difficult to distinguish. Second, the position of the scalp EEG is far away from the electrical source due to the scalp and skull layers, which has the complex anisotropic conductivities which brings more difficulties to create a suitable head model. It could be a solution to set the skull with an isotropic conductivity which make the computation easier but less accuracy. Meanwhile, scalp EEG recording has been reported to be easily influenced by electric fields generated by muscles and other sources (Ball et al 2009) which need to put much more effort to make the signal clear for the following process. Third, application of the FEM model depends critically on the computer technology. A refined head model, number of the electric source and the mathematical equations used to reconstruct the electric field, stronger computing power and larger computing time may need due to those reasons, and sometime in realistic situation, one option might to be sacrificed in order to fulfil others for each individual project which also become one reason of the blurred source localization for the scalp EEG. Comparing with the scalp EEG, intracranial EEG, i.e., ECoG, or as presented in this study the µECoG recording, might have the chance to fill the leak from scalp EEG application and open a new direction for more accurate source localization with FEM modelling because of the well trade-off balance between the less invasive risk and the better spatial resolution. In the present study, simulation with FEM model based on the µECoG recording is first carried out. Regional part of the volume conductor minipig’s head model was refined to be applied in the simulation part in order to improve the solution to solve the forward problem with µECoG recordings. Electric field was simulated with current sources with different size

110 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings applying at different depth in gray matter (cortical layer). Through validation of evoked cortical activity in vivo and the simulated cortical activity with the same µECoG electrode array, we try to explain the relationship between the neural sources and the recorded µECoG signals, especially in different frequency components. Additionally, refining in a local cortical area also serves the purpose of finding the optimal subdural spacing for some clinical context, e.g., the brain-machine interfacing (BMI).

2. METHODS In the present study, experimental in vivo data were acquired previously which are explained more in our other paper (Wang et al 2017). There, two kinds of custom μECoG electrode array (smaller one with electrode contact Ø 0.87mm and 1.68 mm inter-contact distance; larger one with contact Ø1.81mm and 3.5 mm inter-contact distance, both had 12 x 4 electrode contacts, designed and manufactured by IMTEK, University of Freiburg, Henle et al 2009) were implanted over 4 Goettingen minipigs’ somatosensory cortex. Somatosensory evoked potentials (SEPs) were recorded with electrical stimulation at minipigs’ nostril at different combination of stimuli intensities and sites, and which were also analyzed in steps to prove the signal quality recorded with these novel μECoG electrode arrays. One of those analyse method was term as point spread functions (PSFs), which can explain how far across the cortical surface does direct activation by a sensory point stimulus spread across the cortical surface via local cortical currents. In the topographic maps of relative power changes for the animal experimental data (Wang et al 2017), signal distributions were like radial area surrounding the positive / negative significant peaks, and radius of such areas became smaller from lower- to higher- frequency bands / bins with the same stimulation parameters. Simulation is the main topic in this paper, and as explained in the introduction part, conflict between accuracy and computational efficiency is always needed to be considered in a FEM modelling. There are three key points for the FEM modelling: a volume conductor model with which the geometry of the head model with different electric properties depending on the tissues can be described; a source model which can model the electric source inside the brain for each neural activity and a mathematical solution to solve the problem how the activity will be fired and propagated in the volume model. Usually the first two factors decided the computation complication for the whole process. As suggested by Spyrakos (1996), FEM modelling should start with a simple model. The results from the simple model combined with an understanding of the behaviour if the system will help us decide whether and at which part of the model further refinement will be needed. We took this suggestion and separate the whole simulation presented here in two steps: simple model with limited current sources and refined model with enlarged current sources.

2.1. Simple FEM model with limited simulation current sources 2.1.1. Head model of minpig’s brain and placement of µECoG electrode arrays A three-dimensional (3D) brain model was created by segmenting a post-mortem isotropic (0.4 x 0.4 x 0.4mm) T1-weighted MRI of the minipig’s brain obtained on a Bruker 9T Machine (Bruker Corporation, MA; maximum gradient strength of 40 mT/m, TR = 1900 ms, TE = 900 ms, 32-channel receiver coil, section thickness = 0.4 mm, field of view 100 x 100 mm interpolated to 256 x 256 matrix, figure 1 (a), (b)). As the brain was fixed in formalin before the MRI was performed, the tissue contrasts were changed and the segmentation of the cortical gray matter was performed in the following steps. First, the brain was extracted from the background using the BET tool (Smith 2002) provided in the FSL toolbox (Smith et al

111 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

2004, Woolrich et al 2009). The extracted brain volume still had many segmentation faults probably due to the high signal amplitude of background formalin and the fact that the algorithm was originally designed to extract brain tissue from whole head scans, preferentially using T1 and T2 data simultaneously (Smith 2002). Second, segmentation errors were corrected using our self-written semi-automatic gray-value threshold based 26 direction 3D flood filling algorithm, taking into account the faulty extraction to impose local constraints / flexibilities. Once the brain tissue was sufficiently separated from the background and the cerebellum removed, it was divided into gray and white matter (figure 1 (a), (b)). Because the formalin fixed tissue had a very homogeneous contrast, conventional segmentation tools like FAST (Zhang et al 2001) and our semi-automatic 3D flood filling algorithm failed. Third, we created a 6 voxel (6 x 0.4 mm = 2.4 mm) thick layer of gray matter covering the entire outer brain surface. The rest of the brain volume was declared as white matter. Finally, the (partially resected) dura mater was modelled using the mesh_shrinkwrap algorithm implementation for MATLAB by Darren Weber (freely available at http://eeg.sourceforge.net/ as part of the Bio-electromagnetism MATLAB Toolbox). Reconstruction of the µECoG electrode arrays on the above 3D minipig’s brain model is important for the later simulation computing because during the simulation each electrode center is attributed as the closest node at the surface of the model. The potentials at those nodes are then used to compare simulation and measurements. The main challenge in this step was the projection from 2D μECoG electrode arrays onto the 3D surface of the brain and adapting it to the local gyral geometry constrained by the physical properties of the electrode arrays (figure 1(c)). Details about this procedure can be found in (Fiederer et al, 2016b).

Figure 1. MRI data, segmentation and model. (a) Coronal slice through raw T1MRI data. (b) Final segmentation of brain as shown in (a), white: white matter, gray: gray matter, blue: CSF, red: blood, orange: dura mater. (c) 3D rendering of cortical surface (pink), smaller μECoG electrode array (Φ 0.87 mm and 1.68 mm inter-contact distance, blue circles) and larger μECoG electrode array (Φ 1.81 mm and 1.68 mm inter- contact distance, green circles). The cortical surface inside the dashed black line is the somatosensory cortex of minipigs.

2.1.2. Coordinates and orientations of the dipoles The current source can be modelled mathematically by means of a current dipole. For the simple model, idea of the dipoles positions was to put one dipole direct under each electrode contact (48, blue points in figure 2 (a), (b)) and one dipole under the interpolated points which in the middle of each pair of neighbouring contacts (113, cyan points in figure 2 (b), (c)), thus for the simple FEM model for simulation, we got 161 dipoles which can create a signal with twice spatial resolution of the electrode arrays (inter-contact distance, 0.84 mm). Based on the previous study from Oke et al (2010), which show slow-γ (20 - 45 Hz) cycles were approximately came from 1.3 mm (layer V) from pial surface and fast-γ (46 - 80 Hz) cycles were from 0.6 mm (layer III) from rat model, we set our dipoles in those two depths inside the gray matter (figure 2 (c), (d)) for the simple model. For each 3D position of the 161 contacts centres, we found the closest voxel in gray matter for it, then use this voxel as center of a

112 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings sphere with radius of the given depth and as origin point for a line oriented to the gray matter surface, intersection of line and sphere lying in the gray matter would be taken as the position of one dipole. Details for this step can be found in Fiederer’s diploma thesis (2012). Final presentation of the dipoles is shown in figure 2 (c), (d).

Figure 2. Illustration of the source space of the simulation with simple FEM model. (a) 3D rendering of cortical surface (gray matter, gray background) and smaller μECoG electrode array (Φ 0.87 mm and 1.68 mm inter- contact distance, blue circles). (b) Representation of interpolated points (cyan circles) between the contacts shown in (a). For example, cyan circles in the red frame (E12, E13, E24, E34, E14) are the middle points of all neighbouring contacts (E1, E2, E3, E4) as shown in (a). (c) Dipoles placed below each contact and interpolation point at different cortical depths: red arrows indicate the 0.6 mm cortex depth and green ones are at 1.3 mm. (d) another perspective view for (c).

2.2. Refined simulation model with enlarged simulation dipole positions Results from simulation with simple model show a best fitting between experiment data and simulation data at 0.6 mm dipole depth for the gamma-band (45 - 400Hz) activity with smaller electrode array which encouraged us to refine the FEM model to get more precise results. First thing need to clarify here is, the experimental data used for the refined simulated data were changed to a high-frequency-resolution analyzed data which had 10 Hz frequency resolution, details for this experimental data see Wang et al 2017.

