Quantitative Trends and Topology in Developing Functional Brain Networks

Home , Paul Lauterbur

A dissertation submitted to the

Graduate School

of the University of Cincinnati

in partial fulfillment of the

requirements for the degree of

Doctor of Philosophy

in the Department of Physics

of the College of Arts and Sciences

Elveda Gozdas

M.S. University of Cincinnati, August 2013

July 2018

Committee Chairs: Scott K. Holland, Ph.D.

Rohana Wijewardhana, Ph.D.

Abstract

With the advances in MRI, it has become possible to noninvasively observe function and structure of the developing brain in vivo. Functional magnetic resonance imaging (fMRI) of the brain is a non-invasive way to assess brain function using MRI signal changes associated with neuronal activity. The most widely used method is based on BOLD (Blood Oxygenation Level

Dependent) signal changes caused by hemodynamic and metabolic neuronal responses.

Functional connectivity has been defined as inter-regional temporal correlations among spontaneous BOLD fluctuations in different regions of the brain during a task as well as when the brain is idle. By identifying brain regions that exhibit highly correlated BOLD signal fluctuations, we can infer that the regions are functionally connected and co-activation during a particular task or at rest (fcMRI) suggests that these regions work together as part of a functional brain network. This method is now being used widely to study brain networks but has seen limited use in studies of the developing brain, particularly in infants. Functionally connected brain regions can be specified as components of integrated networks that enable specific sensory or cognitive brain functions. These brain networks demonstrate the basic connectivity pattern between brain regions, which can be represented mathematically using graph-theoretical approaches. Graph theory provides a convenient quantitative and visual format to sketch the topological organization of brain connectivity representing complex brain networks. Graph theory analysis also naturally provides quantitative descriptors of both global and regional topological properties of brain graphs. While this approach is now widely used with functional

MRI data as a means of studying the topology of functional brain networks, it has not been applied to study the development of brain networks from birth, nor in the premature infant brain.

The main goal of this dissertation is to use novel functional connectivity acquisition and analysis methods designed for infants and children to examine and better understand the effects of brain injury, prematurity and normal aging on functional brain networks during development.

The work presented here makes significant methodological and scientific contributions to our understanding of developing brain networks with the following discoveries: 1) Infants with brain injury sustained in the perinatal period show evidence of decreased brain activity and functional connectivity during visual stimulation compared to healthy, full-term neonates; 2) Resting-state networks are already established early in development in very preterm infants but functional connectivity and network characteristics in preterm infants differ from those of full-term infants by term-equivalent age; 3) Developmental trajectories of the functional brain connectome in normal healthy children exhibit significant sex-related differences, with different rates of maturation of functional brain networks in boys and girls; 4) Differences in cognitive ability during childhood are supported by differences in regional network topology during brain development; continuing into late adolescence.

Acknowledgements

First and foremost I would like to thank my enthusiastic advisor Prof. Scott K. Holland. It has been an honor to be his Ph.D. student. He has taught me continuously with his patience, motivation and immense knowledge to make my Ph.D. research productive and stimulating. His guidance helped me all the time of my research, even during tough times of my Ph.D. and writing of this thesis. I am also grateful for the excellent example he has provided as a successful scientist and professor. I could not imagine having a better supervisor for my Ph.D. research.

I also would like to thank Prof. Jason Woods for being my academic advisor for the last year. I will always be thankful for your support and kindness. Besides my advisors, I would like to thank the rest of my thesis committee Prof. Rohana Wijewardhana, Prof. Howard Jackson, Dr.

Jean Tkach and Dr. Stephanie Merhar for their time, insightful comments and encouragements.

Also, I would like to thank Dr. Nehal Parikh and Dr. Lili He. The preterm study discussed in this dissertation would not have been possible without their collaboration.

Last but not least, I would like to thank my friends and family for providing me continuous encouragement and support throughout my years of research. Most importantly, I would like to thank my loving, encouraging, patient and supportive husband Okkes whose endless inspiration during all stages of this Ph.D. is so appreciated.

Elveda Gozdas

iii

Table of Contents

List of Figures ...... viii

List of Figures ...... xi

CHAPTER 1: INTRODUCTION ...... 1 1.1 Advances in Functional Magnetic Resonance Imaging ...... 1

1.2 Application of Advanced fMRI Methods in Children in Cincinnati ...... 3

1.3 Bridging the gap in understanding brain development from infancy to adulthood ...... 5

1.4 Hidden Contributions of this Dissertation ...... 5

1.5 Unique contributions of this dissertation ...... 7

CHAPTER 2: THE BASIC PRINCIPLES OF MAGNETIC RESONANCE IMAGING .... 9 2.1 The Quantum Mechanics of MRI ...... 10

2.2 The Classical Description of MRI ...... 14

2.3 Relaxation ...... 16

2.3.1 Introduction of Spin-Lattice (�) ...... 17

2.3.2 Introduction of Spin-Spin (�) ...... 17

∗ 2.3.3 Introduction of � ...... 18

2.4 Bloch Equation with Relaxation Terms ...... 18

2.5 Signal and Spin Density ...... 19

2.6 Spatial Encoding in MRI ...... 20

2.6.1 Slice Selection ...... 20

2.6.2 Phase Encoding ...... 21

2.6.3 Frequency Encoding ...... 22

2.7 Echo Planar Imaging ...... 22

2.7.1 k-space coverage ...... 23

CHAPTER 3: PRINCIPLES OF FUNCTIONAL MAGNETIC RESONANCE IMAGING

...... 25

3.1 Brain Functional MRI and BOLD ...... 25

3.2 fMRI Experimental Designs ...... 27

3.3 General Linear Model (GLM) ...... 28

3.5 Functional Connectivity of fMRI ...... 29

3.6 Graph Theory Analysis ...... 30

CHAPTER 4: FUNCTIONAL CONNECTIVITY OF THE VISUAL SYSTEM IN

INFANTS WITH PERINATAL BRAIN INJURY ...... 34

4.1 Introduction ...... 35

4.2 Methods ...... 36

4.2.1 Participants ...... 36

4.2.2 MRI Acquisition Parameters ...... 36

4.2.3 Visual Task Paradigm ...... 37

4.2.4 Data Analysis ...... 37

4.3 Results ...... 39

4.3.1 Participants ...... 39

4.3.2 Visual Task fMRI ...... 39

4.3.3 Functional Connectivity ...... 40

4.4 Discussion ...... 40

4.4 Conclusions ...... 43

CHAPTER 5: ALTERED FUNCTIONAL NETWORK CONNECTIVITY IN PRETERM

INFANTS: ANTECEDENTS OF COGNITIVE AND MOTOR IMPAIRMENTS? ...... 44

5.1 Introduction ...... 45

5.2 Methods ...... 47

5.2.1 Participants ...... 47

5.2.2 MRI acquisition ...... 48

5.2.3 Data analysis ...... 50

5.2.4 ROI-to-ROI Functional Connectivity ...... 51

5.2.5 Graph theory processing ...... 52

5.3 Results ...... 53

5.4 Discussion ...... 61

5.5 Conclusions ...... 66

CHAPTER 6: DEVELOPMENTAL CHANGES IN FUNCTIONAL BRAIN NETWORKS

FROM BIRTH THROUGH ADOLESCENCE ...... 67

6.1 Introduction ...... 67

6.2 Methods ...... 69

6.2.1 Participants ...... 69

6.2.2 MRI acquisition ...... 70

6.2.3 Data analysis ...... 71

6.2.3.1 Preprocessing ...... 71

6.2.3.2 Functional Brain Networks Construction ...... 72

6.2.3.3 ROI-to-ROI Functional Connectivity ...... 73

6.2.3.4 Correlation of Network Topology with Age, Sex and Behavioral Measures ...... 74

6.2.3.5 Statistical Analysis ...... 74

6.3 Results ...... 75

6.3.1 Age and sex effects on global network properties ...... 75

6.3.2 Neurocognitive measures and sex effects ...... 78

6.4 Discussion ...... 80

CHAPTER 7: CONCLUSIONS AND FUTURE RESEARCH ...... 85

7.1 Conclusions ...... 85

7.2 Future Research Directions ...... 87

BIBLIOGRAPHY ...... 92

vii

List of Figures

Figure 1.1 Number of publication for fMRI studies

(https://f1000research.com/articles/3-313/v2) ...... 1

Figure 1.2 (a) fMRI brain activation map for verb generation task. Light blue color represents greater activation(p<=0.01) (b) fMRI brain activation age correlation map for verb generation

(p<0.05) (Holland et al., 2001) ...... 3

Figure 1.3 Sex effect on Story Processing Syntactic Prosody task activation. Colored regions show greater activation in girls than boys (p<0.05) (Plante et al., 2006) ...... 4

Figure 2.1 The Zeeman energy levels for a spin one-half system. The spin is parallel to the external field B in the lower energy state. The wavy vertical line represents a transition from the higher to the lower state by photon emission...... 12

Figure 2.2 Precession of the magnetization vector. (a) ) In the laboratory frame under the influence of B and B, when w= w the magnetization simultaneously precesses about B at w w and about B at w . (b) In the rotating frame when B = γ (c) In rotating frame the magnetization is precessing about B and the offset frequency is ∆w= γ B-w...... 16

Figure 2.3 (a) EPI sequence diagram (b) k-space coverage in EPI ...... 24

Figure 3.1 (a) Neuronal activity change is associated with the change of relative amount of oxygenated blood (https://www.nature.com/scitable/blog/brain- metrics/what_does_fmri_measure) (b) Hemodynamic response function ...... 26

Figure 3.2 (a) Block design and (b) event related design schematic representation ...... 27

Figure 4.1. Z-score maps of brain activation during the visual fMRI task. The top panel (a) shows controls, the middle panel (b) shows infants with brain injury, and the bottom panel (c) shows the difference between the two. Red areas represent regions of activation during the visual

viii task. Z maps are thresholded using clusters determined by Z > 2.4 and a false-discovery rate corrected significance threshold of P = 0.05 ...... 40

Figure 5.1. Group differences between preterm and full-term neonate for ROI-to-ROI functional connectivity with motor area seed regions. Black circles represent the regions that reached the group level difference in A-C) Superior frontal gyrus (medial), B-D) Superior frontal gyrus

(dorsal), E) Supplementary motor area. Red colors indicate ROIs where the full-term group exhibit greater functional connectivity than the preterm group while blue lines indicate increased functional connectivity to motor area seed ROI for the preterm group relative to the full-term group. Connectivity differences are shown at a threshold of p<0.05 corrected for multiple comparisons using false discovery rate...... 57

Figure 5.2. Group differences between preterm and full-term neonate for ROI-to-ROI functional connectivity with language area seed regions. Black circles represent the regions that reached the group level difference in A) Left Middle Temporal Gyrus, B) Left Angular Gyrus, and C) Right

Fusiform Gyrus. Red colors indicate ROIs where the full-term group exhibit greater functional connectivity than the preterm group while blue lines indicate increased functional connectivity to language area seed ROI for the preterm group relative to the full-term group. Connectivity differences are shown at a threshold of p<0.05 corrected for multiple comparisons using false discovery rate...... 57

Figure 5.3. Group differences between preterm and full-term neonate for ROI-to-ROI functional connectivity with cognitive, executive function area seed regions in: A) Left Middle frontal gyrus and Orbitofrontal cortex, B) Right Middle frontal gyrus and Orbitofrontal cortex .Black circles represent the regions that reached the group level difference. Red colors represent connection where the full-term group had greater functional connectivity and blue lines represent

ix the preterm group had greater functional connectivity with a threshold of p<0.05 corrected for multiple comparisons using false discovery rate...... 58

Figure 5.4. Comparison of global network measures between preterm and full term born infants plotted as a function of the cost threshold or rich club level (k) used to estimate the parameter:

A) Small-Worldness, B) Modularity C) Rich-club Coefficient D) Assortativity. (*) indicates the difference is significant at p<0.05 corrected for multiple comparisons using false discovery rate.

...... 60

Figure 5.5. Group difference maps. Red indicates that the full-term group had greater Global

Efficiency (A) or Degree (B) compared with the preterm group while blue colors represent the converse with the preterm group exhibiting increases relative to the full-term babies. A threshold of p<0.05 was used for all maps, with correction for multiple comparisons using false discovery rate...... 61

Figure 6.1- Graphs showing racial, ethnic and income distribution of participants ...... 71

Figure 6.2- Global network measures as a function of age. (A) Global Efficiency. (B) Local

Efficiency. (C) Small-Worldness. (D) Modularity...... 76

Figure 6.3- Sex difference in brain network topology with age. . (A) Global Efficiency. (B)

Local Efficiency. (C) Small-Worldness. (D) Modularity ...... 77

Figure 6.4 – Rich-club trends in the developing connetome. (A) Hub regions and connections.

The black dots represent the hub regions. (B) Age-related change in normalized rich-club coefficient ...... 78

Figure 6.5 – Local Network Measures as a function of IQ ...... 79

Figure 6.6 -Local Network Measures as a function of Expressive Vocabulary Test scores in inferior frontal gyrus (IFG) and posterior superior temporal gyrus (pSTG) ...... 80

List of Tables

Table 2.1 List of selected nuclear species with their spins, magnetic moments, gyromagnetic ratios (γ) and their abundance in human body ...... 1

Table 5.1 All regions with significant group differences in functional connectivity are listed by anatomical area, along with the t-value and p-value for the between group difference and MNI coordinates of the center of each ROI. These regions were used as seeds (indicated by black circles in Figures 1-3) for the ROI-based connectivity analysis ...... 56

Table 5.2 Regions with significant group differences in regional graph theory measures. A)

Global Efficiency B) Degree. Regions are listed by anatomical designation along with relevant statistical measures between groups and MNI coordinates at the center of each RO ...... 61

Chapter 1: Introduction

1.1 Advances in Functional Magnetic Resonance Imaging

Functional magnetic resonance imaging (fMRI) detects signal changes associated with blood oxygenation level dependent (BOLD) changes in MRI signal intensity that occur when the brain is activated by a task or by spontaneous network fluctuations. For the past two decades, fMRI has emerged as one of the most commonly used imaging modalities for non-invasive mapping of brain functions, with thousands of publication in Pubmed now appearing every year (Figure 1.1).

Current research utilizes fMRI to study normal brain functions as well as to discover functional brain abnormalities in the injured and diseased brain. Initially, fMRI was widely performed using an explicit task to stimulate and examine functional brain networks. More recently, resting state fMRI (rs-fMRI) has been employed effectively to investigate spontaneous neuronal fluctuations of the brain which can be measured in BOLD signals using fMRI when subjects are at rest, with no task to stimulate the brain (Mateo et al., 2017, Biswal et al.,

1995).

Functional connectivity can be derived from the resting state fMRI signal by computing the temporal correlation in the fMRI signals between spatially separated brain regions (Friston,

2002). Seed-based and independent component analysis (ICA) are the most widely used methods for functional connectivity analysis. The seed-based approach computes the cross-correlation 1 between voxels in predefined regions of interest (ROIs), or seed regions; applying a threshold for significance (Biswal et al., 1997). ICA is a data-driven, multivariate approach that provides spatially independent components that share the same temporal fluctuations in the BOLD signal.

Brain activity patterns derived from ICA maps thus provide a natural measure of functional connectivity between voxels within each component (Calhoun et al., 2009, Schmithorst and

Holland, 2004).

The architecture of human brain functional networks, now known as the human connectome, can be studied using rs-fMRI and this has spawned countless, large scale studies in adults and in children. Simultaneous with the explosion of rs-fMRI for studies of the human connectome, graph theory has emerged as a new approach to investigate the complex functional network topology of the human brain (Robinov and Sporns, 2010). Application of graph theory to functional brain networks has revealed important topological properties that influence human behavior including. Efficient network architecture can be described mathematically and heuristically by several topological properties represent small-worldness, which reflects an optimal balance between segregation and integration in information processing between regions

(Salvador et al., 2005, Achard et al., 2007), modular structure, central communication hubs and rich-club organization which is formed by the densely interconnected hubs in healthy adults (van den Heuvel et al., 2011, van den Heuvel et al., 2012, Cao et al., 2014 ). The convergence of advances in MRI scanner technology that yields high quality fMRI data and advanced mathematical approaches to describing topological and information processing capabilities of networks provides an opportunity to explore the human brain in new ways that have never been possible before.

1.2 Application of Advanced fMRI Methods in Children in Cincinnati fMRI has been applied to study brain functions of a wide variety of cognitive processes in adults.

As this dissertation project was being conceived, it was apparent that a significant gap existed between the use of fMRI in adults and it application pediatric populations. The first fMRI study in children was published by Cincinnati Children’s Medical Center (CCHMC) using a verb generation task (Holland et al., 1998, Holland et al., 2001). The research group at CCHMC has made several significant contributions to the literature for studying brain development using fMRI and other neuroimaging methods. First, the verb generation task was successfully applied to children at the age of 5-18, demonstrating on a gross scale that brain activation of this task was lateralized in the left hemisphere and an increasing left lateralization with age in a cross- sectional sample ranging in age from 5 to 18 years (Holland et al., 2001, Szaflarski et al., 2006) as shown in Figure 1.2.

This task was also applied to children longitudinally during the age span of 5-11 years and age- related changes in neuroplasticity of language were explored. The pattern and lateralization of brain activation was found to depend on age and also shown to be related to sex during this task

(Plante et al., 2006), Figure 1.3. Additionally, Wilke et al. showed that the gray matter

subcortical volume was related to

cognitive function in healthy children

aged 7-11 years ( Wilke et al., 2003).

