HOW TO UNDERSTAND IT Pract Neurol: first published as 10.1136/practneurol-2013-000768 on 19 March 2014. Downloaded from Magnetoencephalography Malcolm Proudfoot,1,2 Mark W Woolrich,2 Anna C Nobre,2 Martin R Turner1 1Nuffield Department of Clinical INTRODUCTION Neurosciences, University of The understanding of brain function is Box 1 An introduction to neuronal Oxford, UK 2Oxford Centre for Human Brain moving rapidly towards a systems-level, oscillations Activity, University of Oxford, UK network-based approach. It is now as naive to talk simplistically about what a particu- ▸ Neuronal oscillatory activity is continu- Correspondence to ‘ ’ Dr Martin Turner, Nuffield lar area of the brain does , as it was for ous, but fluctuations in power and Department of Clinical Franz Joseph Gall (1758–1828) to link timing allow rapid alteration in com- Neurosciences, West Wing its performance to the thickness of the munication strength within existing Level 3, John Radcliffe Hospital, overlying skull. Magnetoencephalography structural network architecture, far Oxford OX3 9DU, UK; 1 [email protected] (MEG) is a rapidly developing and unique faster than synaptic modification. tool for the study of brain function, in ▸ Two distinct cerebral regions can facili- Published Online First particular the underlying oscillations in tate preferential information exchange 19 March 2014 neuronal activity that appear to be funda- by synchronising their rhythmic behav- mental (box 1), with real-time resolution iour; the γ band (40–80 Hz), in particu- and potential for application across a range lar, facilitates this process, but is also of brain disorders. We provide a brief modulated ‘top-down’ by lower fre- overview of the technology, broad quencies such as θ (4–7 Hz), reflecting approaches to data analysis, and aspirations factors, such as arousal states. for its application to the study of ▸ α Rhythms (8–13 Hz), so prominent in neurodegeneration. the occipital cortex upon eye closure, reflect more than just an ‘idling’ rhythm FUNCTIONAL BRAIN IMAGING SO FAR but also contribute to active allocation Structural MR imaging of the brain and of attentional resources and suppress 2 spinal cord has revolutionised the accur- irrelevant sensory information. acy of diagnosis in common conditions, ▸ The influential theory ‘Communication such as stroke, and greatly expanded the through Coherence’ developed by 3 http://pn.bmj.com/ taxonomy of neurological disorders. Fries, builds on existing models of Advanced applications of MRI now allow ‘binding by synchronisation’ that may the assessment of white matter tract underpin selective attention, a key func- integrity (diffusion-tensor imaging), tion in prioritising neural events to 4 regional grey matter volume (voxel-based guide awareness and action. morphometry), and cortical thickness (surface-based morphometry). These on September 28, 2021 by guest. Protected copyright. techniques enable the non-invasive and transmissions and physical distance rapid quantification of structure at a covered by brain activity within this time given time point, but it is clear that the frame, it is quickly appreciated that this brain cannot be understood in terms of technology is limited in its ability to deliver a systems-level understanding of Open Access structure alone. Functional MRI (fMRI), Scan to access more brain function if used in isolation. free content based on blood oxygen-level-dependent (BOLD) image contrast, can achieve almost submillimetre accuracy in the THE UNIQUE ADVANTAGES OF MEG spatial localisation of neuronal activity. Ever since the pioneering EEG recordings However, this relatively high spatial reso- made by German neurologist Hans lution is not matched in temporal accur- Berger (1873–1941), it has been possible acy, in essence because the relatively slow to identify the self-generated oscillatory speed of haemodynamic changes in activity of neuronal ensembles and cat- To cite: Proudfoot M, response to neuronal activity fundamen- egorise frequency bands with increasing Woolrich MW, Nobre AC, tally limits the temporal detail that can be accuracy. Electrical potential changes et al. Pract Neurol extracted to a timescale of seconds. related to brain activity measured at the 2014;14:336–343. When one considers the multiple synaptic scalp by EEG are fundamentally limited 336 Proudfoot M, et al. Pract Neurol 2014;14:336–343. doi:10.1136/practneurol-2013-000768 HOW TO UNDERSTAND IT Pract Neurol: first published as 10.1136/practneurol-2013-000768 on 19 March 2014. Downloaded from by the distortive effects of the intervening structures, which severely hamper efforts to localise the signal source precisely. MEG, instead, measures the magnetic field changes induced by intracellular current flow, the