2.2.1. Refine step 1: element expanding Due to the dipole model (venant dipole model, Wolters et al 2007) used in the simulation, the best spatial resolution of the dipole position could be matched in the simple model was 0.4 mm. For the next simulation step, we planned to set dipoles from 0.2 mm to 2 mm in gray matter in 0.1 mm step in order to repeat the results from simple model and get some more accurate results between cortex layer and neuronal activity in different frequency bands, which meant the element. One idea to perform this is to split the tetrahedral element of the simple brain model in the following way: i) interpolated the gravity center for each tetrahedral element from the simple model (e.g., red point P in figure 3a for tetrahedral N1N2N3N4) as a

113 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings new node; ii) then the new interpolated nodes combined with the old nodes in each tetrahedral element from the simple model composed the new expanded elements for the refined model (e.g., expanded tetrahedral PN1N2N3, PN1N3N4, PN1N2N4, PN2N3N4 for tetrahedral N1N2N3N4 in figure 3a). However, we realized if we made this refine step for the whole brain model, the related data size would be extreme large that the computing time would be months long last or stronger computing system need to be set in order to get the results in time which also took time to implement. In the end we decided to perform the refine steps only for the cortex area, i.e., gray matter layer because the main neuronal activity should be elicited in this cortex layer (Kandel et al 2000, Shipp 2007), directly under the electrode array as a compromised solution: i) separate the gray matter layer from the simple model by MATLAB (version R2012b, The MathWorks Inc., Natick, MA) and SCIRun softwares (gray background in figure 3b); ii) depend on the 3D coordinates of the electrode array contacts (blue points in figure 3b), select adequate volume elements direct under the electrode array (white points in figure 3b) and give them a new label in the segmented MRI data (white area in the red circles in figure 3c); iii) base on the new marked MRI data perform the refine steps as described above in Vgrid. figure 3d present the enlarged perspective from part of the final refined model, where the yellow line is the separatrix between the expanded elements (left side from the yellow line) and the original simple elements (right side from the yellow line), which show clear difference between these two parts.

Figure 3. Refine step 1 for the 3D Göttingen minipig brain model: element expanding. (a) Principle algorithm to increase the spatial resolution of the 3D model. Yellow points and lines are the nodes and edges of the element from the first simple 3D model. Red points and dotted line are the interpolated node and edges for each element from the refined 3D model. (b) Area needs to be refined direct under the µECoG electrode array in the simple brain model. Blue points are the µECoG electrode array contacts; gray background is the gray matter of the simple brain model; cortex with white points is the selected area need to be refined. (c) MRI data set of the segmented brain model with new labelled gray matter area, white color in the red circles, which originated from the white points in (b). (d) Enlarged perspective from one part from the final refined brain model combined both refined part (area in the left side of the yellow line) and remained part (area in the right side of the yellow line), which separated by the yellow line. Red lines are the edges from the elements from both refined and remained model. (a), (b) and (d) are visualized through SCIRun and (d) is by the custom MTV Matlab scripts.

114 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

2.2.2. Refine step 2: smoothing After the refine step 1, no matter from the MRI data set (figure 4(a)) or through the SCIRun visualization method (figure 4(c)), the 3D rendering surface of the brain volume model was still obviously rugged and which had a bad influence on the surface orientation and led the following computing for the dipole positions and forward simulation not so precise enough. To get a better surface orientation a spatial filter need to be performed hereafter to achieve the goal of smoothing. For the present study, we adopted median filtering on the 3D MRI images in three dimensions, and each output voxel in new images (e.g., figure 4(b)) contained the median value in a given number size of neighborhood around the corresponding voxel in the original images (e.g., figure 4(a); matlab image filtering function medfilt3, biomecardio toolbox). After some test, we found if we filtered the segmented images with the same parameters, i.e., involved neighborhood voxel number, some information was lost at other segmented slices. Thus, in the end, we filtered the gray matter, whiter matter and new labelled part from refine step 1 with a different parameter compared with the rest segmented slices to get a best trade-off between smoothing and information lost. 3D rendering of the cortical surface was clearly more consecutive and realistic (figure 4(d)).

Figure 4. Refine step 2 for the 3D Göttingen minipig brain model: smoothing. (a, b) MRI data sets after refine step 1 before (a) and after (b) smoothing. ‘Steps’ effect represents clearly at the places where the red arrows point in (a). Both are visualized by the custom MTV Matlab scripts. White areas are the new labelled cortex layer where the refine step 1 performed. (c, d) Refined 3D brain model before (c) and after (d) smooth. ‘Steps’ effect represents clearly over the surface in (d). Both are visualized through SCIRun. Gray backgrounds are the gray matter from the refined brain model and the purple area are the cortex area where the refine step 1 performed.

115 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

2.2.3. Refine step 3: enhancement of dipole positions In the simple model simulation, there were only 161 dipoles with at least 0.84 mm distance in between which made the distributions of dipoles with plenitude space (figure 2(d)). Thus, to have a sufficient number of dipoles which can cover the activated cortex area became an important step in the refine steps. We planned to set each element from the cortex volume (i.e., gray matter) direct under the electrode array as a dipole position to achieve this goal: i) due to the different inter-contact distances used in the animal experiment (3.5 mm for larger electrode array and 1.68 mm for smaller electrode array), signal profile (PSF) was also computed in different distance resolution (2.0 mm for the data recorded with larger electrode array and 1.0 mm for that with smaller one), which made the results from simple model simulation combining from two inconsistent data sets. Additionally, from figure 1(c) we found the area covered by the larger electrode array was out of the boundary of somatosensory cortex. Due to the pure stimulation method for the minipig experiments, we want to limit the simulation area only in the somatosensory cortex area for the refined model. Thus, for the refined model, we decided to use only the data recorded with smaller electrode arrays for the next simulation test. ii) import the refined gray matter volume and 3D coordinates of the electrode array contacts into the MATLAB environment and made a boundary of required elements in the refined cortical volume according to the electrode array contacts positions (red dotted line in figure 5(a)). iii) based on this boundary compute with the custom MATLAB scripts to get a new cortical volume direct under the electrode array (visualized as the yellow area in figure 5(b) by SCIRun). iv) take each element from this new cortical volume as a dipole positions in different cortical depths inside gray matter would be computed (from 0.2 mm to 2.0 mm in step of 0.1 mm, color-coded present in figure 5 (c), (d)). v) separate the outer-surface and inner-surface of the gray matter and find the closest points on the outer-and inner-surface for each dipole obtained at iv) direction between these two points was chosen to be the orientation for each dipole. In the end, we had more than 17,000 single dipoles for each cortical depth which distributed uniformly direct under the electrode array and would provide a more precise simulation in the following step. v) furthermore, for the dipoles at the same cortical depth, different dipole clusters were also selected, which means use each dipole as a circle center with required radium (from 0.5 mm to 3.0 mm in step of 0.5 mm), thus single diploes inside this circle would be considered as a dipole clusters, in order to see the results with different sizes of dipole sources. In the end, we had 133 dipole conditions (19 cortical depth from 0.2 mm to 2.0 mm in step of 0.1 mm, vertically; 7 dipole source size from single dipole to dipole cluster with radius from 0.5 mm to 3.0 mm in step of 0.5 mm, horizontally) for simulated data with refined FEM model. Note, for the dipole clusters, simulated data were not directly from the FEM simulation process, but rather a summation from the single dipole simulation results which were included in the respective dipole clusters, the correlation between the dipoles inside the cluster were taken as ‘1’ in the present study.

116 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

Figure 5. Refine step 3 for the 3D Göttingen minipig brain model: enhancement of dipole positions. a) Minified boundary (red points with black lines in between) for the new segmented cortical volume in the refined gray matter volume visualized in MATLAB environment, where gray background is the gray matter volume from the refined brain model, blue points are the electrode array contacts, purple lines are the boundary of the refined cortical area as described in refine step 1. b) Same content as in (a) but visualized by SCIRun, where yellow area was segmented according to the red dotted line in (a) and purple area is the refined cortical area as described in refine step 1. c, d) two perspective view for the single-dipole positions in different cortical depths which are color-coded showing in (c) and (d) and all visualized by SCIRun.

2.3. FEM simulations Once we had the proper brain volume model and dipole information (position and orientation), FEM forward calculations were computed with SimBio / NeuroFEM (SimBio Development Group, 2009) (freely available at https://www.mrt.uni- jena.de/simbio/index.php/Main_Page) using the Saint Venant dipole model (Wolters et al 2007b) for each dipole separately. To the best of our knowledge, the conductivity of brain tissue has not yet been systematically investigated in pigs, much less in Göttingen minipigs. Thus, each compartment was attributed human conductivity as listed in Table 1 (Haueisen et al 1995, 1997, Manola et al 2005). Visualizations for simulation model (figure 2 - 5) were created with SCIRun (freely available at http://www.sci.utah.edu/cibc/software/106- scirun.html).

117 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

Table 1. List describing which algorithm was used to segment given tissue compartment (kindly provided by Fiederer D.J. L from his diploma thesis (2012)).

Head Tissue Segmentation Conductivity (S/m) References

White Matter Regional Growth 0.1429

Gray Matter Layering 0.3333 (Haueisen et al., 1995)

Liquor Regional Growth 1.5385

Blood Regional Growth 0.6250 (Haueisen et al., 1997)

Dura Mater Shrink wrap 0.0650 (Manola et al., 2005)

2.4. Data analysis After simulation, an amplitude value was assigned to each element node inside the whole 3D brain volume model. Simulated LFP amplitudes at the element nodes closest to each contact � were common averaged and set to be the amplitude for each contact (V�), respectively. Based on ��, together with experimental data, simulated spectral power for each contact and the following analysis can be performed step by step.