Participants in these studies were

selected from one of the first large scale

fMRI studies undertaken in children.

More recently, the Cincinnati group

extended the age range for fMRI studies

in children from the lower age limit of 5 years, all the way to birth, developing specialized methods to study brain activity using fMRI in newborn infants(Vannest et al., 2014). Pediatric participants as young as 27 months in this study successfully completed MRI scans while they were awake (Vannest et al., 2014). Using data from the Cincinnati MR Imaging study of Normal Development, (C-MIND) they made several important and novel discoveries in the developing brain. First, they showed that neurovascular coupling underlying the BOLD effect exhibited an increase with age during childhood (

Schmithorst et al., 2015). Preschoolers selected from the C-MIND study showed increased activation in the left hemisphere with higher vocabulary scores as well as sex differences in brain activation in the language regions during a story listening task (Sroka et al., 2015).

1.3 Bridging the gap in understanding brain development from infancy to adulthood

This dissertation applies advances in knowledge and tools for mapping brain networks in adults to the challenging task of understanding developing connectivity in brain networks as they emerge in neonates and prematurely born infants. Following a brief introduction to the physics of magnetic resonance imaging and the origins of the fMRI signal in Chapters 2 and 3, this treatise will explore the use of task stimulation fMRI methods in neonates with perinatal brain injury with known possible risk of later visual problems (Chapter 4) (Merhar et al., 2016). Using resting state fMRI the effects of prematurity on functional brain networks is explored in Chapter 5.

Finally, connectivity trajectories that relate to cognitive outcomes over the whole age span from birth to 18 years is explained in Chapter 6.

1.4 Hidden Contributions of this Dissertation

Infant/neonate fMRI data processing remains challenging due to poor image quality, inherently poor conventional MRI contrast in the immature brain due to high water content, and lack of available data or standards as a starting point. Atlas-based segmentation has been widely used to process brain tissue segmentation in the adult brain. To make a statistical comparison between individuals and groups, spatial normalization of each individual brain to a common space created from a large number of brain images converging to an average brain size and shape is typically used as starting point. Launching from the ground-breaking work of the Cincinnati group, infant and neonatal brain atlases were created to provide a priori probability maps for segmentation and normalization of pediatric and infant brains (Altaye et al., 2008, Wilke et al., 2008, Shi et al.,

2010, Shi et al., 2011). In this dissertation, the same methodology and neonatal/infant atlases

5 were used a prior probability distributions to guide segmentation and normalization of infant/neonatal brain image data. This is a critical step to permit the use of the most advanced methods for functional connectivity and graph theory analysis of fMRI data, which requires the specification of ROIs in a standard reference frame.

Beyond this critical first step of normalization of infant brain data to known brain templates, one additional barrier exists to utilizing graph theory and connectivity methods provided in currently available software tools. Construction of a covariance matrix of the correlations between brain regions requires some strategy for dividing the brain into meaningful parcels of neurologically or functional homogeneous elements. During the past decade, two main approaches have surfaced for construction of brain regions based on anatomical or functional parcellation. Anatomical brain parcellation derives from anatomical boundaries using structural brain images ( Tzourio-

Mazoyer et al., 2002, Lancaster et al., 2000) and makes use of more than a century of neuroanatomy (Broca, 1861, Wernicke, 1874, Hebb, 1949, Penfield and Jasper, 1954).

Anatomical atlases have been used extensively but the suitability for functional brain connectivity is still not well understood or accepted. Recently, a data-driven method was introduced to parcellate the whole brain using rs-fMRI data to have homogeneous and spatially coherent ROIs (Craddock et al., 2012). In this dissertation, I have used both anatomical and functional parcellation methods specifically adapted to the unique populations included in each study in the following chapters. Neonatal functional connectivity and network measures were generated using an anatomical parcellated neonatal atlas (Shi et al., 2011). On the other hand, a functional whole brain parcellation method was used for parcellation of the resting state fMRI data from older children in the C-MIND data set into 200 regions to evaluate developmental

6 trajectories of functional brain networks. Having solved each of these problems, the methods are now published and can be applied to future studies in children of all ages.

1.5 Unique contributions of this dissertation

In this dissertation I will describe the application of advanced brain connectome methods to mapping the developing connectome in children from the gestational age of 25 weeks (born 15 weeks prematurely) to 936 weeks. Using a visual stimulation task during fMRI, neonates with perinatal brain injury exhibited lower functional connectivity and brain activation in visual brain regions within 2 weeks of birth than normal healthy babies of the same age. This is the first time these types of neurobiological studies, described in Chapter 4, have been performed so early in life and they demonstrate the acute impact of hypoxic ischemic injury on the brain at this early stage. Further investigations are currently underway to determine sensorimotor and cognitive outcomes for these babies as they grow and mature. Do they catch up with their peers? Do brain networks normalize or do compensatory pathways form to allow the children to develop effectively?

In a second study, described in Chapter 5, resting-state fMRI functional connectivity in preterm born neonates showed decreases in frontal, sensorimotor and language brain regions as well as lower functional brain network measures both locally and globally. Interestingly, our data demonstrate for the first time that within days after birth, differences in functional connectivity are seen in the newborn brain compared with full-term infants. Do these difference represent deficits that will endure through childhood? How long will they last? Are developmental differences between prematurely born children related to the degree of connectivity deficits at birth? Many questions remain and many new ones are generated by the work reported. Future

7 studies will be needed to follow the longitudinal progression of behavior and brain networks in individuals as they mature in order to answer many of these questions. For now, I have provided an important window into the premature brain at birth that coincides with some of the findings in adults and older children who were born prematurely.

Finally, expanding the work reported in premature infants , Chapter 6 evaluates developmental changes in connectivity in normal healthy kids as they grow from babies to adulthood. Again, this is the first study of its kind to apply connectome methods across the age span of child brain development. Using graph theory, Chapter 6 reveals developmental changes in connectome topology with age. Not surprisingly, functional brain networks were found to change significantly during childhood, developing a more efficient organization from birth to 18 years.

The work detailed in each chapter of the dissertation opens a window into the emerging networks in the human brain as it develops its full sensorimotor and cognitive capabilities. It’s a first step toward understanding emerging neurological capabilities and developmental milestones during childhood in relation to brain connectivity.

Chapter 2: The Basic Principles of Magnetic Resonance Imaging

This dissertation applies functional magnetic resonance imaging (fMRI) in children, to make new discoveries about the development of human brain networks at various stages of life and under different adverse circumstances. fMRI is a specific experiment that uses MRI in the human brain to create contrast that relates to brain activity. To better understand the origins of these signals, it is appropriate to provide a brief description of the basic principles of magnetic resonance imaging.

The principle of Nuclear Magnetic Resonance (NMR) was first discovered by two scientists,

Purcell and Bloch, independently in 1946 (Bloch, 1946, Purcell et al., 1946). They received

Nobel prize in Physics in 1952 for this discovery. Raymond Damadian claims an early contribution to this field, having used NMR to detect a tumor by using the properties of � and

� relaxation times in tissue (Damadian, 1971). But the most significant breakthrough in the transformation of the NMR signal into an image came in 1973, when Paul Lauterbur and Peter

Mansfield discovered that an image of an object could be constructed from the NMR signal by encoding the signal spatially with time varying magnetic field gradients (Lauterbur, 1973,

Mansfield and Grannell, 1973). Mansfield also later described a method for echo planar imaging

(EPI) to create an NMR image faster than previous imaging methods in 1977. Since then MRI has been used in many biomedical applications to represent a high-quality images of the human body noninvasively. In this chapter, the theoretical principles of MRI are reviewed to provide a basic understanding of fMRI data acquisition.

2.1 The Quantum Mechanics of MRI

Magnetic resonance imaging refers to the absorption and re-emission of electromagnetic radiation by atomic nuclei in an external magnetic field. The quantum mechanical description of atomic nuclei is related to the property of spin angular momentum. In fact, the property of electron spin was observed in the early 1920’s by Stern and Gerlach (Gerlach and Stern, 1924) who experimented with a beam of neutral silver atoms passing through a non-uniform magnetic field and observed that the beam split vertically into two beams corresponding to two discrete values of the angular momentum of the electron. Spin is an intrinsic form of angular momentum carried by elementary particles and atomic nuclei, examples are given in Table 2.1 (Brown et al.,

2014).

The spin angular momentum is characterized by the spin quantum number Ι. Nuclei that have a non-zero value of spin quantum number, Ι (Brown et al., 2014, Sakurai, 1994) exhibit a magnetic spin resonance. The spin angular momentum vector �⃗ has a quantum number Ι that can take an integer or half integer values and its magnitude is given by

� = Ι(Ι + 1)ℏ 2.1 where ℏ = 1.05�10 J.s (Plank’s Constant).

The spin magnetic moment, �, which is proportional to the angular momentum,

� = �� = �ℏΙ 2.2 with the constant of proportionality, �, is called gyromagnetic ratio. The gyromagnetic ratio is a property of the nucleus and has a value of 2.675�10��/�/� for protons. The nucleus of

Hydrogen (1H) contains a single proton and is of primary interest for MRI signals for medical applications because of its ubiquity and high sensitivity due to high gyromagnetic ratio and high natural abundance as shown in Table 2.1. When this magnetic moment is placed in a strong static magnetic field (� ) along the �-direction, the net magnetic moment aligns with the static magnetic field in the direction of the � . In fact, the net magnetic moment represents the difference between spins oriented opposite to the magnetic field direction (higher energy state) and a slightly greater number of the spins will orient in the same direction as the magnetic field

(lower energy state). Thus we get a net magnetic moment oriented along � . The possible value of the z-components of magnetic moment in the presence of the static magnetic field along z-axis are

� = ℏ�� where � = Ι, (Ι − 1), (Ι − 2), … . . , −Ι 2.3 1 1 So for the proton, with spin 2 , there are two possible values for � ± 2 �ℏ (corresponding to a spin state parallel and opposite to applied magnetic field, respectively). Therefore, the discrete magnetic moment leads to discrete energy values.

� = −�. � = −�� = −��ℏ� 2.4 with � = ± 1⁄2, + 1⁄2 spin parallel state, − 1⁄2 spin anti-parallel state to field.

This is known as the Zeeman Effect and is

shown as a quantum mechanical energy

diagram in Figure 2.1, where atomic or

nuclear magnetic moments in the presence

of a �-field lead to atomic or nuclear

energy levels splitting into the spin-down state having a higher energy and spin-up having lower energy.

For a proton, a transition between two states represents an energy change, DE given by:

△ � = �(� = − 1⁄2) − �(� = + 1⁄2) = �ℏ � 2.5

Transition between the two states can be induced by absorption or re-emission of a photon frequency �, such that

△ � = �ℏ� = ℏ� 2.6 where,

� = �� 2.7 is the frequency called the Larmor precession frequency.

In a real system, a large number of nuclei could occupy a particular spin system. Thus, the theory must consider an ensemble of spins. To do this an arbitrary state |�⟩, which can be written as a linear combination of the possible spin states labeled by � for a nucleus is defined

2.8 |�⟩ = �|�⟩ where � represents the amplitude or spin population in each state. When making a measurement on such a system, the expectation value, of the z- component of spin (Ι) is given by:

⟨�|Ι|�⟩=∑|�| � 2.9

where the value |�| represents the probability of finding a single nucleus in the state �. So for the case of spin one-half particles, such as protons where, Ι = ± 1⁄2, the average expectation value can be rewritten with two states

2.10 �⟩ = �⁄|+ 1⁄2⟩ + �⁄ − 1⁄2⟩,

From Boltzmann statistics, at thermal equilibrium the population of the energy levels is

�ℏ�� exp ( ) 2.11 �� |�| = �ℏ�� ∑ exp ( ) ��

where � = 1.38�10 �⁄� is the Boltzmann constant. The ratio of the population of the two energy states for nuclei with � = ± 1⁄2, provided �� ≫ ℏ�� , can be shown, respectively:

2.12 �⁄ ℏ�� = exp (− ) � � �⁄

ℏ�� ℏ�� ℏ�� 2.13 exp = 1 + + �( ) ��

2.14a �⁄ ℏ�� ≈ 1 − � � �⁄ and the difference the between the number of spins in the spin-up state and spin-down states is

ℏ �⁄ − �⁄ ≈ 2.14b 1 If we assume that all the spin-up nuclei have a magnetic moment 2 ℏ� and the spin-down 1 nuclei have a magnetic moment of − 2 ℏ� , the bulk magnetization of the ensemble in the direction of the � -field is given by

ℏ� � 2.15 � ≈ � 2 ��

13 where N is the total number of spins. The last equation implies that the behavior of all the spins of an ensemble in terms of magnetization (the vector sum of the magnetic moments of individual nuclei) leads to transfer from a quantum mechanical description to classical description of MRI.

2.2 The Classical Description of MRI

In classical mechanics, if the spin magnetization vector � is placed in an external magnetic field

�, � will experience a torque. The classical equation of motion for the magnetization � is given

(Brown et al., 2014) by

�� 2.16 = �� which is the simple version of the Bloch equation (Bloch 1946). For a static magnetic field � which is usually taken to be �̂ direction, � = ��̂ , the corresponding solutions become

�� 2.17 = �� , = −�� , = 0 �� which has solutions in matrix form are given by,

2.18 0 �� 0 � �(�) �� 0 �(0) ⎛⎞ =−�� 0 0 � and �(�) = −�� 0 �(0) ⎜ ⎟ 0 0 1 � �(�) 0 0 1 �(0) ⎝ ⎠ where � = ��, which is again the Larmor equation (Eq. 2.7) . In addition to the static �� field applied along z, consider an RF magnetic field �� , applied perpendicularly to �� and oscillating at �. If only the circularly polarized component of �� rotating in the same direction as the magnetization is considered

��(�) = � �� ̂ − � �� ̂ 2.19 which when put into equation of motion yields

�� = �� + � � ��

�� = �[� � �� − � � ] ��

�� 2.20 = �−� � �� − � � ��

when �(0) = �� ,where � = ��. This implies that the magnetization precesses at � and

� simultaneously when a oscillating magnetic field of frequency � applied, as shown in

Figure 2.2a.

Now it is time to introduce a new reference frame to view the evaluation of the magnetization vector, the rotating reference frame, which rotates about z axis with angular frequency � with respect to the fixed field. The equation of motion in the rotating reference frame (�, �, �) can be written

��′ 2.21 = ��′��

Where � is an effective magnetic field, around which the magnetization vector precesses

(Figure 2.2c:

� 2.22 � = � − � + � �′ �� 0 � 1

� and (� , � , �) are unit vectors in the (� , � , �) directions. When (� = �) on resonance, � =

�� and magnetization precesses about � as shown in Figure 2.2c. Applying the � magnetic field has the effect of rotating magnetization vector about the � axis with an angular frequency

� = ��.

When RF field is applied during the time �, the magnetization becomes tilted away from � axis by

an angle � = ��. Most commonly, an RF pulse is used to tilt the magnetization by 90 , 180 , or partial flip angle (<90). If � is 90 degrees, the longitudinal magnetization does not exist and the transverse magnetization is at its maximum, precessing at the Larmor frequency.

2.3 Relaxation

As we discussed in the previous section, the application of a resonant RF pulse disturbs the equilibrium net magnetization component along the direction of �. Once the RF field is off the spin system starts to relax to come back to equilibrium. There are three very important relaxation

∗ processes in MRI, �, � and � (Brown et al., 2014).

2.3.1 Introduction of Spin-Lattice (��)

The spin-lattice (or longitudinal) relaxation time, also known as � relaxation time, is a measure how quickly the longitudinal magnetization (�) reaches a Boltzmann equilibrium with its surroundings (lattice) after a �� pulse is applied. The equation of motion for � is:

�� −(� − � ) 2.23 = ��

The solution can be found as:

/ / �(�) = �(0)� + �(1 − � ) 2.24

This implies that the longitudinal magnetization shows an exponential form displaying the evolution from initial value to equilibrium value after an initial disturbance.

2.3.2 Introduction of Spin-Spin (��)

The other classical component of the magnetization is the transverse magnetization. Immediately after the �� pulse, the transverse magnetization is at its maximum but then it starts to decay because of loss of phase coherence. The spin-spin relaxation time, also known as � relaxation time, is the rate of reduction in transverse magnetization, in the equation describing the evaluation of �:

�� 2.25 = − �� with the solution:

/ � (�) = � (0)� 2.26

This described the exponential decay of the transverse magnetization.

∗ 2.3.3 Introduction of ��

Due to the external field inhomogeneities there is an additional dephasing of the magnetization.

∗ The reduction of the transverse magnetization by replacing � with � takes the form:

/∗ � (�) = � (0)� 2.27

∗ � is defined as the combination of external field induced (�) and thermodynamic (�) effects

1 1 1 2.28 ∗ = + � � �

The quantity of � depends on the MRI machine and the sample and can be much smaller than

�.

2.5 Bloch Equation with Relaxation Terms

Equations for magnetization in the presence of a magnetic field with relaxation terms were defined in the previous section. When combined with the earlier equation of motion we have the differential form of the ��ℎ �� (Brown et al., 2014):

�� 1 1 2.29 = �� − ��̂ + ��̂ + (� − �)� �� where (�, �, �) are unit vectors in the (�, �, �) directions. For � = � the ��ℎ �� in the matrix form can be written:

0 1/� �� 0 � �� 0 ⎛⎞ 0 =−�� −1/� 0 � + with �(��) = −�� 0 2.30 ⎜ ⎟ � 0 0 1 0 0 1 ⎝ ⎠

with solutions:

�(�) � 0 0 �(0) 0 �(�) = � (� �) �(0) + 0 0 � 0 2.31 �(�) �(0) �(1 − � 0 0 �

where �(��) is the rotation matrix about the z-axis at the Larmor frequency � (left-handed rotation).