generation of which obeys the ‘right-hand rule’ in the application of Ampère’s law. Unlike EEG measures, these pass through dura, skull and scalp relatively unaltered. The technique, therefore, offers a safe, non-invasive method to ‘listen’ in to brain activity at rest and during simple tasks, which from the subject’s perspective, despite measuring at several hundred channels, is painless and quick to set up. Mathematical modelling of these data then enables localisation of sources while uniquely maintaining sampling frequencies up to several thousand times per second. Compared to fMRI’s temporal resolution of, at best, several hundred milliseconds, MEG can resolve events with millisecond precision. HOW IT WORKS The neuronal activity captured by MEG is not, as perhaps expected, generated by the (too brief) axonal action potentials of pyramidal cells, but rather by the net contributions of excitatory and inhibitory den- dritic postsynaptic potentials. This current flow through the apical dendrites (represented as a Figure 1 The organisation of cortical microcolumns within the ‘dipole’) generates a magnetic field that projects radi- sulcal bank, tangentially orientated to the skull, allows their ally; thus, MEG excels at detecting dipoles arranged detection with magnetoencephalography since their induced in a tangential orientation to the skull. Fortunately, magnetic fields will project beyond the skull surface. Conversely, the extensively folded sulci of the human cortex apical dendrites orientated perpendicularly to the skull, as promote that orientation for the majority of cortical found at the gyral crown, are better detected by EEG. (From Hansen et al5). microcolumns (figure 1). However, MEG is less sensi- tive to deeper (including subcortical) sources, as mag- netic field change decreases rapidly with distance. developed by his collaborator James Zimmerman. At Compared with a standard clinical MR scanner very low temperatures, SQUIDs are extremely sensi- http://pn.bmj.com/ magnet strength of 1.5 Tesla, the strength of the tive to magnetic field change, which can be recorded signals detected by MEG are 1014 orders smaller and converted into digital signal (‘quantisation’). (figure 2). It has been compared with hearing a pin Sensor arrays have evolved to provide whole-head drop at a rock concert. The smallest measurable mag- coverage via a helmet containing more than 300 netic field changes are thought to be produced by sim- ultaneously active arrays of approximately 50 000 on September 28, 2021 by guest. Protected copyright. pyramidal cells, which in theory covers a cortical surface area of 0.9 mm diameter. It is increasingly recognised that modulation of self-generated oscilla- tory activity is a principal mechanism by which geo- graphically distant network regions interact,6 thus, a brain-wide imaging technique with high temporal sen- sitivity is a prerequisite for interrogation. The ability to detect endogenously generated mag- netic fields was realised in the 1960s by physicist David Cohen at Massachusetts Institute of Technology, who furthered the then recent discovery of ‘magnetocardiography’ by applying a magnetically shielded room to remove the overwhelming noise of the Earth’s magnetic field (figure 3). He could then measure the even smaller magnetoencephalographic Figure 2 Magnetic field strength density measured in signal by making use of superconducting loops super- femtotesla (fT), highlighting the exquisite sensitivity of the conducting quantum interference device (SQUIDs), SQUIDs used in magnetoencephalography. Proudfoot M, et al. Pract Neurol 2014;14:336–343. doi:10.1136/practneurol-2013-000768 337 HOW TO UNDERSTAND IT Pract Neurol: first published as 10.1136/practneurol-2013-000768 on 19 March 2014. Downloaded from THE ACQUISITION The sensor design itself has evolved to meet some of the localisation challenges by using more than one pick-up coil in series. A single ‘magnetometer’ coil measures any orthogonal magnetic field. Pairs of coils place closed together and wound in opposite direc- tions are also used to measure gradients in the mag- netic field over space. These ‘gradiometers’ are particularly sensitive to a gradient in magnetism from nearby (ie, neuronal) sources, but subtract out signal from distant external (and thus artefactual) sources, as these appear similar to both coils. Electromagnetic signals are also generated by move- Figure 3 After 4 s of raw magnetoencephalography data (two ment of the head or eyes (including blinking), skeletal channels contain obvious artefacts), the door to the and cardiac muscle electromagnetic activity. The magnetically shielded room is opened during recording. The sensors, therefore, also pick up this physiological interference caused by external