2.4.1. LFP strength with dipoles located at different cortical depths There is some special computing need to be explained at the beginning that for simulation with simple brain volume model and simulation with refined model but with single dipole, �� was direct from the FEM simulation process. For simulation with refined model but with dipole clusters, �� would be a linear summation of the corresponding �� resulting from single dipole which were included in related circles as we defined in section 2.3.3.. For each dipole position � (� could be a single dipole or a dipole cluster with a total number more than 17,000 at each cortical depth with refined model), maximum of ��,� =1:48 simulated at the electrode contacts was taken as the maximal simulated LFP amplitude for this dipole position ����� ,

����� = max([�1, …,��0]), �0 = 48 ;

Median of ����� (�̃����) with the dipoles at the same cortical depth was considered as the maximal simulated LFP amplitude at this cortical depth, respectively. Then �̃���� was compared for different cortical depths for both simple and refined model. With refined brain volume model and enhanced dipole positions, we had the possibility to test the ‘spatial reach’ from previous study (Lindén et al 2011), that how the LFP amplitude varied with distance away from the neuron source: i) for each contact �, more than 17,000 �� were obtained after simulation at each cortical depth. Dipole which simulated the strongest LFP amplitude was chosen first (������� , cyan dipole related to red contact �� shown in figure 6(a)). ii) two kinds of distances were then computed: one was the distance between contact and each single dipole (contact-dipole distance �1 in figure 6(a)), the other was the

118 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings distance between ������� and other dipoles at the same cortical depth (dipole-dipole distance �2 in figure 6(a)). iii) simulated LFP amplitudes from all dipoles at the same cortical depth were plotted against the related �1,�2 separately from smaller to larger �1,�2 (figure 6(b) for �1 and figure 6(c) for �2). iv) a polynomial fitting curve was computed based on the points cloud in (iii) as the green curve shown in figure 6 (a), (b). v) contact-dipole distance of ������� was marked as ���� (blue star in figure 6(a)), and contact-dipole distance at which simulated the LFP amplitude attenuated to 95% of the maximum LFP amplitude value � � ( �������) was marked 95% (5% of the maximum response termed as the threshold to obtain meaningful signal as used in Lindén’s study, red points inside the red circle in figure 6(a)); vi) ‘spatial reach’ ���11 as borrowed from this study, based on the contact-dipole distance �1 for each contact �, can be defined as,

���11 =�95% −����,� =1∶48 vii) for each cortical depth �, median of ���11 across all electrode contacts was taken as the ‘spatial reach’ ���11

���11 = ���11,�=1∶̃48,�=1:19 viii) Repeat the step (v) - (vii) for another ‘spatial reach’ ���12 based on the dipole-dipole distance �2 for each cortical depth � afterwards. For �� with different circle group of single dipoles, �� was not direct from simulation, thus this ‘spatial reach’ computing in the present study was only performed in single-dipole condition. As was expected, ‘spatial reach’ existed with present simulation condition as suggested by the previous study (Lindén et al 2011).

After we computed �1, �2 for all related contact-dipole or dipole-dipole pairs for each cortical layer, we can also sort the dipoles at the same cortical depth into different groups based on the �1, �2, respectively: i) for all �1 from the same cortical layer, define a distance range from the minimal �1 to maximal �1; ii) separate this distance range from �1 to �2 in 0,1 mm step to get many sub-distance groups; iii) based on those sub-distance groups, sort the dipoles from the same cortical layer into those sub-distance groups; iv) set the correlation of dipoles inside each sub-distance groups as 1, and sum the simulated activity linearly by those dipoles to get a summated activity ����. In the end we had the plot between those summed activities and sub-distance groups (figure 6 (d)). v) initial distance group and the sub-distance group in which maximal summed activities was obtained (�0,���� in figure (d)) can marked directly from the plot, and ‘spatial reach’ from electrode-concentric side for each contact � can be defined as

���12 =���� −�0,� =1∶48 vi) for each cortical depth �, median of ���12 across all electrode contacts was taken as the ‘spatial reach’ ���12,

���12 = ��̃�12,� =1∶48,�=1∶19

Repeat step (i) - (vi) for �2 and we got the same plot as with �1 and ���22based on the dipole- dipole distance (figure 6(e)).

119 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

Figure 6. Illustration of spatial reach (SR) computing with refined brain volume model and enhanced dipole position with single dipole condition. (a) Visualization of the method to calculate the contact-dipole distance and dipole-dipole distance as explained in section 2.4.1. Blue spheres are the electrode contacts, and orange point- cloud is the more than 17,000 dipoles at one cortical depth. Distance we mentioned here are all the Euclidean distance. (b) Point-cloud (blue points) plotting between simulated LFP amplitudes and corresponding contact- dipole distance. Green curve is the polynomial fitting curve based on that points-cloud. Blue star indicates the contact-dipole distance with that dipole simulated the maximal LFP amplitude and magenta horizontal dotted line is the 5% of this simulated maximal LFP amplitude, where it met the green curve is the contact-dipole distance with that dipole simulated LFP attenuate to 5% of the maximal one, marked as red points in red circle. (c) The same plot as in (b) but with the dipole-dipole distance. Symbol and color can be referred as in (b). (d) Summated activities from dipoles in different distance groups. Distance group where produced the maximal summated activity and the initial distance group of the summated activity are marked as ���� and �0 direct from the plot. (e) The same plot as in (d) but with the dipole-dipole distance. Symbol and color can be referred as in (d).

120 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

2.4.2. Topographic maps of relative spectral power changes for both experimental and simulated data For experimental data, baseline variation-corrected relative power changes were presented because they reflected better differences in different frequency bands and computed using the mean over the time bins before stimulus onset as a baseline (��������), by which all recorded power (��������) values were divided (details see Wang et al 2017). Usually �������� can be regard as a mixture of absolute power (��������) and background noise power (��������), and the �������� came from a period just before the ongoing stimulus onset which reflected the background state for the whole ongoing stimulus, thus,

�������� �������� + �������� �������� + �������� �������� = = = ; �������� �������� �������� Topographic maps of the �������� were computed for all defined frequency ranges, time windows and sessions using a linear interpolation method for smoother visualization (figure 7(a)). Different symbols were used to visualize the results of the statistical analyses in these maps (see captions of figure 7 for details; Wang et al 2017).

For simulated data, square of simulated LFP amplitude (V�) was considered as the absolute spectral power (��������) at each contact. Simulation is a noise free environment, thus there would be no difference between different frequency ranges for the signal profile (PSF) if we only computed it with ��������. Theoretically, for each electrode contact �,

��������� + ��������� ��������� = ��������� = ; ��������� and background noises can be formed as,

��������� ��������� = ��������� −1

However, in those equations, �������� should be at the same level at least in a local area for all contacts. Thus, for simulated data we used the respective maxima from all electrode contacts to compute �������� in order to keep the background in the same level for all contacts.

������������ ��������� = ������������ −1

Topographic maps of the �������� would be computed for all defined frequency ranges, time windows and dipole positions with the same method for the experimental data (figure 7(b)).

121 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

Figure 7. Topographic maps of relative power changes and PSF computing. (a) Topographic maps of relative power changes from one experimental session (smaller electrode array in frequency band 264 - 408 Hz in a time window of 5 - 14 ms after stimulation onset and at 8.0 mA from one minipig), where different symbols indicate the statistic state for the electrode contacts (details for statistic calculation see Wang et al 2017): contacts with significant positive and negative potential changes are marked by larger filled black and white circles, respectively (sign test, FDR-corrected at q < 0.01). The white and black dots in the center of electrodes indicate the contacts with the maximum positive and negative power changes, respectively. Among the remaining electrode contacts, which did not reach significance after FDR correction, those with p ≤ 0.3 (sign test) are marked by larger filled gray circles, and those with p > 0.3 (sign test) are marked by the smaller filled black circles. (b) Topographic maps of relative power changes from one simulation session (larger electrode array with a central dipole (n°80) in gray matter at 0.6 mm for frequency bin 80 - 150 Hz). There was no statistic calculating there and electrode contact with filled black circle is the one with positive maxima responses in related simulation condition. (c, d) Comparison of signal profile (PSF distributions) for single experiment session (blue lines) and dipole position in different cortical depths (red line for 0.6 mm and green for 1.3 mm) with simple brain volume model for larger (c) and smaller (d) electrode array respectively. Note, there are distance bins missing with larger electrode array because of the spatial resolution of the electrode array itself (black circles in (c); see 2.4.2 for details). HWHM for the single session under those three conditions (blue for experiment data; red and green for simulated data in different cortical depths, 0.6 mm and 1.3 mm respectively) were also marked in (d).