2.5 Signal and Spin Density

The initial magnetization phase �, � and the receive field � are position independent when the transmitting and receiving RF coil are considered as uniform. It is useful to introduce a proportionality constant Λ and then the signal equation can be written as (Brown et al. 2014)

((,)) 2.32 �(�) ∝ �Λ� � �� (�, 0)� where Ω is reference frequency and the relaxation effects are neglected. The phase �(�, �) =

− ∫ ��(�, �). The initial transverse magnetization can be expressed in terms of spin density �,

�ℏ � 2.33 � = � = ( ) � 2 ��

Substituting this equation into Eq. 2.32, the signal equation can be written:

2.34 �(�) = Λ ��(�)�((,))

( ) ( ) where �(�) is the spin density and � �, � = −�� − ∫ �� , � . If the external magnetic field varies due to an applied linear gradient, the angular frequency and phase can be expressed, respectively as:

2.35 �(�, �) = � + ��(�)� �� (�, �) = −�� − � ��(�)�

The spatial frequency is defined �(�):

� 2.36 �(�) = ��(�) 2�

The signal with the demodulated frequency Ω = � is given by:

2.37 �(�) = Λ ��(�)�

The equation implies that when the externally applied magnetic field is varied spatially by applying a linearly varying magnetic field gradient, the signal �(�) is the Fourier transform of the spin density of a sample (Fong, 2005). The spin density can be found by taking the inverse

Fourier transform of the signal

2.38 �(�) = Λ �� (�)�

2.6 Spatial Encoding in MRI

In MRI, the goal is not only to establish the presence of different nuclei, but also to determine the signal strength at spatial locations in an object or sample. To impose spatial dependence, the uniform static field can be manipulated by a small linear varying gradient field. Three orthogonal gradients are commonly used in MRI for spatial encoding, referred to as: slice selection gradient, phase encoding gradient, and frequency encoding gradient.

2.6.1 Slice Selection

The first step of spatial encoding in MRI is slice selection which is done by exciting a very thin slice of the sample by combining the gradient magnetic fields and frequency selective RF pulses.

In the presence of a slice selection gradient along z-direction, Gz, the frequency of precession becomes a linear function of position along the z-direction. The bandwidth and center frequency of a shaped RF pulse can be calculated and matched with the field gradient to selectively excite a narrow slice of the sample. All spins in the slice thickness (∆�) will have an identical phase and

20 flip angle after slice selection. The frequency at position z is �(�) = � + ��, where � is the

Larmor frequency. The bandwidth of the RF pulse, �� is given by

�� = ∆� = ��Δ� 2.39 and slice thickness ∆� = ��⁄��. As a summary, in the presence of a linear field gradient �, a range of frequencies can be excited to create transverse magnetization in a slice thickness ∆� along the z-direction.

Consider a slice is excited at � = 0 with a constant gradient strength � = � during �

� followed by a reversed gradient during 2. This reversed gradient, called a rephrasing gradient, is added after the RF pulse is turned off to have an accumulated phase which is equal and opposite to the initial phase, thereby cancelling its rephrasing effect. At the end of the rephrasing lobe of the slice selection gradient, all excited spin within the slice are all at zero phase, � = 0. The signal from the whole slice becomes

2.40 � = �� (�, �, �)�� where � is the center of the slice.

2.6.2 Phase Encoding

After the slice excitation pulse, it is followed by a phase encoding gradient � (�) along y- direction. Assume for the moment that no other gradient is applied during this period. The magnetization accumulates a y-dependent phase when � is kept on for a time �. The phase of point (�, �) is determined by

�(�, �) = �� 2.41

After � is switched off, the precession frequency returns to a constant value determined by the main field, while the phase remains proportional to �. The signal immediately after the phase encoding is given by:

2.42 � = �(�, �, �)�� where � = � � .

2.6.3 Frequency Encoding

The final step in spatial encoding is the application of a constant gradient � along x-direction, also called readout gradient �. Due to this gradient, the frequency of precession will change linearly with location:

�(�, �) = �� 2.43

If the signal is received while � is applied, contributions of from the sample at different locations along the x-axis will have different frequencies. With the three encoding steps described above, performed in a pulse sequence, after the beginning of the frequency encoding gradient �, the signal is given by:

2.44 () �(�, �) = �(�, �, �)�� where � = � � and � = � � .

2.7 Echo Planar Imaging

Echo-planar imaging (EPI) (Mansfield, 1977) uses only a single RF pulse for the excitation of a sample, with all of the encoding gradients played rapidly to form an image in just a few hundred

22 milliseconds. The slice selective RF pulse tips the longitudinal magnetization away from its longitudinal axis. Frequency encoding gradient is applied rapidly with a short phase encoding gradient blip applied between oscillation read-out gradients, to obtain gradient echoes. The data collection is reconstructed during each set of echoes. EPI allows us to acquire a complete set of image slices covering the entire object in a time that is typical of a single slice for conventional

MRI. With EPI, T2* image contrast is typically very strong. The fast acquisition and contrast make EPI ideal for fMRI data acquisition. An example of the EPI pulse sequence is illustrated in

Figure 2.3.

2.7.1 k-space coverage

The MRI data is collected in k-space which consist of an array of numbers representing the spatial frequencies comprising an image. The standard EPI imaging sequence diagram and corresponding k-space coverage are shown in Figure 2.3. The bottom line in k-space is obtained by the usual read gradient structure during sampling of the data. Applying a short positive � blip with enough amplitude to carry � up one line moves the position in k-space to the right- side of the second line from the bottom. To bring it back across � requires applying a negative lobe of the same time duration as the positive rephrasing gradient applied to the bottom line.

Repeating the same steps, and each time blipping the phase-encoding gradient while reversing the polarity of the readout gradient, will drive the echo resonance through the (�, �) plane.

The total acquisition time � is defined as � = �(� + 2�), where 2� is the time the read gradient takes from the positive peak to the negative peak and vice versa. The time to complete

one EPI experiment is � = + � + � . A full set of 2D image information during � is collected in the EPI pulse sequence by applying just one RF pulse (Brown et al., 2014).

∗ Therefore, the contrast is governed by � and spin density, not by �. Applying only one RF pulse greatly shortens the image acquisition time and the contrast is governed by spin density.

� Conventional � weighted contrast can also be acquired for a repeated EPI for 2 RF pulse and the signal equation is

2.45 ∗ � ∝ � 1 − � �

Chapter 3: Principles of Functional Magnetic Resonance Imaging

(fMRI)

FMRI is used to measure brain activity that is associated with the increase of blood flow

(Blamire et al., 1992). FMRI is a relatively new research method that gives important perspectives into brain function and it has been used to investigate the functions of normal, injured and diseased brains. The primary method that fMRI uses is blood-oxygen-level dependent (BOLD) (Ogawa et al. 1992) contrast. BOLD fMRI detects the local increase in blood flow and corresponding increase in relative blood oxygenation as a consequence of local neuronal signaling, Figure 3.1a. The primary goal of this chapter to build on the basic physics of

MRI presented in the previous chapter to develop a basic understanding of fMRI principles as a basis for understanding the material presented in the remaining chapters. Functional connectivity based on fMRI is also described briefly. Finally, functional brain network analysis using graph theory approaches will be presented.

3.1 Brain Functional MRI and BOLD

In the oxygenated state the ratio of oxyhemoglobin (HbO2) to deoxyhemoglobin (Hb) increases and can be detected based on the magnetic susceptibility change. This signal change is related to the underlying neuronal activation that stimulates the change in blood flow and oxygenation.

BOLD fMRI uses the different magnetic properties of HbO2 and Hb. Hb is paramagnetic due to the pairing of electrons in the molecule and this causes it’s net magnetic moment to align opposite to the external magnetic field applied in the brain, causing magnetic field distortions in the surrounding tissue. HbO2 is diamagnetic and does not distort the magnetic field in the brain.

∗ The increase in the concentration of paramagnetic molecules in blood tends to decrease the �

∗ relaxation time, thus the signal contribution in the region. However, � becomes longer when the diamagnetic molecules increase in the region of greater neuronal activity and the signal intensity increases relative to Hb state (Ogawa et al.,1990, Ogawa et al.,1992). A hemodynamic response function (HRF) is used to model the temporal behavior of the BOLD signal in a fMRI experiment. The HRF is typically modeled as a gamma variate function as shown in Figure 3.1b, exhibiting an initial dip immediately after neuronal activity increases, then a rise and peak in blood flow at 6 seconds, followed by a return to baseline with an undershoot that persists for a few seconds longer.

The fast method of echo planar imaging (EPI) (Mansfield et al., 1977) described in Chapter 2 is the most commonly used imaging sequence in fMRI experiments. EPI allows collection of whole brain data in a few seconds or less. A large series of images is collected rapidly while the subject performs a task that shifts brain activity between two states or remains quietly at rest during the fMRI experiment.

3.2 fMRI Experimental Designs

Block design and event-related design are two main approaches to the design of fMRI experiments. The block design is the most time efficient approach for comparing different states

(task and rest conditions) of brain responses in fMRI experiments, as shown in Figure 3.2a.

Block design uses relatively long alteration periods (typically 30 seconds) during each state.

Thus, enough data can be accumulated during block design experiments to make statistical analysis feasible. Block designs are powerful for locating activated voxels. As described in the

previous section, during stimulation, the HRF goes through an initial dip and then a gradual rise in the signal to a peak level, eventually returning to a baseline level in the absence of further stimulation. During the task in a block design fMRI experiment, there is a constant stimulation and hemodynamic response does not return to baseline. When the stimulation is turned off (rest condition) the signal decays back to baseline. Voxels do not show the characteristic response and increased level of signal during rest condition compared to task condition. However, block design is appropriate for some applications as it can be difficult to control a cognitive state accurately during a long period of each block. In this case, event-related design may be more

27 appropriate as shown in Figure 3.2b. In this paradigm, fMRI data is accumulated during random stimulation in different conditions,. The event-related design takes longer than block design to achieve a sufficient signal statistics. The statistical analysis of event-related design is much more complex than block design and convolution of the stimulus timing with the canonical HRF is important.

3.3 General Linear Model (GLM)

The raw BOLD fMRI data can be acquired over periods as short as a few minutes. The raw data need to be preprocessed to remove the effects of head motion, heartbeat, breathing, and scanner artifacts. The general linear model (GLM) is typically used as a basic statistical approach for fMRI data analysis: modeling the fMRI data as a linear combination of predictor variables

(Friston et al., 1995). The typical analysis is mass univariate approach performed by construction a separate model for each voxel. Every voxel of the data in the brain is an outcome and the predictors are a series of regressors that are developed based on tasks and conditions. The structural model for the GLM is simple:

� = �� + � 3.1 where the vector Y, representing time course for a given voxel, is usually filled with values of the convolution integral of the original time courses and the chosen HRF, β is the unknown parameter vector, X is the task related design matrix and Ε is the residual or error vector. The columns of the design matrix characterize the explanatory variables to be estimated. Consider an experiment of alternating blocks of task and rest condition. To determine if there is a significant difference in activity between task and rest condition, the similar events are grouped in block design. Then the explicit GLM for this block design can be presented in equation as:

� 1 � � ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ � 1 � � ⎢ ⎥ ⎢ ⋮ ⎥ ⋮ ⋮ � ⎢ ⋮ ⎥ 3.2 ⎢� ⎥ ⎢ � ⎥ ⎢� ⎥ = ⎢0 ⎥ � + ⎢�⎥ 0 � ⎢�⎥ ⎢ ⎥ � ⎢ ⋮ ⎥ ⎢⋮ ⋮ ⎥ ⎢ ⋮ ⎥ ⎣�⎦ ⎣0 �⎦ ⎣�⎦

where the two columns of X are multiplied by the model parameters � (rest) and � (task) which are estimated when the model is fitted in each voxel. � is of particular interested, because it is going to capture the activation amplitude and estimate how much the task versus rest conditions differ in a voxel. When each column is linearly independent in the design matrix, β values can be estimated by the following equation:

� = (��)�� 3.3 which is known as the ordinary least square (OLS) estimate that minimizes the squared residuals.

After the parameter estimation, a hypothesis can be tested using t-statistics which can be obtained by the ratio of contrast value and the standard error of the contrast given by:

�� 3.4 � = ��(��) where c is the contrast vector that defines the weight of each column, so that �� gives a numerical value that can be positive or negative. �� is the simple effect of two different state and scaling of weights does not affect the t-value. �� becomes zero under the null hypothesis that allow us to do the t-test.

3.5 Functional Connectivity of fMRI

Functional connectivity is estimated from the temporal correlation in the fMRI signals among two or more anatomically distinct regions that act together either during a particular task or while

29 the brain is at rest. Conventional fMRI makes use of BOLD contrast to determine regional brain activity. By analyzing the areas of the brain that are functionally connected, distinct resting networks can be derived from fMRI data. Several approaches have been introduced to estimate functional connectivity from BOLD fMRI data (Fransson, 2005, Biswal et al., 1995, Friston et al., 1993, Beckman et al., 2005). However, the most widely used method to explore functional connectivity of a seed region is to correlate the time-series of this predefined seed region across the time-series of all other seed regions. Examining the functional correlations between all possible regions by using a cut-off threshold allows for construction of the graph theory analysis that represents the functional brain networks organization (van den Heuvel et al., 2010).

3.6 Graph Theory Analysis

More recently, graph theory has been applied to fMRI data from the human brain to compute complex network maps (Bullmore and Sporns, 2009). Graph theory is an area of mathematics that can provide quantitative information about networks on both a local and global scale. Brain networks are created consisting of a large number of non-overlapping brain regions and edges

(corresponding link pairs of nodes). A graph can be defined either with directed or undirected connections with respect to how the edges connect to each other. Edges can be either binary or weighted. Brain networks are represented using an �� connectivity matrix where � is the number of nodes. After network construction, the network parameters associated with network topology and efficiency can be quantified. Any given network measure may characterize one or more aspects of global and local brain connectivity (Rubinov and Sporns, 2010). Here, the global and local network parameters are briefly defined as follows.

Global efficiency is defined as the average inverse shortest characteristic path length between all pairs of nodes (Latora and Marchiori, 2001). Global efficiency for a network G with N nodes is given

1 1 3.5 �(�) = �(� − 1) � ,, where � is the characteristic path length between nodes � and � in the network.

Local efficiency is computed as the average of the global efficiency of each node’s neighborhood sub-graph:

1 3.6 � (�) = � (� ) � where � is the sub-graph of the nearest neighbors of node �.

In contrast to global efficiency, local efficiency represents the capacity to transfer information within the neighbors of a given node, and a high local efficiency reflects efficient information transfer in the immediate neighborhood of each node (Lo et al., 2010).

Betweenness centrality of a node is the fraction of shortest paths between all other pairs of nodes in the network that actually pass through the node of interest (Freeman, 1978). This measure has been used to identify the most central nodes in a network, which reflects the importance of the specific node in transferring information to other nodes. A node with high betweenness centrality is crucial to efficient communication.

Node degree is defined as the number of links connected to the node.

Modularity (Q) quantifies the degree to which the network may be partitioned into a subdivision that has higher connections with each other than with the rest of the network (Newman, 2006).

Assortativity quantifies how often nodes with a certain degree tend to link to other nodes with the same or similar degree (Newman, 2002).

Small-worldness is computed using clustering coefficient and characteristic path length (Watts and Strogatz, 1998). Clustering coefficient (C) is the fraction of the node’s neighbors that are also neighbors of each other. Characteristic path length (L) is defined as the average of the shortest path length between all pairs of nodes. Small-worldness (sigma) of a network is then defined by the ratio of normalized clustering coefficient (�) and and normalized characteristic path length (�):

� = (�⁄� )/(�⁄� ) 3.7 where � =�⁄� and � =�⁄� (Watts and Strogatz, 1998 ). The clustering coefficient and characteristic path length of a random network , � and �, are the average of the values calculated from 100 randomized models. It should be noted that a network shows small- worldness if � > 1, �> 1 and � ≈ 1.

Rich-club organization is calculated over a range of degree � after removing all nodes with degree ≤ �. The rich-club coefficient ∅(�) for any � is the ratio of connections between the remaining nodes � and the total number of possible connections for a fully connected network with the same number of nodes N. The rich-club coefficient ∅(�) is computed as follows:

2� 3.8 ∅(�) = �(� − 1)

For comparison across individuals, rich-club coefficient is normalized by the average rich-club coefficient over 1000 random networks ∅(�),

∅(�) 3.9 ∅(�) = ∅ (�)

A normalized rich-club clustering coefficient ∅(�) > 1 over a range of � indicates rich-club organization in the network (van den Heuvel and Sporns, 2011).

These graph theory parameters will be examined in the following analysis of resting state fMRI data from infants and children across the age span, testing for the effects of gestational age, age

32 since birth, sex, IQ and other cognitive variables. This chapter defines the methods and analysis approaches that are applied in subsequent chapters and can serve as a reference for interpretation of those results.

Chapter 4: Functional Connectivity of the Visual System in Infants with Perinatal Brain Injury

This work of this chapter was originally published with the following authors and citation:

Merhar, S.L., Gozdas, E., Tkach, J.A., Harpster, K.L., Schwartz, T.L., Yuan, W., Kline-Fath,

B.M., Leach, J.L., Altaye, M., Holland, S.K. (2016) Functional and structural connectivity of the visual system in infants with perinatal brain injury. Pediatric research, 80:43. It is reproduced in this dissertation with permission of the other authors and from the publisher, Springer Nature on behalf of the journal Pediatric Research. 6 April 2016. Doi:10.1038/pr.2016.49. This work is included here in recognition of the significant contribution that I made to the publication, which included assistance with data acquisition, all of the data analysis and the write up of the methods section of the published paper describing data acquisition and analysis.

The work described in this chapter describes a neuroimaging study in neonates with perinatal brain injury. At the outset of this study, we hypothesized that infants with perinatal brain injury would have less brain activation during a visual stimulation paradigm as measured using fMRI and reduced task-based functional connectivity as compared with healthy controls. Using advanced neuroimaging techniques we set out to detect functional alterations of the visual system in infants with early brain injury, during the neonatal period. Ten infants with perinatal brain injury and 20 control infants underwent visual fMRI during natural sleep with no sedation.

Activation maps, functional connectivity maps, and structural connectivity were analyzed as part

34 of the work toward my dissertation, and these results were compared between the two groups.

Most infants in both groups had negative activation in the visual cortex during the fMRI task.

Infants with brain injury showed reduced activation in the occipital cortex, weaker connectivity between visual areas and other areas of the brain during the visual task as compared with control infants. As an initial step, the chapter establishes a perception of developing brain functions in neonates with hypoxic-ischemic brain injury. Portions of the following chapter are produced verbatim from the published work.

4.1 Introduction

Infants with brain injury sustained in the perinatal period are at risk of later visual problems, including low visual acuity, decreased binocular depth perception, abnormal eye movements, and cortical visual impairment (Mercuri et al., 1997a, Chau et al., 2013). The ability to predict visual outcome based on conventional imaging and ophthalmologic examinations in the neonatal period is poor (Mercuri et al., 1997a, Mercuri et al., 1997b, Bassi et al., 2008). Therefore, children with brain injury who could benefit from early ophthalmologic assessment and early intervention vision therapies may have delayed diagnosis and treatment of visual problems. As advanced neuroimaging techniques have become more common, the visual system has been studied in preterm and term infants. Using task-based fMRI, occipital lobe activation is reliably obtained in infants after term equivalent age (Martin et al., 1999, Yamada et al., 2000) but not in the preterm period (Lee et al., 2012). Functional connectivity of the visual system in the neonatal period using task-based fMRI has not been previously reported.

The overarching goal of the work reported in this chapter was to determine whether fMRI could be used as an adjunct to conventional clinical MRI to add additional information for clinicians

35 and families about potential damage to the visual system in the neonatal period. In this study, we performed a visual fMRI with a flashing light stimulus in 20 term control infants <8 week postnatal age and in 10 infants with brain injury (both term and preterm) at <48 week postmenstrual age. Our hypothesis was that we would find differences in activation and functional connectivity between the two groups, with less brain activation and weaker functional and structural connectivity in infants with brain injury vs. healthy controls. We formulated this hypothesis based on the theory that the infants with brain injury would have lost neurons and the clusters of activation would be smaller, leading to attenuated vascular response and thus reduced amplitude BOLD signal responses.

4.2 Methods

4.2.1 Participants

Twenty term control infants and 10 infants with brain injury were scanned using fMRI with a visual stimulation paradigm before 48 week postmenstrual age. The protocol was approved by the Cincinnati Children’s Hospital Institutional Review Board and informed parental consent was obtained. All imaging was undertaken during un-sedated natural sleep (Vannest et al., 2014).

Hearing protection was provided with earplugs and headphones.

4.2.2 MRI Acquisition Parameters

MRI data were all acquired on a 3T Philips Achieva MRI scanner (Phillips Medical Systems,

Best, the Netherlands) in the Imaging Research Center at Cincinnati Children’s Hospital using a

32 channel phased array head coil. Infants were swaddled for scanning and placed in the head coil with ear phones and ear muffs for hearing protection. A high spatial resolution (1 mm

36 isotropic) 3D T1-weighted scan was obtained using an inversion recovery gradient echo imaging method for anatomic reference. fMRI was performed in the axial plane using a single shot gradient echo echo-planar imaging sequence with echo time 35 ms, repetition time 2 s, slice thickness 3 mm, flip angle = 90°, field of view 180×180 mm, matrix size 64×64, voxel size

2.8x2.8x3 mm, 36 slices, SENSE Factor=2, receiver BW=3423.6Hz/pixel.

4.2.3 Visual Task Paradigm

The visual paradigm was administered in a periodic block design consisting of 5.5 cycles of 30 s

ON/OFF conditions. For the ON condition, a visual stimulus (8 Hz flashing lights) was delivered via a LCD display blinking white, through closed eyelids, contrasted with total darkness during the OFF condition. The total scan time was 5 min 30 s, during which 165 BOLD image volumes were obtained.

4.2.4 Data Analysis fMRI data were analyzed using FEAT (fMRI Expert Analysis Tool, Version 5.0.8) which was part of the FSL processing package (FMRIB, Oxford, UK) (Smith et al., 2004). Standard preprocessing steps were applied to all data steps as implemented in FEAT: motion correction, nonbrain tissue removal, spatial smoothing, normalization (grand mean scaling of the timeseries), and high pass filtering. Data sets were assessed and corrected for motion using

MCFLIRT (FSL’s intramodal motion correction tool), which estimates absolute and relative displacement through the entire image time-series. After motion correction, skull stripping was performed using the BET function implemented in FSL. Images were temporally high-pass filtered and smoothed using a 6mm Gaussian kernel full-width at half maximum algorithm. We

37 used FLIRT (FMRIB’s Linear Image Registration Tool) to register the functional data to the neonate atlas (Shi et al., 2011) with 12 degrees of freedom affine transformation. For each subject, data were fitted to the general linear model as the first level analysis with the six motion parameters modeled as covariates. The second level analysis was performed based on mixed effect analysis using FLAME. For each subject, first level activation maps were merged in a higher-level mixed effect analysis using FLAME (FMRI’s local analysis of mixed effects) to evaluate group differences. All subjects were included in the group analysis, but only negative

BOLD responses were included in the first-level analysis. To ensure that the mixed effect analysis was valid given the heterogeneity of our population, we used the Randomize routine within the FSL program to perform a nonparametric permutation-based statistical analysis, with no difference in the results.

For the visual functional connectivity analysis, all image processing was performed using SPM8

(Welcome Department of Imaging Neuroscience, London, UK). The functional image series were corrected for head motion by realigning all images to the mean of all functional volumes.

The mean functional image was co-registered to a corresponding T1-weighted high-resolution image. T1-weighted images were segmented using neonatal tissue probability maps (Shi et al.,

2011). Finally, the images were normalized to the neonatal template, resliced to a voxel size 2 ×

2 × 2 mm, and smoothed with a full width at half maximum Gaussian Smoothing kernel of 6mm.

After pre-processing for each subject, a general linear model was used to regress the time course of visual task vs. rest as the main effect and the realignment parameters as regressors of no interest. SPM.mat files of each subject were imported into CONN (Whitfield-Gabrieli and Nieto-

Castanon, 2012) to process ROI-to-ROI connectivity analyses. This toolbox implements a noise

38 reduction method (aComp- Cor) (Behzadi et al., 2007) that accounts for physiological noise, and the covariates were included with a principal components analyses reduction of the signal from subject specific white matter and CSF masks. The residual BOLD timeseries was then band pass filtered (0.008

Anatomical Labeling) atlas. Bi-variate correlations were calculated between each pair of ROIs.

All ROIs were imported as possible connections for our selected ROIs. The unpaired t-test was used with a threshold set at P < 0.05 false discovery rate corrected to determine whether significant differences existed between the control and brain injured groups.

4.3 Results

4.3.1 Participants

The control group consisted of 20 healthy term infants scanned at <48 week postmenstrual age with a mean gestational age of 39 week and a mean postmenstrual age of 44.7 week at the time of scan. 60% were female and 40% were male. The 10 infants with brain injury consisted of 6 term infants and 4 preterm infants all scanned at <48 week postmenstrual age, with a mean postmenstrual age of 44.2 week at the time of scan. 60% were female and 40 % were male.

4.3.2 Visual Task fMRI

On individual analysis, 14/20 control participants and 8/10 of those with brain injury had negative brain activation in the expected visual cortical areas. One control participant had positive activation in the visual cortex, and the rest had no activation. Figure 4.1 shows the group

39 average activation maps for the controls (a), infants with brain injury (b), and the difference between the groups (c), as measured by fMRI during visual stimulation, using a mixed effects model. The controls had a significantly higher percent signal change, which represents the change in BOLD signal intensity from baseline, of 1.78% as compared with 1.4% for the brain injury group, P=0.007.

4.3.3 Functional Connectivity

The control group had stronger

connectivity between the visual

areas and other areas of the

brain. Functional connectivity

was derived from BOLD time

courses modeled during the task in the block design using a general linear model. We examined several areas of the brain known to be related to vision for their functional connectivity with other brain regions in the network.. In the control group, the lingual gyrus, supramarginal gyrus, cuneus, superior occipital gyrus, middle occipital gyrus, and inferior occipital gyrus were found to have strong functional connectivity with other regions. By contrast, these regions in children with brain injury showed very few strong connections with other brain regions.

4.4 Discussions

Normal vision requires intact anterior (eyes, optic nerves, optic chiasm) and posterior (lateral geniculate nucleus, optic radiations, and primary visual cortex) pathways. Visual information is

40 then relayed to association cortices or to the extra geniculate visual system (Chau et al., 2013).

The corpus callosum allows for integration of the two halves of the visual field (Pietrasanta et al.,

2012). In cortical visual impairment, visual loss is due to injury to the primary visual cortex (V1) and/or extrastriate visual cortical areas (V2, V3, V4, V5). In the term infant, cortical visual impairment is most commonly due to hypoxic–ischemic injury or severe hypoglycemia (Tam et al.,2008, Lehman, 2012). In the preterm neonate, cortical visual impairment is often due to brain injury directly impacting the visual pathways, most often the optic radiations or interruption of the thalamocortical fibers due to intraventricular and/or intraparenchymal hemorrhage (grade 3–4

IVH) (Chau et al., 2013, Lehman, 2012). In our study, we compared fMRI in healthy term-born control infants with term and preterm infants with brain injury. Other investigators have attempted to use task based fMRI to evaluate the visual system in preterm and term neonates with varying degrees of success. In preterm infants before term equivalent age, activation of the visual cortex in response to a strong visual stimulus is inconsistent (Lee et al., 2012, Born et al.,

2000). From term age up until several months or years of life, BOLD response can often be elicited but most authors have found a positive BOLD signal in these infants (Martin et al., 1999,

Yamada et al., 2000, Lee et al., 2012, Born et al., 2000, Born et al., 1998, Erberich et al., 2003).

In contrast, we found fairly consistent negative visual BOLD activation in the medial occipital lobe in our population. Negative BOLD signal during activation can be caused by a decrease in cerebral blood flow, increase in cerebral blood volume, or increase in oxygen extraction. Recent animal studies demonstrate that, unlike in adults, the delivery of oxygenated hemoglobin does not overcompensate for the elevated local metabolic demand in the newborn (Kozberg et al.,

2013). An increase in the local deoxyhemoglobin concentration results from focal brain stimulation, leading to a negative BOLD response, as was observed in our study. Another

41 possible explanation is that the strong visual stimulus creates a very high local metabolic demand, and blood flow is unable to increase sufficiently in the neonates to compensate. In addition, recent data shows that the coupling between the BOLD signal and hemodynamic response is age-dependent in children, being weaker in younger children, which could imply that this response is not sufficiently mature in neonates to meet metabolic demand (Schmithorst et al.,

2015). Notably, in one study, sleeping adults showed a negative BOLD response to the same strong visual stimulus we used, which may be due to direct neural inhibition of the visual cortex during sleep (Born et al., 2002).

In a heterogeneous group of infants and older children with suspected visual deficit, visual impairment was variably associated with smaller cortical volumes of activation on fMRI (Born et al., 2000). In former preterm infants with PVL, visual impairment evaluated near the time of fMRI correlated with abnormal BOLD response (Yu et al., 2011). Both of these studies are consistent with our results as shown in Figure 4.1, in which the control group had less activation during the visual task than the brain injury group. However, these studies were both done in infants/children with established visual deficits.

Neural networks can be characterized by observing temporal fluctuations in the functional MRI signal to identify areas that are functionally connected. Correlation found between two regions suggests that the regions work together and have similar temporal responses. Negative connectivity refers to temporally inverse correlation in BOLD fluctuation, where when one node demonstrates increased BOLD signal intensity relative to baseline, the other node shows lower

BOLD signal intensity relative to baseline. Functional connectivity can be computed for resting state BOLD time courses, as well as time courses derived from the task in the block design, as

42 we did here. In adults, networks have been identified in regions of the brain involved in vision, hearing, sensation, motor function, attention, and behavior (Damoiseaux et al., 2006). Resting state networks that correspond with those seen in adults are known to be present at term

(Fransson et al., 2007, Smyser et al., 2010) and even prior to birth (Thomason et al., 2015). A resting state network located in the medial aspects of the occipital cortex appears as early as 24–

26 week gestation (Smyser et al., 2010, Thomason et al., 2015) and is similar to the visual network found in the adult brain (Damoiseaux et al., 2006). Intact networks are thought to be necessary for normal neural processing, and disrupted networks have been seen in preterm infants as compared with normal controls (Smyser et al., 2010). We evaluated the connectivity of the visual system using task-based fMRI and found that healthy term neonates have a more robust visual network than infants with perinatal brain injury. We expect that this finding will have implications for visual outcomes in the infants with brain injury, though we will not have outcome measures from this group for another 2-3 y.

4.4 Conclusions

Infants with brain injury sustained in the perinatal period showed evidence of decreased brain activity and functional connectivity during a visual task as compared with healthy term neonates.

Chapter 5: Altered Functional Network Connectivity in Preterm

Infants: Antecedents of Cognitive and Motor Impairments?

This paper has recently been published with the following citation:

Gozdas, E., Parikh, N.A., Merhar, S.L., Tkach, J.A., He, L., and Holland, S.K. (2018). Altered

functional network connectivity in preterm infants: antecedents of cognitive and motor

impairments? Brain Struct Funct.

As primary author of the published work I have included the full text of the manuscript in this chapter, with permission of the publisher, via License Number 4387660662347 on July 14, 2018, and my coauthors. My contributions to the published work were essential to the publication and included, assistance with data acquisition, quality assurance of the reconstructed imaging data, all of the pre-processing and final analysis for functional connectivity. Pre-processing of this data set presented some particular challenges as noted in Section 1.4 of the Introduction.

Without these contributions, the work reported in this chapter would not have been possible.

Chapter 4 provided evidence that functional connectivity and brain activation in neonates with perinatal brain injury were altered during a visual task fMRI . In this chapter, we further tested whether preterm infants, without overt evidence of brain injury, would demonstrate alterations in connectivity measures both globally and in specific networks related to motor, language and cognitive function. To do this I applied functional connectivity and graph theory methods to resting state fMRI data (fcMRI) from premature infants with no anatomical imaging evidence of injury. Fifty-one healthy full-term controls and 24 very preterm infants without significant neonatal brain injury, were evaluated at term-equivalent age with fcMRI. Preterm subjects showed lower functional connectivity from regions associated with motor, cognitive, language

44 and executive function, than term controls. Examining brain networks using graph theory measures of functional connectivity, very preterm infants also exhibited lower rich-club coefficient and assortativity but higher small-worldness and no significant difference in modularity when compared to term infants.

5.1 Introduction

Preterm birth is a serious health problem, often resulting in significant morbidity and mortality and later cognitive and motor impairments (Botting et al., 1998, Hack 2009, Mansson and

Stjernqvist, 2014). Brain injury and altered brain development related to preterm birth have been well described using conventional and advanced magnetic resonance imaging (MRI), but predicting later adverse neurodevelopmental outcomes based on neonatal structural MRI remains difficult (Van’t Hooft et al., 2015 Parikh, 2016). Functional connectivity analysis of functional

MRI (fMRI) data shows promise as a sophisticated technique to elucidate more subtle adverse effects of preterm birth on the developing brain.

Numerous studies have demonstrated the cognitive deficits associated with preterm birth, including decreased memory and attention skills (Kerr-Wilson et al., 2012, de Kieviet et al.,

2012). Even preterm infants with normal IQ remain at high risk for school failure, possibly due to deficits in executive function as compared to full-term peers (Edgin et al., 2008, Burnett et al.,

2015). One in three children born prematurely also show signs of language delay even before school age (Foster-Cohen et al., 2010); motor impairments and cerebral palsy are also common

(Woodward et al., 2009, Moore et al., 2012, Synnes et al., 2015). These difficulties persist through adolescence and young adulthood (Breeman et al., 2015). Previous studies evaluating

45 alterations of functional brain networks soon after birth in preterm infants suggest a mechanism for these numerous issues (Doria et al., 2010, Smyser et al., 2010, Smyser et al. 2014). Few studies have directly related network alterations in the neonatal period to later outcomes, Rogers et al has related neonatal amygdala connectivity to later social processing at 2 years (Rogers et al., 2017).