2.4.3. Point spread function computing After the simulated relative spectral power changes (relSP) for all frequency bins / bands were calculated, point spread functions (PSFs) for the simulated data were ready to be computed as well as for the experiment data, which intended to reveal the relationship between power intensity and distance between electrode contacts. To do this, i) the ‘maxima contact’ was chosen as a reference contact for each recording / simulation condition (frequency range, time window, recording session for experimental data (contact with significant positive maxima which marked with black fulfilled circles with white point inside in figure 7(a)) and simulated dipole position for simulated data (contact with positive maxima which marked as fulfilled black circles in figure 7(b))). ii) relative spectral power changes from all contacts were divided by that from reference contact as the normalized relative spectral power changes

122 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

(nrelSP), and meanwhile distances between the reference contact and other contacts were also calculated (figure 7 (a), (b)). iii) contacts were sorted depending on the distances between the reference contacts, i.e., contacts which in the same range far away from the reference contacts would be in the same contacts group. iv) for each contacts group, median values of nrelSP across all suitable contacts were calculated as the nrelSP standing for the respective distance range. In the end a PSF profile can be plotted as a dot-dash line as contact-contact distances vs nrelSP changes (figure 7 (c), (d)). Note, for the larger electrode array, due to its 3.5 mm inter-contact distance, distance range was chosen to be a 2.0 mm radius circle in 1.0 mm step, i.e., in figure 7(c), distance bin ’0’ in x-axis stands for the reference contact, ‘1’ for a range 0 - 2 mm, ‘2’ for 3 - 4 mm, ‘3’ for 4 - 5 mm, ‘4’ for 5 - 6 mm, and so on. In result some contacts were assigned to two contact groups in order to get enough available distance bins. Even though, some distance bins were still void, e.g., the distance bins marked as black circles in figure 7(c). Conversely for the smaller electrode array, distance ranges were a 1.0 mm circle in 1.0 mm step, i.e., in figure 7(d), distance bin ’0’ in x-axis stands for the reference contact, ‘1’ for a range 0 - 1 mm, ‘2’ for 1 - 2 mm, ‘3’ for 2 - 3 mm, ‘4’ for 3 - 4 mm, and so on. And there were no void distance bins as for the larger electrode array which was one reason for the simulation with refined model only the data with smaller electrode array would be included. These steps then performed for each recording / simulation condition as described above respectively. In the end, median of the PSF for each recording / stimulation condition were calculated across valid sessions (this is specified for experiment data where the significant reference contact exist) for all distance bins. Half width at half maxima (HWHM) is a convention used in , which means the half width of a spectrum curve measured between those points on the y-axis which are half the maximum amplitude. In the present study we borrowed this definition and calculated the HWHM for the PSF profile in each recording / simulation condition (figure 7(d)) and a mean HWHM across all valid sessions was also computed for experimental and simulated data i.e., larger and smaller respectively or dipole in different cortical depth, respectively, in order to find out to which contact distance the recorded neural activity would attenuate to at least the half the maximum.

2.4.4. Comparison of PSF profiles between experimental and simulated data Comparison of PSF profile between experimental and simulated data with simple model was simple and direct, because there were only three PSF profile need to be compared for larger and smaller electrode array separately in each frequency band and time window: experimental data, simulated data with dipole in cortical layer 0.6 mm and 1.3 mm, as the examples shown in figure 7 (c), (d). Thus, it appeared clearly that with dipole in which cortical layer match the experimental data better. However, with refined model combined with high-frequency resolution analyzed experimental data, even only with smaller electrode array, for each frequency bin and time window, there were 133 (19 x 7, 19 cortical layer depth with 0.2 mm to 2.0 mm in step of 0.1 mm and 7 type of dipole groups from single dipole to dipole cluster radium from 0.5 mm to 3.0 mm in step of 0.5 mm) PSF profiles need to be compared. Thus we computed an error- map based on ������ difference at each distance bin (we term it as disbin error hereafter) between experiment data and simulated data for each frequency bin and time window: for each distance bin � showing in the PSF profile with refined model (e.g., figure 8 (a), (b)), we had ������ from experimental data ���������� , and simulated data for each dipole condition �, ����������� . Thus, for each distance bin � and each dipole condition �, we got a distance bin error ���,

123 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

��� = ����������� − ����������,� =1:17,�=1:133; Impression from the PSF profile that the 3rd, 4th and 5th distance bin from the PSF profile (the one inside the black rectangle in figure 8 (a), (b)) were the 3 distance bins which decided the similarity degree between experiment and simulation data, while distance bins after those three for the experimental data usually had the negative effects or some twists and turns due to the existing of the multiple significant spatial local maxima as shown in figure 7(a) (details see Wang et al 2017) which could not be presented through the current simulation method. Thus, only these three distance bins were included in the following calculation and distance bin error �� between experiment data and simulated data with dipole condition � for each frequency bin and time window was updated as, 2 5 2 ∑5 (������ − ������ ) ∑�=3 ��� �=3 ����� ���� � = = ,� =1:133; � 3 3

To make sure the accuracy of the ��, cross validation (leave-one-out) was used here for the experimental data: for valid experimental sessions (the one which had reference contact for each frequency range and time window respectively, e.g., number �), we took one of them out of the experimental data set sequentially, and performed the steps as described above to compute the ��� with the rest of experimental data, then at the end we got a group (number �) of ��� for each dipole condition, and median of ��� across the valid session (number �) was taken as the final distance bin error ��� for each dipole condition, each frequency bin and time window.

∑� ��� � = � �

Based on the ��, at least for each frequency bin and time window, a color-coded image map (figure 8(c)) can be made to give a direct impression of the similarity degree of those 133 dipole conditions, that smaller �� with cold color indicated better matching degree with the related experimental data. Furthermore 5 minimum �� locations were marked in such color- coded image (white circles in figure 8(c)) in order to see with which dipole condition simulation data had better similarity degree (smaller ��) to the related experimental data.

Figure 8. Comparison of PSF between experimental and simulated data with refined brain model. (a) Median PSF profile for high-resolution analyzed experimental data at frequency bin 130 Hz in time period 5 - 105 ms across all valid recording sessions, blue dashed line with error bar at each distance bin in the same color which indicated the standard error across all valid recording sessions. Other solid lines are the PSF profile from the simulated data combined experimental data for the same frequency bin and time period as presented in (a). Different color indicated the dipole condition from single one increasing to a cluster with radius from 0.5 mm to 3.0 mm in step 0.5 mm, but all in the same cortical depth 1.3 mm, where cold to warm color are the circle radius from smaller to larger one. Error bar are the standard error across all dipole in the same dipole condition. (b) Blue dashed line presents the same experimental data as in (a). Other solid lines are the PSF profile are also from the simulated data combined experimental data for the same frequency bin and time period as presented in (a), but different color indicated the dipole condition in cortical steps from 0.2 mm to 2.0 mm in step 0.1 mm, but

124 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings all in the same dipole cluster radius 1.0 mm, where cold to warm color imply the smaller to larger distance away from the cortical surface. Error bar are the standard error across all dipole in the same dipole condition. (c) Color-coded distance bin error map for frequency bin 130 Hz and time period 5 - 105 ms. Different color stand for the difference of relSP between experimental and simulated data as show in (a) and (b), where cold color for the smaller difference and warm color for larger one. White circles imply the dipole conditions where the first five minimum differences were computed, number near the circle are the ranking for minimum degree.

3. RESULTS 3.1. Topographic maps of spatial distribution of signal simulated in FEM model Topographic maps based on the simulated signal, no matter with the LFP amplitude or with background adjusted relative power changes through experimental data, spatial distributions show the same tendency, that all maps were composed by one contact with maximal positive response surrounded by other contacts with decreasing responses in radius direction (figure 9). Topographic maps with multiple response-peaks which separated the maps into different activated areas like presented by the experimental data (figure 7(a); more experimental results can be found in Wang et al 2017) were not observed in the simulated data. Topographic maps with simulated data for the larger electrode array had a relative smaller activated area compared with the maps based on the simulated data for the smaller electrode array which had the similar peaks area and were more like a zoom-out of the activated area obtained by the large electrode array if simulated with the same dipole (comparison between upper and lower panels in figure (a), (b), (d), (e), (g) and (h)). This zoom-out result between larger and smaller electrode array were also observed in the experimental data (Wang et al 2017). Topographic maps with signal differences between different dipole locations in the cortical layer (figure 9 (c), (f) and (i)) showed that signal strength was stronger if the dipole was located in the superficial cortical layer especially for the data simulated with the smaller electrode array.

3.2. Comparison of signal strength between experiment and simulation data To quantify the difference of simulated signal strength with different dipole location or with larger / smaller electrode array, we also compared the maximal positive responses obtained in experimental and different simulation conditions. Comparison of potential amplitude from experimental data between larger and smaller electrode array showed no clearly difference for the early responses (14 - 23 ms after stimulation onset, N20 of the somatosensory evoked potentials (SEPs), blue bars in figure 10(a)), and the same comparison from simulated data with simple FEM modelling showed that with smaller electrode array stronger LFPs were obtained with different cortical dipole depths (red and green bars in figure 10(a)). Furthermore, dipoles simulated at 0.6 mm cortical depth produce stronger LFP than dipoles located at 1.3 mm cortical depth as measured by the smaller and larger μECoG electrode arrays, respectively (figure 10 (a), (b))), and the difference of simulated LFP with larger and smaller electrode array was smaller with dipole located in deeper cortical layer, at least for the maximal positive responses for all comparison components. Comparison of relative spectral power changes (relSP) was more complicate. For the experimental data, we observed stronger response with larger electrode contacts in lower frequency bands (till to 45 Hz), and in higher frequency bands, i.e., gamma frequency bands, larger responses were obtained with smaller electrode contacts (blue bars in figure 10(b)). However, for background adjusted relSP for simulated data, maximal relSP was obtained always with larger electrode contacts, except for

125 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings the responses in 45 - 80 frequency bands. The same as potential data, relSP was also stronger with dipole located more superficial cortical layer and the difference of relSP between larger and smaller electrode contacts, or between different cortical dipole depth, was always stronger in lower frequency bands.

Figure 9. Topographic maps of simulated cortical dipolar activity with different dipole depths in simple FEM model and with larger and smaller electrode arrays, respectively. (a, b) LFP amplitude produced by central dipole (n°80) in the depth of 0.6 mm (a) and 1.3 mm (b) measured by larger (upper panel of (a) and (b)) and smaller (lower panel of (a) and (b)) μECoG electrode array respectively. (c) LFP amplitude differences obtained between this central dipole (n°80) locating at 0.6 mmand1.3 mm depth measured by larger (upper panel) and smaller (lower panel) μECoG electrode array, respectively. (d, e) Background adjusted (by experimental relative spectral power changes in 20 - 45 Hz) relative spectral power changes (rel.power) produced by the same dipole as in (a) and (a) in the depth of 0.6 mm (d) and 1.3 mm (e) measured by larger (upper panel of (d) and (e)) and smaller (lower panel of (d) and (e)) μECoG electrode array respectively (detail see Method 2.4.2). (f) Background adjusted relative spectral power differences obtained between this dipole locating at 0.6 mm and 1.3 mm depth measured by larger (upper panel) and smaller (lower panel) μECoG electrode array respectively. (g, h, i) the same data structure source as in (d), (e) and (f), but present for the relative spectral power changes in 80 - 150 Hz. Contacts which got the maximal positive responses are marked as fulfilled black circles here and other contacts as empty black circles.

126 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

Figure 10. Comparison of signal strength between experimental data and simulated data with simple and refine FEM model with dipole in different conditions. (a) Median maximal amplitude from experimental data (blue bar, during the time period 14 - 23 ms after stimulation onset) and simulated data with simple FEM modelling of 161 dipoles at 0.6 mm (red bars) and 1.3 mm (green bars) cortical depths for larger and smaller μECoG electrode arrays, respectively, and related error bars are the standard errors across involved recording sessions for experimental data or dipole positions for simulated data. (b) Median maximal relative power changes from experimental data (blue bar, during the time period 5 - 105 ms after stimulation onset) and background adjusted simulated data with simple FEM model in 7 defined frequency ranges. Color coding and error bars are the same as in (a), except the light colors are for the data measured with larger electrode array and dark colors are for the smaller electrode array. (c) Color-coded images of maximum simulated LFP amplitude with different dipole conditions (19 cortical depths and 7 dipole source sizes (details see Method 2.2.3.), where cold to warm color imply the smaller to larger LEP amplitude values. White circles indicate 5 dipole conditions which simulated the LFP amplitudes which had the first five minimal difference compared with the experimental data. (d) Box plots of median maximal relative spectral power changes (relSP) from experimental data and background adjusted simulated data with different dipole conditions for all frequency bins. Boxes are the interquartile range (IQR) of all contacts at each cortical depth, whisker extend to upper adjacent value (largest value = 75 percentile + 1.5 x IQR) and lower adjacent value (smallest value = 25 percentile - 1.5 x IQR). Color-coded dotted lines are the median maximal relSP from all frequency bins for each cortical depth with the same dipole cluster size (with different color presented).

With refined FEM modelling, only smaller electrode array was utilized, and the results showed the same content as with simple FEM modelling that stronger signals (both LFP amplitude and background adjusted relative spectral power changes) were obtained with dipole located at more superficial cortical layer no matter for larger or smaller dipole cluster size from the same cortical layer (figure 10 (c), (d)). However, for dipole from the same cortical layer, the simulated LFP amplitudes were directly proportional to the dipole cluster size (figure 10(c)) but the background adjusted relative spectral power changes were inversely proportional to it (figure 10(d)). And with increasing dipole cluster size, the difference in different cortical depths became smaller which could give a clue that the ere might be a neuronal generation range for the LFP in result of out of this range the LFP strength would not be influenced (figure 10(d)).

127 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

With this clue and the high spatial resolution of dipole positions in the refined modelling step (more than 17,000 dipoles for each dipole condition), how the simulated LFP strength varied with the contact-dipole distance directly or dipole-dipole distance stand for the possible source distance in the same cortical layer indirectly, can be observed similar as presented by the previous study (Lindén et al 2011). Related results showed that, (i) simulated cortical activity attenuated with increasing contact-dipole or dipole-dipole distance in a distance range, otherwise simulated cortical activity kept at the same level if out of this distance range (figure 11 (a), (d)). Corresponding contact-dipole or dipole-dipole distance for this distance range (figure 6 (b), (c); details see Method 2.4.1.) can be defined as ‘spatial reach’ based on the population-centric theory as suggest in Lindén’s study (1991); ii) if we look the simulated data from electrode-centric side (Lindén et al 2011), summation of simulated activities from the dipoles inside a defined distance range based on the contact-dipole or dipole-dipole distance (play the role as neuronal sources with correlation = 1 between each dipole inside) increased quickly in a certain distance range, and then decreased in a relative slower decay and in the end seem to flatten out in the end of the maximal dipole numbers were included in (figure 11 (b), (e)). Results from this part is different from the Lindén’s study (1991), that the LFP increased but not to flatten out to a stable level beyond a certain distance range rather decreased first and flatten out at the end. This could be due to the reason that µECoG electrode contacts were not direct inside the cortical layer, so that the number dipoles inside each distance group is much more than in the Lindén’s study (1991). However, we can still take the distance range where got the maximal summated activity as a ‘spatial reach’ for the electrode-centric theory. Both ‘spatial reach’ defined with contact-dipole or dipole-dipole distance were increased from superficial to deeper cortical depth below the cortical surface (figure 11 (c), (f)), however, the ‘spatial reach’ with dipole-dipole distance were more consistent for all cortical layer (figure 11(f)).

Figure 11. Spatial Reach of simulated LFP with refined FEM model and single-dipole at different cortical depths. (a, d) Illustrations of spatial reach (SR) computing based on the population-centric idea for contact- dipole distance (a) and dipole-dipole distance (d) as explained in figure 6 (b), (c) (details see Method 2.4.1) from the contact which obtained the maximal simulated LFP at two cortical depths, respectively. (b, e) Illustrations of SR computing based on the electrode-centric idea for contact-dipole distance (b) or dipole-dipole distance (e) as explained in figure 6 (d), (e) (details see Method 2.4.1) from the same contact and the same cortical depths as shown in (a), (d), respectively. For explanation of details in (a), (b), (d), (e) refer to figure 6 (b) - (e). (c, f) Box plots of SR of simulated LFP amplitude for all electrode contacts at each cortical depth based on the cortical- dipole ((c), cyan for the SR from (a) and magenta for the SR from (b)) or dipole-dipole (((f), cyan for the SR from (a) and magenta for the SR from (b)) distance, respectively, and boxes are the interquartile range (IQR) of all contacts at each cortical depth, whisker extend to upper adjacent value (largest value = 75 percentile + 1.5 x IQR) and lower adjacent value (smallest value = 25 percentile - 1.5 x IQR) , points beyond the whiskers are displayed using +. Dotted lines are the median SR from all contacts at each cortical depth.

128 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

3.3. Comparison of PSF function between experiment and background adjusted simulation data From the comparison of PSF profile between the experiment and background adjusted simulated relative spectral power changes with simple FEM modelling, we can see that if we ignore the influence of missing distance bin for the larger electrode array (figure 12(a)), then for the larger electrode array, simulated data with dipole position at 1.3 mm cortical depth had a better similarity compared with the experimental data , and for the smaller electrode array dipole position located at 0.6 mm cortical depth could get a more similar PSF profile plot (figure 12(b)). Related HWHM bar plots presented the same tendency that, for smaller μECoG electrode arrays, HWHM was around 1.3 mm similar as experimental data for smaller electrode arrays while HWHM was around 2.0 mm for the larger electrode arrays (figure 12(c)), respectively.

Figure 12. Comparison of PSF based on experimental data and simulated data in simple FEM model with different cortical dipole depths sampled by larger and smaller electrode arrays, respectively. (a, b) Mean PSFs of background adjusted relative spectral power changes in 45 - 400 Hz as sampled by larger (blue line in (a)) or smaller (blue line in (b)) μECoG electrode array in measurements (blue line) and simulations of 161 dipoles (red line) placed 0.6 mm (red lines in (a) and (b)) and 1.3 mm (green lines in (a) and (b)) below cortical surface, respectively. Error bars are the standard error across all involved recording sessions (blue ones) or dipole positions (red and green ones) for each distance bin in x-axis. (c) Median of HWHM computing from the single PSFs of relative power changes in 45 - 400 Hz from each recording session (blue bars in (c)) sampled by larger (green bars and smaller) electrode array, respectively, error bars are the standard error of HWHM across involved recording sessions. Median of HWHM computing from the single PSFs of background adjusted relative spectral power changes in the same frequency band with 161 dipoles (red line) placed 0.6 mm (red bar in (c)) and 1.3 mm (green bar in (c)) below cortical surface, respectively, error bars are the standard error of HWHM across all dipole positions.

The same comparison of PSF profile between experimental and the background adjusted simulated data with refined FEM modelling was also performed. However, for the results with the refined model, there were many parameters need to be considered, i.e., 19 cortical depths, 7 dipole source sizes and 24 frequency bins (till to 230 Hz) in the end. Of course, the corresponding dipole condition which produced the best and worst PSF profiles compared with the experimental one can be selected for each frequency bin (upper panels from figure 13(a)) as for the simple model, but it was not enough to show a clear direction. Thus, an error map based on the distance bin errors between experiment and simulated data from all dipole conditions was computed for each frequency bin, and the dipole conditions which produced the minimal distance bin errors were also selected afterwards (lower panels from figure 13(a)). Furthermore, those dipole positions selected for all frequency bins were summarized in one figure (figure 13(b)), and the median distance bin errors across all dipole cluster size for each distance bin and each cortical depth were also computed (figure 13(c)), which all showed the same tendency that, for lower-frequency activity, dipole located at deeper cortical

129 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings layer produced the PSF profile which was more similar with the experimental PSF profile in the same frequency range, while for higher-frequency activity, cortical dipole position which simulated more similar PSF profile as experimental one was more superficial. HWHM from all simulated PSF profiles with refined FEM model was also calculated hereafter for all frequency bins and dipole conditions (figure 13(d)). HWHM from the experimental PSF profiles with smaller electrode array in frequency range 0 - 230 Hz was around 1.4 mm (median across all frequency bins), and HWHM for simulated PSF profiles increased with both cortical depth and dipole cluster size but with larger dipole cluster the increasing rate was clearly slower. In the end, a color-coded image based on the HWHM differences between experimental and simulated PSF profiles across all dipole cluster size for each frequency bin and each cortical depth was also achieved (figure 13(e)), and the tendency was consistent with the error map of distance bins (figure 13(b), (c)), that for lower-frequency activity, the HWHM of simulated PSF profile, produced by the deeper cortical dipole position, was more comparable with the experimental HWHM in the related frequency bins, and for higher- frequency activity, with more superficial cortical dipole location the simulated PSF profile would have a closer HWHM value comparing with the experimental data.