The large majority of studies of functional brain connectivity in children born prematurely have reported the effects of preterm birth on motor, language, and cognitive development in older children, several years after preterm birth, and at school age (Wilke et al., 2014, Degnan et al.,

2015, Lee et al., 2011). Consequently, we are beginning to recognize the causal relationship between deficits in connectivity within brain networks that persist into school age in children born prematurely and neurocognitive deficits associated with these networks, even in the absence of anatomical lesions (Rowlands et al., 2016). Presumably, these infants exhibit microscopic/microstructural injury and/or delayed brain maturation soon after birth (Volpe JJ,

2009, Chau et al., 2013). The investigation of the effect of preterm birth on functional connectivity during the neonatal period remains poorly understood. However, recent investigations have demonstrated a relationship between alternations in structural connectivity

(i.e. white matter development) and poor cognitive and motor outcomes in children born prematurely (Fischi-Gómez et al., 2015, Liu et al., 2010, Chau et al., 2013, Parikh, 2016). The rationale for the current study is to identify deficits in functional connectivity in brain networks that exist soon after birth in infants born prematurely, that may underlie motor and cognitive deficits that emerge as the children develop. By examining specific networks that are now known to control sensorimotor, language and executive functions in older children and adults, we can

46 identify potential sources of future deficits in these domains in infants born prematurely, providing a potential biomarker for poor outcomes as well as an indicator of children who may benefit most from early interventions.

In this study, we investigated functional connectivity and network measures using graph theory in very preterm infants without significant structural brain injury and healthy full-term control infants using resting state fMRI data collected at term-equivalent age. We tested that preterm and full-term infants will differ in global network measures as well as specific network measures related to motor, language, and executive function, with decreased connectivity in the preterm group during the neonatal period. This will be tested by examining both whole brain and network-specific brain connectivity differences between pre-term and full term infants based on fMRI data obtained within 2 weeks of term equivalent age.

5.2 Methods

5.2.1 Participants

For this cross-sectional study, we included 24 very preterm infants (≤31 weeks gestational age) without significant brain injury on anatomic MRI from a longitudinal cohort study of very preterm infants cared for in the neonatal intensive care unit at Nationwide Children’s Hospital.

We compared these infants to a control population of 51 healthy full-term infants (38–41 completed weeks gestation), also conducted at Nationwide Children’s. Term control infants were recruited from two well-baby nurseries at The Ohio State University Medical Center and

Riverside Methodist Hospital. Infants with known structural congenital central nervous system anomalies, congenital chromosomal anomalies, or congenital cyanotic cardiac defects were

47 excluded. Additionally, we excluded full-term infants that were small or large for gestational age and those with a history of perinatal distress or complications. None of the term infants exhibited any signs of parenchymal brain injury or delayed development on anatomic MRI at term.

The Nationwide Children’s Hospital Institutional Review Board approved this study and written parental informed consent was obtained for every subject prior to imaging.

5.2.2 MRI acquisition

Full term infants (mean postmenstrual age at scan 40 weeks, range 40-42 weeks) and preterm infants (mean postmenstrual age at scan 39.8 weeks, range 39-41 weeks) were scanned on a 3T

Siemens Skyra scanner using a 32-channel head coil during natural sleep. Previously we have found it feasible to obtain high quality fMRI data from sleeping infants and toddlers. (Patel et al.,

2007, Tan et al., 2013, Deshpande et al., 2016, Merhar et al., 2016 ) It is not ethical to use sedation exclusively for research involving healthy, therefore sleep is a good option to minimize motion during MRI scanning in infants for research ( Dehaene-Lambertz et al., 2002). Inducing and maintaining natural sleep is a challenging element to add to any imaging experimental protocol. Nevertheless, development of these protocols has taken place at a few sites, including ours (Dehaene-Lambertz et al., 2002, Anderson et al., 2001, Redcay et al., 2007; Vannest et al.,

2014 ). Previous work in our laboratory demonstrates that fMRI is capable of imaging the status of auditory cortex and subsequent processing in response to sound stimulation ( Holland et al.,

2004, Patel et al., 2007 ). Indeed, auditory response is one of the last cortical functions to shut down under the influence of sleep or anesthesia and auditory responsiveness is used by anesthesiologists to assess level of sedation ( Malviya et al., 2006, Malviya et al., 2007).

Therefore a passive auditory stimulation paradigm was a natural choice to evaluate functional

48 connectivity for this study in infants( Cheour et al., 2004 ). We probed language networks using a well-characterized task ( Karunanayaka et al., 2007, Schmithorst et al., 2006): passively listening to stories (Story Listening Task). Given the similar (though attenuated) brain activation patterns we observed in the sleeping infants relative to our previous studies in sleeping( Merhar et al., 2016, Vannest et al., 2014, Wilke et al., 2003 ), sedated(Patel et al., 2007, Holland et al.,

2004, DiFrancesco et al., 2013) and awake children( Szaflarski et al., 2012, Karunanayaka et al.

2007, Vannest et al., 2009, Holland et al., 2007 ) using this task, we have a good measure of confidence that the BOLD effect is somewhat preserved in the sleeping infant cohorts studied here. Similar arguments apply to our assessment of sensory motor function in this study.

Although, objective measurements of the depth of sleep were not performed for this study, the acquisition of the resting state fMRI was begun when infants were already in a deep stage of sleep to insure that they would not be awakened by the scanner noise which as a weighted sound pressure level (SLP-A) of approximately 104 dB for the resting state fMRI protocol. A more detailed analysis of the literature on fMRI connectivity and natural sleep is provided later in the

Discussion.

High-resolution anatomical T1-and T2-weighted images were acquired for alignment of fMRI data and for inspection of brain anatomy. T1-weighted 3D MPRAGE images were acquired using TE=2.9ms, TR=2130ms, 1x1x1 mm3 voxel size, flip angle=12°, and FOV=174x192 mm2.

T2-weighted images were obtained with TE=147ms, TR=9500ms, 0.93x0.93x1.1 mm3 voxel size, flip angle=150°, and FOV=180x180 mm2. The resting state fMRI data was collected from a gradient echo planar sequence with TE=30ms, TR= 3000ms, FOV=212x212 mm2, matrix size=

104x104, slice thickness 2.5mm and 150 measurements.

Prior to quantitative analysis of resting state fMRI data for connectivity using graph measures, all anatomical, T1 and T2-weighted, images were reviewed by a pediatric neuroradiologist and a neonatologist skilled in reading neonatal MRI scans. No subjects with evidence of defects in white and/or grey matter were included in the subsequent analysis because anatomical lesions provide separate evidence for predicting poor cognitive or motor outcome and are also likely to affect brain connectivity.

5.2.3 Data analysis

SPM8 was used for pre-processing of the resting state functional imaging data. The functional image series were corrected for head motion by realigning all images to the mean of all functional volumes. Although the neonates were asleep during the fMRI procedures, head motion was still problematic and was corrected as described in detail in the section below. The mean functional image was co-registered to a corresponding T2-weighted high-resolution image.

T2-weighted images were segmented into gray matter, white matter, and cerebrospinal fluid

(CSF) using the UNC neonatal tissue probability maps (Shi et al., 2010). Tools developed for segmentation of adult brain images do not work as well in neonatal brain images due to reduced contrast between gray matter and white matter that arises from the higher water content and limited myelination in the neonatal brain. Consequently, we utilized a neonatal brain atlas developed specifically to address this low contrast condition by introducing age appropriate prior probability estimates (Shi et al., 2010). As a final step, the images were normalized to the neonatal template (Shi et al., 2011), re-sliced to a voxel size 2x2x2mm, and smoothed with a full width at half maximum (FWHM) Gaussian Smoothing kernel of 6mm. T1-weighted images in

50 neonates do not provide sufficient contrast for segmentation and co-registration and were not used in our analysis.

5.2.4 ROI-to-ROI Functional Connectivity

To test our hypothesis that preterm and full-term infants will differ in specific network measures related to motor, language, and executive function, during the neonatal period, we used ROI-to-

ROI connectivity methods to examine connectivity between all brain regions. The assessment of functional connectivity for pre-processed resting state data was carried out using the Conn functional connectivity toolbox (Whitfield-Gabrieli and Nieto-Castanon, 2012). Conn is a

Matlab-based toolbox for the computation, display and analysis of functional connectivity in fMRI (fcMRI). The residual BOLD time-series was extracted from gray matter voxels. The resting state fMRI time series from each voxel was corrected using a strict noise reduction method called aCompCor, which removed the principal components attributed to white matter and cerebrospinal fluid signals (Behzadi et al., 2007) and eliminated the need for a global signal regression (Chai et al., 2012, Murphy et al., 2009). In addition, subject-specific six motion parameter time-series and their first derivatives, and extreme outlier volumes detected with the threshold 1mm in ART ( http://www.nitrc.org/projects/artifact_detect/) were also introduced as potential confounds. ART is a graphic tool for automatic and manual detection of global mean and motion outliers in fMRI data. In addition, the residual BOLD time-series in each voxel was band-pass filtered at 0.008 to 0.08 Hz to focus on low frequency fluctuations (Fox et al., 2005).

Following these temporal preprocessing steps, an ROI-to-ROI functional connectivity analysis was performed by grouping voxels into the 90 ROIs defined specifically for the neonatal brain.

These 90 ROIs are defined objectively using a neonatal brain image atlas, neonate.aal

(Automated Anatomical Labeling) (Shi et al., 2011). The correlation matrix among these ROIs is then estimated for each subject by computing Pearson’s correlation coefficients between each

ROI time-series and the time-series of all other ROIs. The correlation values for each subject were Fisher transformed to normally distributed scores and used in subsequent analysis.

The normalized correlation coefficients were averaged within each group for each region and correlation matrices were created for each group for all 90 ROIs. Additionally, a two-sample t- test was run between the groups for each ROI-ROI pair, to examine whether differences in connectivity between full-term and preterm groups were significant for any of the 3960 possible connections. For this analysis an FDR-corrected threshold p<0.05, was used. All ROIs that reached this level of difference between preterm and full-term infants were then defined as seed regions for further connectivity analysis within the specific networks shown in Figure 5.1-3 ( 5 sensory/motor, 2 auditory/language, 6 cognitive/executive function,). These ROIs are listed in

Table 5.1 and identified as the black circles on the connectivity plots in Figures 5.1, 5.2 and 5.3.

5.2.5 Graph theory processing

To test our hypothesis that preterm and full-term infants would differ in global network measures during the neonatal period, we use graph theory to examine whole brain network connectivity based on global and local topological measures. The unweighted covariance matrix for these 90

ROIs for each subject from the Conn functional connectivity toolbox described above was used as input to the graph theory network analysis in the Brain Connectivity Toolbox

(https://sites.google.com/site/bctnet/). Both global and regional topological properties were

52 investigated using the graph theory parameters: global efficiency, local efficiency, degree, betweenness centrality, assortativity, small-worldness, modularity and rich-club coefficient.

These parameters characterize the properties of the network and the efficiency of information transfer within it. Graph theory parameters were tested for differences between the groups on a regional basis using the same 90 ROIs defined above in the correlation analysis. The specific graph theory parameters we examined are defined in Chapter 3.

Measures of network degree, global efficiency, local efficiency and betweenness centrality (BC), were computed for every region by means of graph theory analysis implemented in the Brain

Connectivity Toolbox. Additionally, assortativity, small-worldness, modularity and rich-club coefficient were computed for the entire brain using a cost threshold. Cost is a measure of the proportion of connections for each ROI in relation to all connections in the network.

A two-sample t-test was used to compare graph measures between full-term and preterm groups to determine if these parameters differed significantly over the entire brain or within any of the

90 ROIs tested. Regions where any of these parameters differed significantly are tabulated

(Table 5.2) and plotted (Figure 5.5).

Finally, to determine the influence of post-menstrual age and numbers of days post-delivery, both graph theory measures and ROI-ROI connectivity values were entered into a general linear model (GLM) to analyze these effects in full-term and preterm babies.

5.3 Results

The mean gestational age at birth for full term infants was 39.5 weeks (±0.9, range 38-41 weeks, median 39 weeks) and the mean gestational age at birth for preterm infants was 28.6 weeks 53

(±2.6, range 24-31 weeks, median 29 weeks). High resolution T2-weighted and T1-weighted,

MPRAGE MRI scans of the participants in both the preterm and full-term cohorts did not exhibit evidence of defects or lesions in white and/or grey matter.

The mean (SD) age of the mothers of the preterm group was 31.4 years (4.8) and for healthy term controls mean maternal age was 28.2 years (5.7) (P=0.02). Although this mean difference of

3 years is statistically significant, it is not clinically significant in terms of fetal gestation and is unlikely to affect brain development of the fetus. Five mothers of preemies (20.8%) and one mother of a term control infant (2.0%) stated they smoked during pregnancy (NS). Each of these smokers stated they smoked less than a half a pack per day. Two mothers of preemies admitted to using street drugs (8.3%). Both stated they used marijuana only. None of the moms of healthy term controls admitted to using street drugs. One mother of a preterm infant (4.2%) and 11 mothers of term controls (20.6%) stated they drank alcohol during pregnancy (NS). All but one of these mothers said they stopped once they learned they were pregnant; the other mother (of a term control infant) said she drank infrequently during her pregnancy (<2 alcoholic drinks per week).

Our hypothesis that preterm and full-term infants will differ in brain connectivity in specific networks related to motor, language, and executive function during the neonatal period was initially tested using whole brain analysis to identify any brain regions that differed significantly between the groups based on the two-sample t-test described above. Based on the ROI to ROI connectivity analysis we found that preterm infants differed from full-term infants in the expected brain networks. These differences are listed in Table 5.1 and shown as black circles in

Figure 5.1-3. No other ROIs or networks exhibited significant differences in brain connectivity based on the statistical thresholds defined by our rigorous methodology. All of these regions fall

54 within motor, language and cognitive networks as we hypothesized based on known behavioral outcomes from preterm birth as outlined in the Introduction. No other significant differences in brain connectivity were identified among the 90 ROIs tested. Using ROIs where significant differences were detected as seeds regions in a seed-to-ROI connectivity analysis, we further explored connectivity differences between the groups in each network.

The sensory/motor network exhibited decreased connectivity of the medial and dorsal superior frontal gyrus (SFGmed, SFGdor) (premotor area) (Picard and Strick 2001) bilaterally and left supplementary motor area (SMA) (Goldberg 1985) to other areas of the brain as compared with the term group. The sensory/motor seed ROIs are listed in Table 5.1 and indicated in black in

Figures 5.1 A-E, with connections that are stronger in the full-term infants than in preterm neonates highlighted in red. Conversely, increased functional connectivity was found in the preterm group relative to the full-term group between the seed region in the left SMA and left olfactory (OLF) and rolandic operculum (ROL) bilaterally; highlighted in blue in Fig 5.1E.

We also found differences in expressive and semantic language networks between preterm and full-term babies. The specific regional differences we found within the language network are listed in Table 5.1 and indicated by the black circles in Figure 5.2. Preterm infants showed decreased connectivity in expressive language regions (Friederici and Gierhan, 2013, Vigneau et al., 2006) in the left angular gyrus (ANG) (word reading and comprehension) and right fusiform

56 gyrus (FFG) (object recognition and reading) as compared with term controls; highlighted in red in Fig 5.2B and 2C. However, increased connectivity in preterm relative to full-term infants was found between the left middle temporal gyrus (MTG) (semantic processing) and hippocampus

(HIP) bilaterally and amygdala (AMYG) in the left hemisphere as indicted by the blue circles and lines in Fig 5.2A. Preterm infants also showed increased connection between right FFG and left thalamus (THA) and posterior cingulate gyrus (PCG) (Fig 5.2C-blue).

Connectivity within cognitive and executive function networks also differed between the groups of infants as indicated by the black circles in Figure 5.3A and 5.3B. Coordinates of these seed regions are listed in Table 5.1. Strikingly, we found decreased connectivity for the preterm group between the orbitofrontal cortex (ORB) (cognitive, executive function) (Kringelbach,

2005) as well as middle frontal gyrus (MFG) (working memory) (Fan et el. 2005) and multiple brain regions bilaterally (Fig 5.3 A and B -red). We also found increased connectivity for the preterm group between the right inferior ORB and right superior temporal pole (TPO) and right middle temporal pole (TPO) and between the right medial ORB and left superior TPO (Fig 5.3B

– blue).

Using graph theory analysis over the entire brain, we also found some significant differences between preterm and full-term infants. Assortativity was significantly decreased in preterm infants (p<0.05) across the full range of cost function values tested (Fig 5.4A). Differences in small-worldness (� > 1) were not significant over a range of cost threshold (0.06 to 0.3), but at the cost level of 0.08-0.11, preterm infants showed higher small-worldness (p<0.05) (Fig 5B).

Modularity (Q) was not significantly different between preterm and full term infants (Fig 5.4C) over the entire range of cost levels we tested of 0.06-0.3. Fig 5.4D shows that preterm infants exhibit characteristic rich-club organization (∅(�) > 1) for the range of � = 18 to � = 30 whereas full term infants exhibit characteristic rich-club organization (∅(�) > 1) for the range of � = 15 to � = 30. Rich-club organization was significantly reduced in preterm infants for values of the Rich-club level >15 (p<0.05).

Finally, we present region specific group differences in graph theory parameters in Figure 5.5.