Figure 13. Comparison of PSF with experimental data and simulated cortical activity with refined FEM model and different dipole conditions sampled by the smaller µECoG electrode array. Simulations were done by putting current dipoles in different cortical depths (19 cortical depths for x-axis) and different dipole source sizes (7 types, single and dipole cluster radium from 0.5 mm to 3.0 mm in step of 0.5 mm which define which and how many dipoles will be included in) in a refined 3D Göttingen minipig brain model (details see Method 2.3.3). (a) Upper: blue curve is the median PSF plot across involved recording sessions in 10 Hz (left) and 170 Hz (right)

130 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

frequency bin, and related error bars are the standard error across those sessions for each distance bin. Red curve is the median PSF plot from baseline adjusted simulated data in 10 Hz and 170 Hz bin across all dipole positions in the condition where we got the minimal distance bin error (details see Method 2.4.4.) compared with the experimental data in the same frequency bin and the error bars are the standard errors across all dipole positions for each distance bin. Green curve is the median PSF from the simulated data where we got the maximal distance bin error compared with the experimental data. Lower panel are the color-coded image based on the distance bin errors from all dipole conditions in 10 Hz (left) and 170 Hz (right) frequency bins. White circles indicate the 5 dipole conditions where we got the first 5 minimal distance bin errors in the related frequency bins. (b) Median of distance bin errors across all involved frequency bins (0 - 230 Hz in 10 Hz step) for all dipole conditions. White circles indicate the dipole condition where in each frequency bin we got the minimal distance bin error, corresponding frequency number was marker nearby, e.g., position of white circle with number 3 indicate for frequency bin 40 Hz, we had the minimal distance bin error with single-dipole located at 1.6 mm cortical depth. (c) Error map of median distance bins between measured and simulated PSF of cortical activity in all frequency bins with single-dipole condition in all 19 cortical depths. Cold to warm color indicate smaller to larger distance bin errors. (d) Box plots of HWHM from the PSF with simulated data for all frequency bins (0 - 230 Hz in 10 Hz step) for each dipole condition (19 in x-axis and 7 in y-axis). Color-coded dotted lines are the median HWHM across frequency bins for all dipole conditions, boxes are the interquartile range (IQR) of all frequency bins for each dipole condition, whisker extend to upper adjacent value (largest value = 75 percentile + 1.5 x IQR) and lower adjacent value (smallest value = 25 percentile - 1.5 x IQR). (e) Error map of median HWHM differences between measured and simulated PSF of cortical activity in related frequency bins across all dipole cluster sizes in all 19 cortical depths, respectively. Cold to warm color indicate smaller to larger difference values.

4. DISCUSSION 4.1. Influence of cortical depth on LFP strength As convinced by most neuroscientists, signals recorded by (µ)ECoG electrode array are composed of extracellular electrical currents synchronized at a given point where the electrode contact located, which also termed as local field potentials(LFPs).Any excitable membrane and any type of transmembrane currents contribute to yield LFP, thus how large is the cortical area which generates the LFP or how far can LFP extend in cortical region, is always the neuroscientists debate for. Normally, the further away from the recording electrode, the more neuron population will be involved in contributing the LFP generation. However, the attenuation effect, that the potential generated by a single neuron decreases with increasing distance between the sources and recording point, can’t be ignored. Due to these two conflict facts, one can assume that, in some certain distance range between source and recording point, the LFP amplitude usually increases with the distance, but it will increase slowly till to stable level when out of this range. This distance range was defined as ‘spatial reach’ in Lindén’s study (Lindén et al 2011) based on the same hypothesis. In their study, through an ‘electrode-centric’ view (us the size of LFP generators to measure the spatial reach), they proved the existence of LFP spatial reach from different perspectives, such as, single-cell LFP distance dependence, dependence of LFP reach on cell morphology, synapse distribution and electrode depth. In the present simulation study, our simulation model provides another view, ‘population-centric’ view suggested by Lindén (2011), to find out how far the LFP can spread outside an active source. Typically prove for this idea need to fix the dipole source position and move away the electrode position to see the changes. In the present study, our electrode contact positions were always at the same place, and the dipole positions were varied in the given distance range. Both methods should explain the same thing from two perspectives. In result, at least for the single-dipole source in the present simulation model, in some certain distance range, the LFP amplitude decreased within the distance range between source and recording place and flattened out beyond a certain distance away from the contact position at all investigated dipole depths (figure 6 (a), (b); cyan box plot in figure 11), and this distance range was depended on the dipole depths inside cortex, i.e., it is larger in the

131 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings deeper cortical layers than in the superficial cortical layers (figure 11 (c), (f)), which are all consistent with the finding from Lindén’s study (2011) that the ‘spatial reach’ was larger if electrode located at layer 5/6 compared with electrode at layer 2/3. However, for the electrode relative far away from the neuronal source, .i.e., for EEG, (μ)ECoG, which is completely above the source, sizeable of contributed neurons direct under the contact play an important role of the recorded LFP at the electrode contacts, and the correlation between these contributed neurons will also influence the recorded LFP at the electrode contacts. In the present study, a modified definition of ‘spatial reach’ which measured a distance range to get the maximum summated activity (magenta box plot in figure 11) was also a tiny increased with dipoles located at deeper cortical depth. Correlation of dipoles inside the distance ranges was set to ‘1’. With different cortical state which depends on the individual experiment situation, correlation between the contributed dipole source will be influenced, and thus the reach to get the maximum summated activity will also varied, i.e., might be larger than with correlation ‘1’ as reported in Lindén’s study. And this ‘spatial reach’ to measure a cortical region to get the maximum summated activity will help us to choose a proper contact array size for each individual experimental situation.

4.2. Influence of cortical depth on LFP in different frequency ranges Frequency components are one interesting characteristic which bring more and more focus on. Laminar recordings in different animal models and neural dynamics analyzed in modelling studies suggested that slow band synchrony, like the alpha rhythm (< 15 Hz), derives from the excitatory pyramidal cells in deeper layer (layer 5 / 6), which was 1.1 mm below the cortical surface in the visual cortex of dog (Lopes Da Silva and Storm Van Leeuwen 1977). And fast band synchrony, i.e., gamma band rhythm (30 - 120 Hz), is generated by rhythmic recurrent synaptic excitation and inhibition in superficial layer (layer 3), which was 0.5 mm below the cortical surface in the visual cortex of rat (Oke et al 2010). Comparison of PSF profile between the experiment and simulation data in this part are in line with those previous studies. Results from the first simulation model gave a first impression that the PSF profile based on stimulation data in gamma band (45 - 400 Hz) match the profile with experiment data in the same frequency band (figure 12). Later on, simulation data with a refined model support this conclusion, that till to 80 Hz, PSF profile of simulation data had a best similarity compared with the experiment data in respective frequency band if the dipoles located in deeper layer, approximately 1.4 mm or deeper below the cortical surface. And up to 80 Hz, these dipoles position was shifted to more superficial layer and most of them around 0.7 mm below the cortical surface (figure 13 (b), (c)). Further analysis of HWHM from the PSF profiles with background adjusted simulated data also show the same tendency (figure 13(e)). This dividing frequency bin 80 Hz was consistent with that investigated in Oke’ study (2010). Results in the present study confirm the layer-specific roles of the LFP frequency components as suggested by previous studies, that the lower-frequency component should come from deeper cortical layer and higher-frequency activity might more originate from the superficial cortical layer. However, data obtained direct after the FEM simulation were frequency-independent, applied power data for the simulation part were all background adjusted by the experimental data as we described in the Method section, thus this frequency dependent cortical layer distribution is due to the extracellular medium itself is frequency dependent (Bédard et al 2004). To this step, combine with the discussion of LFP amplitude spatial reach, there might be also the ‘spatial reach’ for the LFP in different component. Is this hypothesis in a right direction, if yes, how this ‘spatial reach’ will behave at different cortical layer, which are all need further study concentrate on them.

132 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

Additionally, not only single-dipole simulation method was performed in the present study, different sizes of dipole cluster model (correlation inside these clusters were also set to ‘1’) were also tested in order to see the influence by the dipole source sizes. Results were similar as with the single-dipole condition and didn’t give us more clues. It might be due to the cluster radius is not in the right way. For the further study, we could reduce the radius size, or even with different correlation inside the cluster, to find out, does this population size influence the LFP generating, especially for the different frequency components.