By examining all 90 ROIs and all of the topological measures we calculated, we constructed the ball and stick representations shown in Figure 5.5, of the specific areas where differences occur in the graphical measures between groups. These ROIs are also listed in Table 5.2. Differences are highlighted showing full-term infants with increased global efficiency and degree in frontal networks (SFGmed, SFGdor, MFG, and superior ORB) (Fig 5.5A and B-red). Preterm infants exhibit increased global efficiency relative to full term in auditory, language networks (left

Heschl’s gyrus-HES, superior temporal gyrus-STG) and primary motor networks (precentral gyrus-PreCG) and increased degree in auditory and language networks (HES, STG) (Fig 5.5A and B-blue, respectively). Local efficiency and betweenness centrality were not significantly different between preterm and full term infants.

Regression analysis for

connectivity measures with

post-menstrual age and

numbers of days post-delivery

did not return any significant

influence from these factors

on the differences between

full-term and preterm babies

for either the ROI-to-ROI or whole brain analysis for any of the measures we calculated.

5.4 Discussion

The aim of this study was to identify topological differences in brain networks associated with very preterm birth, during the neonatal period of development that exist even in the absence of anatomical lesions and that might lead to the poor outcomes often observed in these children later during infancy and childhood. In particular, we focused on network measures and functional connectivity alterations in regions related to language, motor, cognitive and executive functions because these domains have been shown to be adversely affected in children born preterm as compared to their term counterparts. Alterations in network functional connectivity measures were found in the preterm group in all of the networks we hypothesized. Remarkably, although we began our analysis with a full set of 90 ROIs covering the entire brain and full range of brain networks, differences between the preterm and full-term infants were limited to the hypothesized networks.

Consistent with our hypothesis of decreased network connectivity in preterm infants, we found alterations in neural connectivity for motor regions in preterm infants (Fig 5.1). Furthermore, we found clear evidence of pervasive decreased connectivity in language and executive function pathways in the brains of very preterm infants at term-equivalent age, as compared to healthy full-term neonates (Figs 5.2 & 3). However, our data also highlight selected pathways of enhanced connectivity in the preterm group in each of these networks. Understanding why this occurs and how it affects long-term development could provide important insights about the optimal conditions and trajectories for environmental stimulation of preterm infants.

Importantly, we documented decreased functional connectivity from the left ANG and right FFG

(Fig 5.2B & C) in the preterm group. However, the preterm group had increased connectivity from the left MTG and right FFG to several other areas of the brain (Fig 5.2A & C).

Additionally, preterm infants had higher total number of connected links to the left STG than full term infants (Fig 5.5B). Similar findings were noted in preterm children at school age and during adolescence, with increased connectivity in Wernicke’s area in the left STG (left Brodmann’s area 22) using task-based fMRI (Gozzo et al. 2009, Myers et al. 2010). These data suggest either the engagement of a broader network of phonologic processing of language or a delay of semantic processing of language in preterm infants. Additionally, preterm born adults showed increased whole brain connectivity in left and right temporal parietal language regions (Scheinost et al., 2012). In addition to demonstrating increased connectivity, a recent functional connectivity study showed that infants born preterm showed decreased lateralization of connectivity in left frontal temporal language areas when compared the term born infants (Kwon et al., 2015). It is notable that in this study, the connections to reading and comprehension areas of the developing language network in the IFG, ANG and FFG appear to suffer from decreased network connectivity in preterm infants while semantic processing elements of the language network in the MTG seem to exhibit increased functional connectivity in the preterm group relative to the full-term infants. Considering possible explanations for this apparently inconsistent finding we are forced to consider the relative developmental advantage of allocating neural resources to receptive language processing in the newborn infant. Is it possible that enhanced connectivity between receptive language areas of the brain with attention and striatal and motor pathways conveys a developmental advantage in a preterm infant that warrants selective allocation of resources to this pathway? Alternatively, the increased connectivity may simply reflect the impact on brain development of the exposure of preterm infants to types and amounts of increased auditory stimuli that do not match the “expected” stimuli (i.e. in utero environment) for a given GA (Lawrence et al., 2010, Lubsen et al., 2011, Ment et al., 2006, Schafer et al., 2009).

Reduced connectivity was also found between regions related to cognitive and executive functions in the ORB and MFG bilaterally (Fig 5.3). We believe that these findings provide strong support for previous studies (Smyser et al., 2014, Doria et al., 2010) in which preterm infants demonstrated reduced covariance in fronto-parietal networks suggests that functional connections are less complex in preterm infants with attentional networks forming at different rates with gestation.

Using network measures from graph theory that reflect whole brain functional organization we demonstrated that preterm birth consistently alters network topology globally and that such effects are detectable by term-equivalent age. Comparing preterm and full-term infants at term- equivalent age, preterm infants showed reduced rich-club coefficient, assortativity and no significant difference in modularity (Fig 5.4A, D &C). Rich-club coefficient identifies the tendency for high degree nodes to be more densely connected among themselves than with other nodes. This is directly consistent with a recent finding that rich-club organization and assortativity lowered in preterm infants but modularity is not significantly different between preterm and full term infants (Scheinost et al., 2016). Although a recent study (Scheinost et al.,

2016) suggests reduced clustering coefficient in preterms, small-worldness was increased in preterm infants in our study, indicating increased clustering coefficient. Since network measures are very sensitive to cost threshold, we repeated the analysis across a range of cost thresholds that might effect results. Results remain reliable across a range of cost values. Furthermore, preterm infants had lower global efficiency and degree in frontal regions including premotor and cognitive function areas. However, connectivity within the auditory network appears to be enhanced in the preterm group. For this network, preterm infants exhibited increased global

63 efficiency and degree; findings that are consistent with the seed-based analysis presented in Figs

5.1-3.

As mentioned previously, a likely explanation for the increased ROI to ROI connectivity in the motor area in preterm infants (Fig 5.1 – blue) could be the earlier and additional time of exposure to sensorimotor stimulation ex-utero in the preterm group. At the time of MR imaging, preterm infants had been exposed to 9-16 weeks of direct sensory and motor stimulation that was not experienced by full term infants. Hence, it is conceivable that preterm infants formed additional connections to primary auditory and sensorimotor brain regions that are not present in the full- term infants. However, evidence of later speech and language delays as well as sensory and motor impairments in preterm children suggests that this early enrichment of network connections may not be complemented by projections of these areas to secondary brain regions necessary for temporal sequencing, coordination, association and other functions essential to optimal motor and cognitive function.

Increased functional connectivity in preterm relative to full term infants from left but not right

SMA could arise from several factors. As noted below, sleep tends to attenuate the BOLD fMRI signal, resulting in decreased functional connectivity. Therefore it is possible that the absence of bilateral differences in SMA connectivity within the motor network could arise from the decreased signal to noise of the fMRI signal due to sleep. Given the prevalence of right hand dominance in humans, it is also possible that left SMA is already becoming more strongly connected to other sensory, motor, visual and other brain regions than the right hemisphere analog. There is some evidence supporting this suggestion in adult subjects (Sarfeld et al., 2012)

The alterations in network characteristics revealed by this study clearly demonstrate that differences in brain networks associated with preterm birth can be observed as early as term-

64 equivalent age. Our results reflect regional and network-based differences in connectivity measures in preterm infants at term-equivalent age. These findings provide a possible explanation for later impairments in sensorimotor, cognitive, and behavioral function commonly seen in children who are born very prematurely (Lawrence et al., 2010, Lubsen et al., 2011, Ment et al. 2006, Schafer et al., 2009, Gozzo et al., 2009). The networks we examined are associated with sensorimotor and cognitive neural systems known to reflect a delay or alteration in maturation in preterm infants. Documentation of the developmental trajectory of the observed network alterations is an important first step towards determining the best way to improve network connectivity and developmental outcomes in terms of timing and methodology (Doria et al., 2010, He and Parikh, 2016, Keunen et al., 2017). As such, these results may inform the development of intervention strategies to minimize the long-term impact of premature birth on sensorimotor and cognitive outcomes. Alternatively, network measures could be further developed and used as early surrogate outcome measures for neonatal neuroprotective interventions (e.g., erythropoietin), rather than waiting for 2+ years for neurodevelopmental outcomes.

It is important to note that the differences in connectivity we have identified between preterm and full-term infants are not due to differences in head motion during the resting state fMRI scans. Differences in motion between groups have been shown to influence estimates of functional connectivity differences from resting state fMRI data (Power et al., 2015, Garrison et al., 2015, Power et al., 2013, Power et al., 2012). Motion effects have been shown to differentially decrease long-range connectivity, including characteristic path length and global efficiency while apparently causing increases in local efficiency and betweenness centrality. In our study, fMRI scans in both groups of infants were performed during natural sleep. When we

65 compared preterm and full-term infants, there were no significant differences in motion

(p=0.2839) as indicated by the root mean square (RMS) displacement and also we did not detect a significant effect of motion on the parameter estimates.

5.5 Conclusions

The findings provide evidence that functional connectivity exhibits deficits soon after birth in very preterm infants in key brain networks responsible for motor, language and executive functions, even in the absence of anatomical lesions. These functional network measures could serve as prognostic biomarkers for later developmental disabilities and guide decisions about early interventions.

Chapter 6: Developmental Changes in Functional Brain Networks from Birth through Adolescence

The studies described in Chapter 4 and 5 demonstrate that early in infancy, functional brain networks are significantly effected by perinatal brain injury and premature birth. These findings are among the first to demonstrate these effects so early in life. In this chapter, we will examine the influence of age, sex and neurocognitive measures on developmental changes to the global and regional topology of functional brain networks derived from fMRI data in healthy children in the age range of birth to 18 years. In these experiments I have applied graph theory analysis which has emerged as a new approach to study brain development, cognitive function and neurophysiological disorders using fMRI data from the brain at rest. We observed that global efficiency and rich-club coefficient increased with age and local efficiency and small-worldness decreased with age, while modularity at the global level showed an inverted U-shaped trajectory during development. Boys had higher values of local efficiency, small-worldness and modularity at a global level than girls throughout development. We also examine the effects of neurocognitive measures in boys and girls globally and locally. These results and their interpretation provide new insight to understand the maturation of functional brain connectome and its relation to cognitive development from birth through adolescence.

6.1 Introduction

Graph theory approaches has been used in the previous studies to analyze fMRI data have demonstrated age-related changes in the brain functional connectome. Some of the earliest

67 studies examining the efficiency of functional networks in younger and older adults reported that older adults showed decreased global and local efficiency (Achard and Bullmore, 2007) while network modularity is not significantly changed from younger adults to older adults (Meunier at al., 2009). More recently topological properties of the functional connectome have been found to change significantly during child development (Bullmore and Sporns 2009, 2012, Di Martino et al., 2014), raising the question of how they emerge and change from birth to adolescence.

Some recent studies have shown that sex and cognitive functions correlate with specific features of functional brain networks during development. Examining sex effects on intelligence and functional connectivity in language networks boys with higher IQ exhibited more modular functional architecture with age, while girls with higher IQ exhibited a more connected functional architecture with age (Schmithorst and Holland, 2006). Similarly, girls developed a greater interhemispheric connectivity for intelligence with age, while boys developed a greater connectivity in the left inferior frontal gyrus (Schmithorst and Holland, 2007). There is also evidence that significant sex differences was indicated in the network properties of structural and functional brain networks in adults (Smith et al., 2014). Based on these earlier studies on the interaction of sex and developmental on trajectories of brain networks, we sought to explore the influence of age, sex and cognitive ability on more specific topological measures of developing brain networks from birth through adolescence. Using graph theory to analyze r-fMRI data from

189 healthy children aged from 0 to 18 years. Following age appropriate normalization and parcellation of the brain imaging data into 200 regions of interest, we estimated functional connectivity from the temporal correlations between the fMRI time series from pairs of the 200 regions. The resulting correlation matrix from each individual was used to construct a binary

68 network reflecting the topological organization of a functional brain network. Specifically we tested that sex and neurocognitive measures would influence topological properties of the functional brain connectome at both the global and regional levels. Using graph theory, we tested whole brain functional network topology for differences in global network properties between boys and girls and cognitive ability. Hypothesizing a significant relationship between specific brain networks and developing cognitive and language skills, we further explored the influence of IQ and language measures on specific brain networks where significant relationships were detected in network topology.

6.2 Methods

6.2.1 Participants

One hundred eighty-nine participants were selected from Cincinnati MR Imaging of

Neurodevelopment (C-MIND) (http://research.cchmc.org/c-mind) database of functional neuroimaging and behavioral data from typically developing children. The C-MIND database only contains data from normally developing children with strict exclusion of chronic illness, gestation less than 37 or greater than 42 weeks, birth weight less than the 10th percentile, history of head trauma with a loss of consciousness, special education, orthodontic braces or other metallic implants and standard MRI compatibility contraindications. The study was approved by the Institutional Review Board at Cincinnati Children’s Hospital Medical Center; informed consent and assent (where appropriate) were obtained from a parent/ guardian and the participant. In participants of age 5 years and under, additional preparation time was allowed at each session to acclimate the child to the scanner environment. During this time, study personnel used a step-by-step approach to introduce the equipment, including the headphones and

69 microphone for in-scanner communication. Details of methodology for scanning young children are available (Vannest et al., 2014). Participants between the ages 0 to 2 years of age were scanned while asleep. Participants between the ages 3 to 18 years of age were scanned while awake. All participants were native English speakers with demographics reflecting the regional racial and ethnic distribution (64% Caucasian, 23% African-American, 9% Multi-ethnic, 1%

Asian, Pacific Islander, 2% unknown) and a median household income of $62,500, see Figure

6.1 for detailed demographic information.

6.2.2 MRI acquisition

156 subjects were imaged on a Philips 3T Achieva system and a 32-channel head coil. 33 subjects were imaged on a Siemens Magnetom Trio-Tim scanner. EPI–fMRI scan parameters between the scanners were homogenized to minimize systematic variation in the data from the two machines. TR/TE = 2000/35 ms, FOV = 24 cm × 24 cm, matrix = 80 × 80, and slice thickness = 4 mm, yielding spatial resolution of 3 x 3 x 4 mm. Thirty-six slices were acquired, covering the entire brain. One hundred fifty whole-brain volumes were acquired. Prior to acquisition of the functional image data, the scanner was programmed to execute a “dummy

70 scan” interval consisting of 2 TR periods, for which the scans were not recorded. In addition, the first acquired image was also discarded during post-processing to insure image contrast at relaxation equilibrium. Total scan time for resting state acquisition plus dummy scans was 5 min

8 s. Techniques detailed elsewhere (Byars et al., 2002, Vannest et al., 2014) were used to acclimatize the participants to the MRI procedure and render them comfortable inside the scanner. Soft head restraints were used to minimize head motion. In addition to the fMRI scans, whole-brain T1 weighted inversion recovery fast gradient echo scans were acquired for anatomical co-registration. All imaging was performed on 3 T MRI scanners. Comparison of resting state fMRI data between scanners was performed as described below to insure compatibility of the data sets.

6.2.3 Data analysis

6.2.3.1 Preprocessing

For the resting state functional connectivity analysis, all images are preprocessed in SPM8. The resting state functional image series were corrected for head motion by realigning all images to the mean of all functional volumes. The data of 9 subjects were discarded from further analysis due to head motion that exceeded a mean framewise displacement(mFD) threshold of >1mm.

Among the remaining 189 subjects, the head motion parameters showed no significant correlation with age. The head motion corrected functional data were processed as follows. The mean functional image was co-registered to a corresponding T1-weighted high-resolution image.

To optimize segmentation and normalization across the age span of all subjects, we created age specific templates for five age groups: 0-2 years, 3-5 years, 6-8 years, 9-12, and 13-18 years.

After motion correction, first structural images were brain-extracted and next tissue-type

71 segmentation was implemented. The resulting images were then aligned to MNI space using nonlinear registration and then averaged to create an age-specific template. T1-weighted images were re-segmented into gray matter, white matter, and cerebrospinal fluid (CSF) using the age specific probability maps and then all functional images were normalized to this template and resampled to voxel size 2x2x2 mm. Subsequently, normalized functional images were smoothed with 8 mm FWHM kernel.

6.2.3.2 Functional Brain Networks Construction

A major consideration in graph theory approaches to functional brain imaging data is how to define the nodes of the network. Many previous studies have used anatomical atlases as nodes

(Merhar et al., 2016, Wu et al., 2013). However, anatomical atlases often have poor functional homogeneity and do not present functional connectivity patterns accurately (Craddock, 2012). A whole brain parcellation method was recently introduced based on a data-driven approach to generate functionally homogeneous regions of interest (ROIs) by spatially parcellating whole- brain resting state fMRI data into regions of maximal coherence in the time course of voxels within each ROI (Craddock et al., 2012). This method can be used to create a parcellation atlas of any desired number of ROIs, though it has been shown that 150-200 ROIs provides a good compromise between functional homogeneity of the regions, interpretability of connectivity results and computational data reduction. We implemented this method for parcellation of normalized whole brain data. The pyClusterROI software

(http://ccraddock.github.io/cluster_roi/) was used to parcel whole brain functional data of each group into 200 regions.

6.2.3.3 ROI-to-ROI Functional Connectivity

With 200 ROIs defined in the normalized and pre-processed resting state data brain imaging from all subjects, we computed functional connectivity using the Conn functional connectivity toolbox (Whitfield-Gabrieli and Nieto-Castanon, 2012). The residual BOLD time-series was extracted from gray matter voxels within the 200 defined ROIs in the preprocessed resting state fMRI data. The resting state fMRI time series was corrected on a voxel-by-voxel basis using the anatomical component correction (aCompCor) method, which removes the principal components attributed to white matter and cerebrospinal fluid signals (Behzadi et al., 2007) and eliminates the need for a global signal regression (Chai et al., 2012, Murphy et al., 2009). Additional post- processing of time series data was performed using the Artifact Detection Tool (ART) (Mazaika et al., 2009) to identify image frames at the subject level with extreme motion (>1mm relative to time series average) as outliers. These frames as well as six subject-specific motion parameter and their first derivatives, were also calculated as potential confounds in subsequent analysis stages. The residual BOLD time-series in each voxel was band-pass filtered at 0.008 to 0.08 Hz to focus on low frequency fluctuations (Fox et al., 2005).

Following the temporal preprocessing of fMRI data, an ROI-to-ROI functional connectivity analysis was performed by grouping voxels into the 200 ROIs defined from the age specific atlases, and estimating the 200 x 200 correlation matrix for each subject by computing Pearson’s correlation coefficients between each ROI time-series and the time-series of all other ROIs. The correlation maps for each subject were Fisher transformed to normally distributed scores and submitted to one sample t-tests to compute group correlation maps for each ROI.

6.2.3.4 Correlation of Network Topology with Age, Sex and Behavioral Measures

To assess whether the observed changes in network properties relate to age, sex or neurocognitive abilities in the children, we evaluated the effects of association of these measures with global and regional network properties. For each child, the age in months at the time of the resting state fMRI scan and the sex were extracted from the C-MIND database. All children in the C-MIND study participated in an extensive neurocognitive assessment battery within a few weeks with respect to the functional MR imaging session. Neurocognitive testing included tests of general intellectual function (Wechsler Intelligence Scales: Wechsler Pre- school & Primary

Scale of Intelligence, Third Edition (WPPSI-III, ages 2:6–5); Wechsler Intelligence Scale for

Children, Fourth Edition (WISC-IV, ages 6–16) Wechsler Adult Intelligence Scale, Fourth

Edition (WAIS-IV, ages 17 and 18, Wechsler 1981), vocabulary (the Expressive Vocabulary

Test-2 (EVT-2) ( Williams, 2007). We found no significant correlation between age and neurocognitive measures (p>0.5) and no significant difference between boys and girls (p>0.6).

Scores from these tests will next be included in statistical analysis models with brain network measures to pin point brain regions and connections that facilitate greater cognitive ability during development. Younger participants were excluded from the analysis because neurocognitive data were not available.

6.2.3.5 Statistical Analysis

To determine the changes in the functional networks across development, a general linear model

(GLM) was applied to analyze the effects of age, sex and neurocognitive measures and their interactions on the network parameters. We performed multiple linear regressions which included age, sex and neurocognitive measures and their interactions as covariates to detect the

74 linear and quadratic developmental trajectories. The GLM models were determined as follows:

�

To examine the sex-related differences and their development we used another GLM model including age, sex and sex-by-age interactions to examine both positive (boys>girls) and negative (girls>boys) contrasts as well as positive and negative age-by-sex interactions.

�

For the analysis of the IQ-related differences, we first applied a multiple linear regression � on both IQ and each network parameter to model the effects of age, sex and their interactions.

All statistical analysis were performed in RStudio (https://www.rstudio.com). For statistical evaluations a significance threshold of p<0.05 was used.

6.3 Results

6.3.1 Age and sex effects on global network properties

Brain functional network topology correlated significantly with age during development. Figure

6.2 demonstrates the developmental trajectories of key global network parameters from 0 to 18 years of age. Children who slept during the r-fMRI scan are indicated by open circles as the data point in Figure 6.2 to emphasize sleep as a potential confound for parameter estimates in this part of the curves. Positive age related changes (p<0.0001) were found in global efficiency (Figure

6.2A), but negative age related changes were found in local efficiency and small-worldness

(Figure 6.2B and C). Modularity exhibited negative quadratic age effect (inverted U-shaped)

(p=0.0005145) (Figure 6.2D).

We found significant sex differences (p<0.05, corrected) in local efficiency, small-worldness and modularity in which the boys showed significantly higher values when compared with the girls

(Figure 6.3B, C and D). Age-by-sex interaction in global network measures as reflected in global efficiency and small-worldness did not reach significance (Figure 6.3A and C).

Using degree as a measure of which nodes are most highly connected within a functional brain network and betweenness centrality, we identified the hub regions across all subjects. We found that the hubs differed slightly in the youngest group, but the brain hubs were predominantly located in the default mode, attention related regions and visual cortex (Figure 6.4A-dark dots).

The rich club architecture was also observed in the topology as the hub regions were more densely connected among themselves than with other brain regions. We observed that the rich club architecture increased during childhood (p=0.04149) with high variability from 5 to 12 years (Figure 6.4B). No significant sex differences were found in rich club coefficient of the functional brain networks across the age span.

6.3.2 Neurocognitive measures and sex effects

Examining all participants together we found no significant IQ effect on global network measures. However, when boys and girls were analyzed separately, significant IQ-related differences were observed in relationship with global efficiency, local efficiency, small- worldness and modularity. For the developmental trajectories, the boys had higher local efficiency, small-worldness and modularity but lower global efficiency.

When boys and girls are examined together we find regional positive correlations between IQ and the local graph measures in specific regions in interest. Regions with significant correlations in this analysis are defined by functional parcelleation as described above but labeled using anatomical brain atlases for reference purposes. Linear decreases between global efficiency and

IQ, and degree and IQ, but linear increase between average path length and IQ were found in the left middle temporal gyrus (MTG(L)). Linear increases were also found between local efficiency and IQ, and clustering coefficient and IQ in the right superior frontal gyrus (SFG(R)) (Figure

6.5A-E). Significant positive quadratic changes were found in the right posterior parietal cortex

(PPC(R)), lateral prefrontal cortex (LPFC(R)), and inferior frontal gyrus (IFG(R)) between local efficiency and IQ, and clustering coefficient and IQ (Figure 6.5F-K).

Treating boys and girls as one group, we also identified that there were correlations between specific brain regions and language ability as reflected in the EVT-2 measure. There was a negative linear correlation between nodal global efficiency and EVT-2 in the left inferior frontal gyrus (IFG(L)). In contrast, positive linear correlations were found between the average path length in IFG(L) and IFG(R)) and EVT-2, and small-worldness, clustering coefficient and IQ in

IFG(L). Similarly, correlations between average path length and IQ, and clustering coefficient and IQ exhibited positive linear changes, but negative quadratic change between small-worldness and EVT-2 in the left posterior superior temporal gyrus (pSTG(L)) (Figure 6.6). There was no significant sex effect between nodal network parameters in language regions and EVT-2. In this analysis 113 subjects aged between 3-11 years were included.

6.4 Discussion

In this study, we investigated the topological organization of functional brain networks derived from r-fMRI in healthy children in an age range from 0 to 18 years and analyzed the effects of age, sex and neurocognitive measures on the network properties both globally and locally. The main findings included the following: (i) functional brain networks from birth through adolescence showed linear increases in global efficiency and rich-club coefficient, but linear decreases local efficiency and small-worldness, while modularity showed negative quadratic change (Figure 6.2) ; (ii) the boys had a higher local efficiency, small-worldness and modularity when compared with the girls (Figure 6.3); (iii) the functional brain hubs were located in default mode, fronto-parietal and visual cortex over development (Figure 6.4); (iv) the regional network 80 parameters positively correlated with several neurocognitive measures (Figures 6.5 & 6).

Collectively, we detected significant topological modifications of the human brain connectome from birth through adolescence which appear to be influenced by sex and cognitive abilities.

Earlier studies have reported age-associated changes in the global network properties showing that local efficiency and small-worldness were reduced while global efficiency was not affected by age (Cao et al., 2014, Geerligs et al., 2014). In this study, we find that global efficiency of the brain functional networks increased but local efficiency decreased with age. A high global efficiency indicates the capacity of rapid information exchange among the distributed elements

(Bullmore and Sporns, 2012). In contrast to global efficiency, a high local efficiency reflects efficient information transfer in the immediate neighborhood of each node (Latora and

Marchiori, 2001). Previous functional brain network studies have reported no significant change in global efficiency during development and adulthood and increases in local efficiency during development and its reduction with aging which are partly in accordance with our findings (Cao et al., 2014). Differences between those results and our findings may be related to the regional parcellation of the brain that was used in the previous study for graph construction (90 vs. 200 nodes in the present study), which has a limited ability to represent functional networks due to coarse nodes which encompasses different functional areas (Power et al., 2011). In addition, we detected a negative age effect on small-worldness during childhood which is consistent with previous results that have shown a reduction in small-worldness with age. Recent studies showed that small-worldness was established in functional brain networks early during development

(Scheinost et al., 2016, Fransson et al., 2011), remaining stable in children and young adults (Wu et al., 2013). Therefore, our study provides further support for the previous results that the

81 functional brain networks display consistent small-world properties from birth through adolescence.

Along with these changes, functional brain networks were found to be highly modular reflecting higher intra-network connectivity than inter-network connections in younger adults. In this study we have found that modularity changed throughout brain development (Figure 6.2D) which aligns with previous findings (Fair et al., 2009, Cao et al., 2014). Similarly, we detected modularity with an inverted U-shaped developmental trajectory, indicating that the brain’s functional modularity increased until approximately 8 years of age and decreased at older age.

This indicates that functional brain networks may become less differentiated with age which is also in accordance with previous studies (Geerligs et al., 2014).

Highly connected hub regions in the “rich-club” play a crucial role in global information integration among parallel and distributed brain networks. Previous studies observed that the brain hubs of functional networks were mainly located in the default mode, fronto-parietal and visual cortex (Cole et al., 2010; Liang et al., 2013, Zuo et al., 2012). Also one recent study suggested that functional brain hubs were stable over development (Hwang et al., 2012). Our results in Figure 4 are consistent with these findings. Recent work elucidated the rich-club configuration of human brain structural and functional networks in healthy adults (Collin et al.,

2013, van den Heuvel et al.,2012, Cao et al.,2014). We also observed the existence of rich-club organization in the functional connectome during childhood with a significant, positive age effect, in accordance with previous studies (Cao et al., 2014). The rich-club organization provides an important contribution to information transfer between different brain regions.

Previous fMRI studies indicated that functional interactions between multiple brain regions are strongly related to the neurocognitive measures during a task or at rest (Gray et al., 2003, Song et al., 2008). Previous research has shown that a positive correlation between IQ and fMRI brain activation in the middle temporal gyrus for a verb generation task (Schmithorst and Holland

2006). Therefore, in addition to age and sex effects we investigated the correlation between behavioral measures and functional brain network topology. No significant correlation was observed between full-scale IQ and global network measures. However, when we examined girls and boys separately, we found significant IQ related differences with age between boys and girls. This is not consistent with the recent study (van den Heuvel et al., 2009) which showed that intelligence is highly correlated with both local and global efficiencies of functional brain networks in adults. IQ-related changes in the regional nodal parameters were found in the frontal, parietal and temporal brain regions (LPPC, PPC, MTG, STG and IFG). Consistent with our results, Wu and colleagues also reported IQ-related changes in local nodal parameters in frontal and temporal regions during development (Wu et al., 2013). Also a previous study on functional connectivity indicated that brain network connectivity is related to intelligence (Song et al., 2008). Furthermore, several brain regions mainly involved in attention showed a positive correlation with IQ. Though we did not find local network correlations with IQ, we did discover significant correlations between language ability (EVT-2 standard scores) and network topology in several language-related brain regions (IFG-L, STG-L).

Overall, our finding suggests that topology of brain networks related to attention, memory and language, underlie differences in intelligence and language ability in children throughout development.

In prior studies of developmental changes in structural and functional brain networks, the effect of age has been demonstrated in different age groups (Cao et al.2014, Wu et al. 2013, Fransson et al., 2011, Meunier et al., 2009). However, the interaction of brain network topology with sex and cognitive abilities through development have not been studied thoroughly. Our study adds to the understanding of functional brain network development from birth through adolescence in healthy children by examining the growth trajectories of global network topology as a function of age and sex during childhood. An important observation is that key graph measures characterizing global and regional brain network topology do not appear to follow linear or monotonic trajectories as networks organize for improved efficiency during childhood.

Chapter 7: Conclusions and Future Research

Overall, the main focus of this dissertation has been the application of advanced brain connectome methods to mapping the developing connectome in order to construct a better model for the organization of functional brain networks through child development. In a series of complex imaging experiments I have been able to map the developmental trends in brain connectivity in children from the gestational age of 25 weeks (15 weeks before full-term) to 18 years; when the brain is nearly fully mature. In this concluding chapter, the contributions of this dissertation are summarized and opportunities for future research directions are outlined.

7.1 Conclusions

Taking the work presented in this dissertation from earliest to latest stages of development, consider first the work presented in Chapter 5 (Gozdas et al., 2018). This work demonstrates that resting-state brain networks are already well established early in development in very preterm neonates, but that functional connectivity and network characteristics in preterm infants differ from those of full-term infants by term-equivalent age. Using functional connectivity analysis and graph theory, we also demonstrated significant differences in key brain networks between preterm and full-term infants, even at this early stage in development. Previous studies have examined brain networks of prematurely born children later in life, at the age of 7, or 9 years, or during adulthood. But the work presented in Chapter 5 is the first to report observations during the neonatal period. Previously it was assumed that these aberrations were present from the earliest stages of life, but our work proves that this assumption is correct and provides new insights into the consequences of prematurity as well as possible considerations for

85 early interventions. It is possible that slight differences in the post-menstrual age at the time of the MRI scans could have influenced the findings of connectivity differences between groups due to the rapid brain development that occurs during the neonatal period. We, therefore, examined the influence of post-menstrual age and did not detect a significant effect of postmenstrual age on any of the connectivity parameters. Many other corrections applied to the data to avoid confounding influences of scanning protocols, normalization of infant brains to standardized brain space, and definition of relevant regions of interest provide a high degree of confidence in the findings presented in Chapter 5, which have now been published in the journal,

Brain Structure and Function (Gozdas et al., 2018).

In Chapter 4, we examined the impact of brain injury during the perinatal period (i.e. within a few weeks before or after birth) on brain networks. In this work, we used a different approach, choosing an active brain stimulation paradigm with a controlled visual stimulus that is known to cause activation of the visual cortex. We showed that this visual paradigm can reliably cause negative activation in the visual cortex during natural sleep in neonates scanned at <48 weeks postmenstrual age and that infants with brain injury sustained in the perinatal period have altered task-based functional connectivity and decreased brain activation in the occipital cortex as compared with healthy term neonates. This work provides further evidence for the early and profound influence of early brain insults on the developing brain. This work may provide explanations for some of the prior literature in which alterations in various brain networks have been observed in older children with history of early brain injury. While the long term cognitive and sensorimotor impact of such injuries is well known (e.g. cerebral palsy is one consequence of perinatal brain injury to the sensorimotor cortex); the work described in Chapter 4 is arguable

86 the first of it’s kind to demonstrate these effects within the first two months of life. Continuing, longitudinal observations in the cohort of children described in Chapter 4 are underway now that they are more than 2 years old and can be tested for cognitive and motor outcomes. The results of that work will lead to stronger conclusions about cause and effect of early brain network alterations and developmental outcomes.

Finally, working with a large cohort of normally developing children without brain injury or prematurity we examined normal development trajectories of brain networks in children from birth to age 18 using the advanced connectome tools described in Chapter 3 and applied in the studies of Chapters 4 and 5. Chapter 6 examines age- and sex-related trends in the developmental trajectories of the functional brain connectome and show different rates of maturation of functional brain networks in boys and girls. We also explored the correlation of neurocognitive measures and the regional network properties by combining neuroimaging data and behavioral measures. This approach provides novel insight into the neural underpinnings of behavioral and cognitive variability in individuals during development. Chapter 6 reveals the dynamics of the functional brain connectome globally and locally and provides a background for evaluation of network anomalies associated with neurological disorders in children.

7.2 Future Research Directions

The work reported in this dissertation and now published in several scientific manuscripts, lays a solid foundation for future research exploring the development of brain networks in young children. But there is much more work to be done to expand on these preliminary findings and to further explore the impact that atypical developmental conditions might have on the brain

87 during the early stages. In Chapter 4, our study includes small sample size, heterogeneous etiology of brain injury, and lack of longitudinal outcome data to determine long-term visual function. Due to the small sample size and variability of pathology, visualization of results in the figures is highly threshold dependent. We used a strict threshold in order to avoid misrepresenting sources of activation. Despite the considerable heterogeneity of the groups, we still found a significant difference in patterns of activation and functional connectivity, suggesting that the effect is strong despite the considerable variance. We have shown on a group basis that fMRI can discriminate between controls and infants with brain injury, but in order to individualize prediction (for example, using machine learning), a larger sample size and a more homogenous population using a prospective, longitudinal design would be necessary. It is possible that the flash stimulus we used may be biased to identify anterior visual pathway disease

(such as optic nerve and retinal damage) and future directions could include more complex visual stimuli to obtain a maximal response from V1 and association areas. As mentioned above, the longitudinal experiments in these subjects is currently underway. A future study with a larger, longitudinal cohort will provide more conclusive results and better guidance for physicians and parents caring for infants with perinatal brain injury.

The premature infants discussed in Chapter 5, would ideally be followed longitudinally as well.

Then it would be possible to obtain outcome measures in each child at key developmental age points during preschool, kindergarten and school age years. Sensorimotor and cognitive outcomes were not available to us in this study because most participants are still too young to provide reliable test measures. The sample size for the preterm cohort of 24 infants is also relatively small. This limits the significance of some findings. With a larger sample size, we

88 would expect to find more pervasive differences in brain networks and additional graph measures that exhibit significant differences on a smaller scale. However, the current study provides evidence that such network alterations exist during the earliest stages of development and therefore provide important support for further exploration of graph theory measures as predictive biomarkers.