4.3. Subdural BMI optimal spatial resolution One important goal for electrode designing for some clinical application, e.g., BMI, is to find out the optimal spatial resolution in order to obtain the maximum amount of brain activity information, either for the normal neural recording or for some clinical application, e.g., the most famous one brain-machine-interface (BMI), which is also one motivation of this simulation part in the present study. However, it is reported that the optimal spatial resolution for normal neural recording and for the BMI should be different (Rouse et al 2016), and they point that for monkey with single-contact decoding the optimal spacing is around 3 mm, and for human it must be larger than 3 mm due to the thicker dura matter. Indeed, they didn’t completely refuse the application with even smaller electrode array, performance with a smaller electrode array partly depend on the decoding strategy. In Slutzky’s study (2008, 2010) they also used finite model (FM) mathematic simulation method in both rat and human model, furthermore compared with the data recorded in rats under both aesthesia and awake conditions. They concluded that for rats, the optimal spacing for both epidural and subdural was around 0.7 mm (Slutzky et al 2008). In human, theoretically, the optimal spacing for subdural electrodes was reported to be 1.25 mm (Freeman et al 2000). With dipole centered at 0.9 mm below cortical surface and CSF layer of 0.2 mm thick, the threshold crossing by 90% attenuation to the maximal responses was examined as 2.6 mm subdurally with realistic human model, and this threshold crossing was increased with CSF thickness and also with the dipole depth from 0.9 - 1.4 mm below the cortical surface (Slutzky et al 2010). In fact, if we compare the research from Slutzky (2008, 2010) and Lindén (2011), we can find that they discuss the same thing with the same method but from different perspectives, one from the electrode contact side to ask how large the contact size should be in order to get maximum information (Slutzky et al 2008, 2010), one from neuroscience side to solve the problem how far away can brain signal propagate after generated from the source (Lindén et al 2011). Now if we put them together we can see that they have the same goal, that, only during the ‘spatial reach’ the contact has the possibility to obtain the maximum brain activity. In the present study, we used a minipig volume conduct head model combined the human tissue conductivities in the FEM simulation. Results were as expected consist with Slutzky’s study (2010), that with dipole located at 0.9 mm below the cortex surface and CSF thickness of 0.4 mm, the subdural spatial reach (5% of the maximum responses from the same cortical layer) was around 3.0 mm (figure 11(f)), which was reported as 2.6 mm with CSF of 0.2 mm thickness at 90% attenuation to the maximum responses in Slutzky’s study (2010) with electrode located subdurally. Meanwhile, this spatial reach was also broadened with increasing dipole depth (figure 11(f)). The difference might due to the FEM model itself (minipig head model combined with the human tissue conductivity parameters somehow not the best model situation) and the CSF thickness. In this part, dipole depth varied from 0.2 mm to 2.0 mm in 0.1 mm step, and the spatial reach for LFP attenuated to 5% of the maximal results were also changed from 2.4 mm to 3.9 mm based on the dipole-dipole distance at the same cortical depth, while for the reach to get the maximum summated activity varied from

133 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

3.2 mm to 4.5 mm at least for the single-dipole condition. Furthermore, HWHM of PSFs profile of the relative spectral power changes based on simulated data adjusted by the background activity of the experiment data were also computed, that till to 230 Hz, with dipole from superficial to deeper position, HWHM values were from 1.1 mm to 1.9 mm with single-dipole condition and are not larger than 2.5 mm even with different dipole cluster size (figure 13(d)), which make the full width at half maxima value is from 2.2 mm to 3.8 mm, which is consistent with the spatial reach analysis. Spatial reach to get the maximal summated activity is in a little bit larger region which could due to the definition itself. If we also set the distance range to the 95% maximal, resulting spatial reach might also in the same range. However, method to quantify this definition can be improved for the further study to get better results. And for the present stud, at least for the large animal model, the optimal spacing should between 2 - 4 mm to record the largest meaningful responses. Later with more precising model and parameters, the optimal spacing might even smaller.

4.4. Limitations of the simulation method in present study FEM model provides a realistic representation of the complicated head volume conductor which can let us investigate the electric propagation in the cortex (Fiederer 2012). Thus, a well segmented head model with suitable conductivity parameters are the preconditions to make the simulation work well. In the present study, white matter, gray matter, CSF, dura matter and blood are successfully segmented. Unfortunately, the conductivity of brain tissue has not yet been systematically investigated in pigs, much less in Göttingen minipigs. Thus, for the further more precision study, we could can put more effort to make the segmentation more in detail, like the pia matter or the arachnoidea structure, or compressing the CSF thickness which was proved to have influence on the optimal spacing range in Slutzky’ study (2010). Exactly conductivity parameters from the respected animal model is also important, which of course need related scientific studies, then maybe later utilizing a real human head model will help us learn more about the results. About the dipole model method, single one or dipole clusters with stronger correlation have been performed here. However, simulation with dipole clusters was performed based on the results of the single-dipole simulation. The subsequent mathematic computation seems too simple for the realistic. For further study, we can test with different correlation parameters for the dipoles in each dipole-cluster, even more, we can bring this idea directly in the simulation part, not just the simple mathematic computation based on the results of the single-dipole simulation

5. CONCLUSIONS AND OUTLOOKS To summarize the results presented in this study, simulations with the simple model confirm the previous finding from previous study and give a clue that the higher-frequency activity may come from the layer near the cortex surface. This result encourages us to refine the model and current dipole in order to reproduce the results from the simple model. However, the results after refining the model become more complex as we thought, but the tendency it still clear, i) spatial region to generate LFP is dependent on the cortical depth; ii) the current source of the lower-frequency activity should be in the deeper cortical layer and for higher- frequency activity it may locate in the superficial cortical layer. Reasons why the results become complex is still in process to be understood which are mainly due to the conductivity parameters assigned to the cortical tissues and the simplified mathematical relationship between the dipoles. In the future, proper head model with more precise conductivity

134 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings parameters is the most important step to be improved. Otherwise, testing with different current dipole model or another signal analyzing method, are all worth to improve the solution of the forward problem with µECoG electrode array. Meanwhile, simulations with regional refined model give a suggestion that the optimal subdural BMI spatial resolution should be smaller than 2 mm, at least for the minipig experiment.

ACKNOWLEDGMENTS CH is affiliated with the new spin-off company CorTec GmbH that was funded by a BMBF- grant of the Federal Republic of Germany and developed the μECoG electrode arrays used in this study. MG, TS, TB are scientific partners of CorTec GmbH and do not have any financial interest in or affiliation with CorTec GmbH. The remaining authors (XW, LF, IM) did and do not have any affiliation with CorTec. This work funded by the German Federal Ministry of Education and Research, BMBF (grants 01GQ0830 to BFNT Freiburg/Tübingen and 16SV5834 NASS) and the Deutsche Forschungsgemeinschaft, DFG, (grant EXC1086 BrainLinks-BrainTools). The authors thank Katharina Foerster and Prof. Joerg Haberstroh for maintaining the anesthesia in the animal experiments and Wolfgang Meier for doing the inter-connections of the µECoG electrodes.

REFERENCES Ball T, Kern M, Mutschler I, Aertsen A, Schulze-Bonhage A 2009 Signal quality of simultaneously recorded invasive and non-invasive EEG NeuroImage, 46: 708 - 16 Bédard C, Kröger H and Destexhe A 2004 Modeling extracellular field potentials and the frequency-filtering properties of extracellular space Biophys. J. 86: 1829 - 42 Brodmann K 1909 Vergleichende Lokalisationslehre der Grosshirnrinde in ihren Prinzipien dargestellt auf Grund des Zellenbaues Comparative Localization Studies in the Brain Cortex, its Fundamentals Represented on the Basis of its CellularArchitecture Published in Leipzig: Barth. Buzsáki G, Anastassiou C A and Koch C 2012 The origin of extracellular fields and currents--EEG, ECoG, LFP and spikes Nat. Rev. Neurosci., 13(6): 407 - 20 Fiederer L D J, Lahr J, Vorwerk F L, Aertsen A, Wolters C H, Schulze-Bonhage A and Ball T 2016 Electrical Stimulation of the Human Cerebral Cortex by Extracranial Muscle Activity: Effect Quantification With Intracranial EEG and FEM Simulations IEEE Trans Biomed Eng., 63(12): 2552 - 63 Fiederer L D J 2012 Diploma Thesis: Multiscale Electrophysiological Neuroimaging: High-Resolution Forward Modeling, University of Freiburg, Freiburg, Germany Freeman W J, Rogers L J, Holmes M D and Silbergeld D L 2000 Spatial spectral analysis of human electrocorticograms including the alpha and gamma bands J. Neurosci. Methods 95: 111 - 21 Hallez H, Vanrumste B, Grech R, Muscat J, De Clercq W, Vergult A, D'Asseler Y, Camilleri K P, Fabri S G, Van Huffel S and Lemahieu I 2007 Review on solving the forward problem in EEG source analysis J. Neuroeng. Rehabil., 4: 46 Hamalainen M S and Sarvas J 1989 Realistic conductivity geometry model of the human head for interpretation of neuromagnetic data IEEE Trans. Biomed. Eng., 36: 165 - 71 Haueisen J, Ramon C, Czapski P and Eiselt M 1995 On the influence of volume currents and extended sources on neuromagnetic fields: a simulation study Ann. Biomed. Eng., 23: 728 - 39 Haueisen J, Ramon C, Eiselt M, Brauer H and Nowak H 1997 Influence of tissue resistivities on neuromagnetic fields and electric potentials studied with a finite element model of the head IEEE Trans. Biomed. Eng., 44: 727 – 35 Henle C, Meier W, Schuettler M, Boretius T and Stieglitz T 2009 Electrical Characterization of Platinum, Stainless Steel and Platinum/Iridium as Electrode Materials for a New Neural Interface World Congress on Medical Physics and Biomedical Engineering, Munich, IFMBE Proceedings, 25(9): 100 - 3 Kandel E R, Schwartz J H and Jessell T M 2000 Principles of neural science 4th ed, New York: McGraw-hill. Lindén H, Tetzlaff T, Potjans T C, Pettersen K H, Grün S, Diesmann M and Einevoll G T 2011 Modeling the Spatial Reach of the LFP Neuron, 72: 859 - 72

135 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings

Logothetis N K 2003 The underpinning of the BOLD functional magnetic resonance imaging signal J. Neurosci., 23 (10): 3963 - 71 Lopes Da Silva F H and Storm Van Leeuwen W 1977 The cortical source of the alpha rhythm Neurosci. Lett., 6: 237 - 41 Maier A, Adams G K, Aura C and Leopold D A 2010 Distinct superficial and deep laminar domains of activity in the visual cortex during rest and stimulation Front. Syst. Neurosci., 4: 31 Manola L, Roelofsen B H, Holsheimer J, Marani E and Geelen J 2005 Modelling motor cortex stimulation for chronic pain control: electrical potential field, activating functions and responses of simple nerve fibre models Med. Biol. Eng. Comput., 43, 335 - 43 Nicholson C and Llinas R 1971 Field potentials in the alligator cerebellum and theory of their relationship to Purkinje cell dendritic spikes J. Neurophysiol., 34(4): 509 - 31 Nunez P L 1981 A Study of Origins of the Time Dependencies of Scalp EEG: II-Experimental Support of Theory IEEE Trans. Biomed. Eng., 28: 281 - 8 Nunez P L and Srinivasan R 2006 A theoretical basis for standing and traveling brain waves measured with human EEG with implications for an integrated consciousness Clin. Neurophysiol. Off. J. Int. Fed. Clin.Neurophysiol.,117: 2424 - 35 Oke O O, Magony A, Anver H, Ward P D, Jiruska P, Jefferys J G R and Vreugdenhil M 2010 High-frequency gamma oscillations coexist with low-frequency gamma oscillations in the rat visual cortex in vitro Eur. J. Neurosci., 31: 1435 - 45 Rouse A G, Williams J J, Wheeler J J and Moran D W 2016 Spatial co-adaptation of cortical control columns in a micro-ECoG brain-computer interface J. Neural. Eng., 13(5): 056018 Shipp S 2007 Structure and function of the cerebral cortex Curr. Biol., 17(12): R443 - 9 Slutzky M W, Jordan L R and Miller L E 2008 Optimal spatial resolution of epidural and subdural electrode arrays for brain-machine interface applications Conf. Proc. IEEE Eng. Med. Biol. Soc., : 3771 - 4 Slutzky M W, Jordan L R, Krieg T, Chen M, Mogul D J and Miller L E 2010 Optimal Spacing of Surface Electrode Arrays for Brain Machine Interface Applications J. Neural Eng., 7: 26004 Smith S M 2002 Fast robust automated brain extraction Hum. Brain Mapp., 17: 143 - 55 Smith S M, Jenkinson M, Woolrich M W, Beckmann C F, Behrens T E J, Johansen-Berg H, Bannister P R, De Luca M, Drobnjak I, Flitney D E, Niazy R K, Saunders J, Vickers J, Zhang Y, De Stefano N, Brady J M and Matthews P M 2004 Advances in functional and structural MR image analysis and implementation as FSL NeuroImage, 23(Suppl. 1): S208 - S219 Spaak E, Bonnefond M, Maier A, Leopold D A and Jensen O 2012 Layer-Specific Entrainment of Gamma-Band Neural Activity by the Alpha Rhythm in Monkey Visual Cortex Current Biology, 22(24): 2313 - 18 Spyrakos, C.C., 1996. Finite element modeling in engineering practice: includes examples with ALGOR Published by AlgorInc, Pittsburgh. Wagner, S., Lucka, F., Vorwerk, J., Herrmann, C., Nolte, G., Burger, M., Wolters, C.H., 2016.Using reciprocity for relating the simulation of transcranial current stimulation to the EEG forward problem. BioRxiv, 043554 Wang X, Gkogkidis C A, Ilijina O, Fiederer L D J, Henle C, Mader I, Kaminsky J, Stieglitz T, Gierthmuehlen M and Ball T accepted in J. Neural Eng. in Jun. 2017 Mapping the fine structure of cortical activity with different µECoG electrode array geometries Wolters C H, Grasedyck L and Hackbusch W 2004 Efficient computation of lead field bases and influence matrix for the FEM-based EEG and MEG inverse problem Inverse Problems, 20: 1099 - 116 Wolters C H 2007a The finite element method in EEG/MEG source analysis Siam News, 40: 1 - 2 Wolters C H, Köstler H, Möller C, Härdtlein J and Anwander A 2007b Numerical approaches for dipole modeling in finite element method based source analysis Int. Congr. Ser., New Frontiers in Biomagnetism Proceedings of the 15th International Conference on Biomagnetism, Vancouver, BC, Canada, 1300: 189 - 92 Woolrich M W, Jbabdi S, Patenaude B, Chappell M, Makni S, Behrens T, Beckmann C, Jenkinson M and Smith S M 2009 Bayesian analysis of neuroimaging data in FSL NeuroImage, 45: S173 - S186 Yan Y, Nunez P L and Hart R T 1991 Finite-element model of the human head: scalp potentials due to dipole sources Med. Biol. Eng. Comput. 29(5): 475 - 81 Zhang Y, Brady M and Smith S 2001 Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm IEEE Trans. Med. Imaging, 20: 45 - 57

136 Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings — Supplements

Table S1: Literature survey of the layer specificity of neuronal frequency bands. Publications are sorted by ascending publication date. Courtesy of Xi 137 Wang. Author Subjects Task Results

visual cortex, Lopes Da Silva intracortical eye-close the phase reversal (180 °) of alpha rhythms (11.2 – 13.1 Hz) was and Storm Van 3 dogs electrodes and anesthesia induced alpha found at about 1100 µm below the cortical surface Leeuwen, 1977 EEG rhythm Silva et al., sensorimotor many pyramidal neurons in layer 5 of the neocortex showed rats slice recording 1991 cortex prolonged, 5- to 12- Hz rhythmic firing patterns at threshold neocortical neurons in layers 2/3 (15 – 25 Hz) and layer 5 (1 – 5 somatosensory Flint et al., 1996 rats slice recording Hz) can independently generate two distinct forms of rhythmic cortex population activity auditory areas in superficial layers II/III, a gamma rhythm was observed Roopun et al., and secondary (frequency 37.5±4.5 Hz), whereas in deep layers (V and VI), a rats slice recording 2006 somatosensory beta frequency rhythm was observed (frequency 25.4±3.2 Hz), cortical areas and in layer IV, both rhythms were observed to coexist primary visual low-frequency (~2 Hz) activity is widely observed in layer 2/3 Sun and Dan, cortex(V1), rats slice recording (L2/3), a narrow-band fast oscillation (10 – 15 Hz) is prominent 2009 spontaneous in layer 5 (L5) activity visual cortex, fast-γ (46-80 Hz) was associated with rhythmic current sink- intracortical medicine source sequences in layer III (0.5 mm from pia surface) and Oke et al., 2010 rats anesthesia electrodes induced gamma slow-γ (20-45 Hz) with rhythmic current sink source sequences oscillations in layer V (1.2 mm from pia surface) primary visual cortex (V1), Maier et al., intracortical spontaneous higher gamma power (30 – 100 Hz) in the superficial layers than 2 monkeys awake 2010 electrodes neural activity in the deep layers was observed and visual stimulation V1, selective multiple, separable alpha (7 – 13 Hz) current generators were Bollimunta et intracortical attention task 2 monkeys awake identified in SG, G, and IG layers, with the layer 4C and layer 6 al., 2011 recording (visual vs generators being consistently the strongest across penetrations auditory) Cortical-Depth and Frequency-Band Dependency of the Spatial Extent of the Generators Underlying µECoG Recordings — Supplements

spike-field coherence in the gamma (40–60 Hz) frequency range Buffalo, Fries P intracortical visual cortex, was largely confined to the superficial layers, whereas the deep 2 monkeys awake et al 2011 electrodes motor task layers showed maximal coherence at low frequencies (6–16 Hz), which included the alpha range LFP power, which was concentrated in the γ-band (20–60 Hz), Xing et al., intracortical 9 monkeys anesthesia V1, visual task was greatest at the cortical depth corresponding to 2012 recording corticocortical output layers 2, 3, and 4B γ-waves (40 – 90 Hz) are initiated in input layer 4 and propagate van Kerkoerle intracortical texture 6 monkeys awake to the deep and superficial layers of cortex, whereas α-waves (5 et al., 2014 recording segregation task – 15 Hz) propagate in the opposite direction during normal visual stimulation scene information peaks in Muckli et al., healthy human visual fMRI awake mid-layers. Conversely, we found that contextual feedback 2015 subjects stimulation task information peaks in outer, superficial layers. influences along feedforward projections predominate in the Michalareas, gamma band (40 – 75 Hz) which usually originate from 43 healthy fixation visual Fries P, et al., MEG/MRI awake superficial layers, whereas influences along feedback human subjects task 2016 projections predominate in the alpha-beta band (7 – 17 Hz) which originate from infragranular (deeper) layers γ-band EEG power (40 – 100 Hz) correlates positively with the Scheeringa et 34 healthy fixation visual superficial layers’ BOLD signal and that β-power (15 – 30 Hz) EEG-fMRI awake al., 2016 human subjects task is negatively correlated to deep layer BOLD and α-power (8 – 12 Hz) to both deep and superficial layer BOLD A current-source density analysis reveals top-down inputs in the V1, attention van Kerkoerle intracortical superficial layers and layer 5, and an increase in neuronal firing 2 monkeys awake demanding et al., 2017 recording rates most pronounced in the superficial and deep layers and tracing task weaker in input layer 4. 138