While the sample size of normally developing children reported in Chapter 6 is substantial and larger than any study of normal brain connectivity development to-date, the variance in network topology estimates is large as illustrated in the scatter plots in the figures. Consequently, the parameters describing the quadratic regression fits to the trajectories have substantial uncertainty.

While the figures clearly illustrate the dynamic trends of brain network topology through development, they are not sufficiently precise to allow individual predictions about brain age, IQ or other cognitive outcomes. In this respect the results of Chapter 6 can be considered as a starting point for studies of developing network topology, with solutions presented to some of the most challenging problems of working with data spanning such a wide range of development.

WPPSI-III and EVT-2 IQ and language measures were only administered to children age 30 months and older. The sample size for testing IQ and language effects on global and regional brain network topology is limited. While we found differences in developmental trajectories for global network measures when we treated boys and girls separately, the smaller sample size for the IQ and language measures may be responsible for not detecting these differences as a function of cognitive ability in the sample tested. A large sample should provide more exact estimates of average trajectories as well as allowing for additional effects to be tested and potential for training algorithms to make individual predictions of outcomes. Longitudinal

89 studies within individuals through development rather than a cross-sectional study as described here should reduce variance in growth trajectories (Szaflarski et al., 2006, Szaflarski et al., 2012) of human brain connectomes during development and both of the studies mentioned above include longitudinal components. It is also worth mentioning that several large-scale studies are now underway in the US and Europe that will enroll tens of thousands of individuals in longitudinal imaging studies of brain development. As that data becomes available, the work reported in this dissertation and its supporting publications will provide a reference for how such data should be treated.

As noted, children under the age of 3 years included in our studies were asleep during the fMRI scan. Ethical standards prohibit the use of medication-induced sedation for research involving children, so natural sleep is the only viable condition under which the youngest children can be engaged in fMRI research( Dehaene-Lambertz et al., 2002). Functional brain connectivity during sleep has been studied extensively and the findings are somewhat inconsistent. For example, a significant effect of resting state connectivity was detected using EEG as well as fMRI during deep sleep (Dang-Vu et al., 2008). Similarly, some reduction has been observed in functional connectivity within default network during deep sleep (Horovitz et al., 2008). On the other hand, a recent work has shown that functional connectivity within attention, default and executive networks and primary sensory networks were maintained from wakefulness to light sleep

(Larson-Prior et al., 2009). A recent study in infants has reported no significant differences between functional connectivity in deep sleep and sedation (Doria et al., 2010). In our experience sleep and sedation both have the impact of attenuating the BOLD fMRI signal itself as well as selectively influencing connectivity in specific brain networks mediating attention and

90 consciousness. ( Wilke et al., 2003, DiFrancesco et al., 2013) However, we have also demonstrated that it is feasible to obtain robust and meaningful brain activation maps from naturally sleeping infants in Chapter 4. Recently Laumann et al. have concluded that “resting- state BOLD correlations do not primarily reflect moment-to-moment changes in cognitive content. Rather, resting-state BOLD correlations may predominantly reflect processes concerned with the maintenance of the long-term stability of the brain's functional organization” ( Laumann et al. 2016) Given this background and remaining uncertainty, we emerge with some confidence that connectivity within the predefined networks we have selected remain somewhat intact during natural sleep, with some level of attenuation. Sleep is likely to reduce the significance of

ROI-ROI correlations and resulting functional connectivity and differences between groups.

However, this suggests that our results would be stronger in a wakeful state if that could be achieved in this young cohort.

A major contribution of our work is to demonstrate to the scientific community that exploration of functional connectivity in infants and toddlers is possible using non-invasive fMRI methods and advanced analysis methods. As more developmental neuroscientists take note of our work and expand on it with larger, longitudinal samples, a more complete picture of the relationship between developing brain networks and developmental, behavioral milestones will emerge.

Genetic and environmental factors clearly have an impact on child brain development as well.

We have not been able to consider such factors in our analysis. Ideally, such factors and others combined with the methods we have now outlined will facilitate studies that lead to improved strategies for fostering the best possible developmental outcomes for all children.

Bibliography

Achard, S., Bullmore, E., 2007. Efficiency and cost of economical brain functional networks.

PLoS computational biology 3, e17.

Altaye, M., Holland, S.K., Wilke, M., Gaser, C., 2008. Infant brain probability templates for

MRI segmentation and normalization. Neuroimage 43, 721-730.

Anderson, A.W., Marois, R., Colson, E.R., Peterson, B.S., Duncan, C.C., Ehrenkranz, R.A.,

Schneider, K.C., Gore, J.C., Ment, L.R., 2001. Neonatal auditory activation detected by functional magnetic resonance imaging. Magnetic resonance imaging 19, 1-5.

Bassi, L., Ricci, D., Volzone, A., Allsop, J.M., Srinivasan, L., Pai, A., Ribes, C., Ramenghi,

L.A., Mercuri, E., Mosca, F., 2008. Probabilistic diffusion tractography of the optic radiations and visual function in preterm infants at term equivalent age. Brain 131, 573-582.

Beckmann, C.F., DeLuca, M., Devlin, J.T., Smith, S.M., 2005. Investigations into resting-state connectivity using independent component analysis. Philosophical Transactions of the Royal

Society of London B: Biological Sciences 360, 1001-1013.

Behzadi, Y., Restom, K., Liau, J., Liu, T.T., 2007. A component based noise correction method

(CompCor) for BOLD and perfusion based fMRI. Neuroimage 37, 90-101.

Biswal, B., Zerrin Yetkin, F., Haughton, V.M., Hyde, J.S., 1995. Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magnetic resonance in medicine 34,

537-541.

Biswal, B.B., Kylen, J.V., Hyde, J.S., 1997. Simultaneous assessment of flow and BOLD signals in resting-state functional connectivity maps. NMR in Biomedicine 10, 165-170.

Blamire, A.M., Ogawa, S., Ugurbil, K., Rothman, D., McCarthy, G., Ellermann, J.M., Hyder, F.,

Rattner, Z., Shulman, R.G., 1992. Dynamic mapping of the human visual cortex by high-speed magnetic resonance imaging. Proceedings of the National Academy of Sciences 89, 11069-

11073.

Bloch, F., 1946. Nuclear induction. Physical review 70, 460.

Born, A., Miranda, M., Rostrup, E., Toft, P., Peitersen, B., Larsson, H., Lou, H., 2000.

Functional magnetic resonance imaging of the normal and abnormal visual system in early life.

Neuropediatrics 31, 24-32.

Born, A.P., Law, I., Lund, T.E., Rostrup, E., Hanson, L.G., Wildschiødtz, G., Lou, H.C.,

Paulson, O.B., 2002. Cortical deactivation induced by visual stimulation in human slow-wave sleep. Neuroimage 17, 1325-1335.

Born, P., Leth, H., Miranda, M.J., Rostrup, E., Stensgaard, A., Peitersen, B., Larsson, H.B., Lou,

H.C., 1998. Visual activation in infants and young children studied by functional magnetic resonance imaging. Pediatric research 44, 578.

Botting, N., Powls, A., Cooke, R.W., Marlow, N., 1998. Cognitive and educational outcome of very-low-birthweight children in early adolescence. Dev Med Child Neurol 40, 652-660.

Breeman, L.D., Jaekel, J., Baumann, N., Bartmann, P., Wolke, D., 2015. Preterm cognitive function into adulthood. Pediatrics 136, 415-423.

Broca, P., 1861. Remarques sur le siège de la faculté du langage articulé, suivies d’une observation d’aphémie (perte de la parole). Bulletin et Memoires de la Societe anatomique de

Paris 6, 330-357.

Brown, R.W., Haacke, E.M., Cheng, Y.-C.N., Thompson, M.R., Venkatesan, R., 2014. Magnetic resonance imaging: physical principles and sequence design. John Wiley & Sons.

Bullmore, E., Sporns, O., 2009. Complex brain networks: graph theoretical analysis of structural and functional systems. Nature Reviews Neuroscience 10, 186.

Bullmore, E., Sporns, O., 2009. Complex brain networks: graph theoretical analysis of structural and functional systems. Nature Reviews Neuroscience 10, 186-198.

Bullmore, E., Sporns, O., 2012. The economy of brain network organization. Nature Reviews

Neuroscience 13, 336.

Burnett, A.C., Scratch, S.E., Lee, K.J., Cheong, J., Searle, K., Hutchinson, E., De Luca, C.,

Davey, M.-A., Roberts, G., Doyle, L.W., 2015. Executive Function in Adolescents Born< 1000 g or< 28 Weeks: A Prospective Cohort Study. Pediatrics 135, e826-e834.

Byars, A.W., Holland, S.K., Strawsburg, R.H., Bommer, W., Dunn, R.S., Schmithorst, V.J.,

Plante, E., 2002. Practical aspects of conducting large-scale functional magnetic resonance imaging studies in children. Journal of child neurology 17, 885-889.

Calhoun, V.D., Liu, J., Adalı, T., 2009. A review of group ICA for fMRI data and ICA for joint inference of imaging, genetic, and ERP data. Neuroimage 45, S163-S172.

Cao, M., Wang, J.-H., Dai, Z.-J., Cao, X.-Y., Jiang, L.-L., Fan, F.-M., Song, X.-W., Xia, M.-R.,

Shu, N., Dong, Q., 2014. Topological organization of the human brain functional connectome across the lifespan. Developmental cognitive neuroscience 7, 76-93.

Chai, X.J., Castañón, A.N., Öngür, D., Whitfield-Gabrieli, S., 2012. Anticorrelations in resting state networks without global signal regression. Neuroimage 59, 1420-1428.

Chau, V., Taylor, M.J., Miller, S.P., 2013. Visual function in preterm infants: visualizing the brain to improve prognosis. Documenta Ophthalmologica 127, 41-55.

Cheour, M., Imada, T., Taulu, S., Ahonen, A., Salonen, J., Kuhl, P., 2004.

Magnetoencephalography is feasible for infant assessment of auditory discrimination.

Experimental Neurology 190, 44-51.

Cole, M.W., Pathak, S., Schneider, W., 2010. Identifying the brain's most globally connected regions. Neuroimage 49, 3132-3148.

Collin, G., Sporns, O., Mandl, R.C., van den Heuvel, M.P., 2013. Structural and functional aspects relating to cost and benefit of rich club organization in the human cerebral cortex.

Cerebral cortex 24, 2258-2267.

Craddock, R.C., James, G.A., Holtzheimer, P.E., Hu, X.P., Mayberg, H.S., 2012. A whole brain fMRI atlas generated via spatially constrained spectral clustering. Human brain mapping 33,

1914-1928.

Damadian, R., 1971. Tumor detection by nuclear magnetic resonance. science 171, 1151-1153.

Damoiseaux, J., Rombouts, S., Barkhof, F., Scheltens, P., Stam, C., Smith, S.M., Beckmann, C.,

2006. Consistent resting-state networks across healthy subjects. Proceedings of the National

Academy of Sciences 103, 13848-13853.

Dang-Vu, T.T., Schabus, M., Desseilles, M., Albouy, G., Boly, M., Darsaud, A., Gais, S.,

Rauchs, G., Sterpenich, V., Vandewalle, G., 2008. Spontaneous neural activity during human slow wave sleep. Proceedings of the National Academy of Sciences 105, 15160-15165.

de Kieviet, J.F., van Elburg, R.M., Lafeber, H.N., Oosterlaan, J., 2012. Attention problems of very preterm children compared with age-matched term controls at school-age. J Pediatr 161,

824-829.

Degnan, A.J., Wisnowski, J.L., Choi, S., Ceschin, R., Bhushan, C., Leahy, R.M., Corby, P.,

Schmithorst, V.J., Panigrahy, A., 2015. Alterations of resting state networks and structural connectivity in relation to the prefrontal and anterior cingulate cortices in late prematurity.

Neuroreport 26, 22-26.

Dehaene-Lambertz, G., Dehaene, S., Hertz-Pannier, L., 2002. Functional neuroimaging of speech perception in infants. science 298, 2013-2015.

Deshpande, A.K., Tan, L., Lu, L.J., Altaye, M., Holland, S.K., 2016. fMRI as a Preimplant

Objective Tool to Predict Postimplant Oral Language Outcomes in Children with Cochlear

Implants. Ear and hearing 37, e263-e272.

DiFrancesco, M.W., Robertson, S.A., Karunanayaka, P., Holland, S.K., 2013. BOLD fMRI in infants under sedation: comparing the impact of pentobarbital and propofol on auditory and language activation. Journal of Magnetic Resonance Imaging 38, 1184-1195.

Di Martino, A., Fair, D.A., Kelly, C., Satterthwaite, T.D., Castellanos, F.X., Thomason, M.E.,

Craddock, R.C., Luna, B., Leventhal, B.L., Zuo, X.-N., 2014. Unraveling the miswired connectome: a developmental perspective. Neuron 83, 1335-1353.

Doria, V., Beckmann, C.F., Arichi, T., Merchant, N., Groppo, M., Turkheimer, F.E., Counsell,

S.J., Murgasova, M., Aljabar, P., Nunes, R.G., 2010. Emergence of resting state networks in the preterm human brain. Proceedings of the National Academy of Sciences 107, 20015-20020.

Edgin, J.O., Inder, T.E., Anderson, P.J., Hood, K.M., Clark, C.A., Woodward, L.J., 2008.

Executive functioning in preschool children born very preterm: relationship with early white matter pathology. Journal of the International Neuropsychological Society 14, 90-101.

Erberich, S.G., Friedlich, P., Seri, I., Nelson Jr, M.D., Blüml, S., 2003. Functional MRI in neonates using neonatal head coil and MR compatible incubator. Neuroimage 20, 683-692.

Fair, D.A., Cohen, A.L., Power, J.D., Dosenbach, N.U., Church, J.A., Miezin, F.M., Schlaggar,

B.L., Petersen, S.E., 2009. Functional brain networks develop from a “local to distributed” organization. PLoS computational biology 5, e1000381.

Fan, J., McCandliss, B.D., Fossella, J., Flombaum, J.I., Posner, M.I., 2005. The activation of attentional networks. Neuroimage 26, 471-479.

Fischi-Gómez, E., Vasung, L., Meskaldji, D.-E., Lazeyras, F., Borradori-Tolsa, C., Hagmann, P.,

Barisnikov, K., Thiran, J.-P., Hüppi, P.S., 2015. Structural brain connectivity in school-age preterm infants provides evidence for impaired networks relevant for higher order cognitive skills and social cognition. Cerebral cortex 25, 2793-2805.

Fong, W., 2005. Handbook of MRI pulse sequences. Medical Physics 32, 1452-1452.

Foster-Cohen, S.H., Friesen, M.D., Champion, P.R., Woodward, L.J., 2010. High prevalence/low severity language delay in preschool children born very preterm. Journal of Developmental &

Behavioral Pediatrics 31, 658-667.

Fox, M.D., Snyder, A.Z., Vincent, J.L., Corbetta, M., Van Essen, D.C., Raichle, M.E., 2005. The human brain is intrinsically organized into dynamic, anticorrelated functional networks.

Proceedings of the National Academy of Sciences of the United States of America 102, 9673-

9678.

Fransson, P., 2005. Spontaneous low-frequency BOLD signal fluctuations: An fMRI investigation of the resting-state default mode of brain function hypothesis. Human brain mapping 26, 15-29.

Fransson, P., Åden, U., Blennow, M., Lagercrantz, H., 2010. The functional architecture of the infant brain as revealed by resting-state fMRI. Cerebral cortex 21, 145-154.

Fransson, P., Skiöld, B., Horsch, S., Nordell, A., Blennow, M., Lagercrantz, H., Åden, U., 2007.

Resting-state networks in the infant brain. Proceedings of the National Academy of Sciences

104, 15531-15536.

Freeman, L.C., 1978. Centrality in social networks conceptual clarification. Social networks 1,

215-239.

Friederici, A.D., Gierhan, S.M., 2013. The language network. Current opinion in neurobiology

23, 250-254.

Friston, K., 2002. Functional integration and inference in the brain. Progress in neurobiology 68,

113-143.

Friston, K., Frith, C., Liddle, P., Frackowiak, R., 1993. Functional connectivity: the principal- component analysis of large (PET) data sets. Journal of Cerebral Blood Flow & Metabolism 13,

5-14.

Friston, K.J., Holmes, A.P., Poline, J., Grasby, P., Williams, S., Frackowiak, R.S., Turner, R.,

1995. Analysis of fMRI time-series revisited. Neuroimage 2, 45-53.

Garrison, K.A., Scheinost, D., Finn, E.S., Shen, X., Constable, R.T., 2015. The (in) stability of functional brain network measures across thresholds. Neuroimage 118, 651-661.

100

Geerligs, L., Renken, R.J., Saliasi, E., Maurits, N.M., Lorist, M.M., 2014. A brain-wide study of age-related changes in functional connectivity. Cerebral Cortex 25, 1987-1999.

Gerlach, W., Stern, O., 1924. über die richtungsquantelung im magnetfeld. Annalen der Physik

379, 673-699.

Goldberg, G., 1985. Supplementary motor area structure and function: review and hypotheses.

Behavioral and brain Sciences 8, 567-588.

Gozdas, E., Parikh, N.A., Merhar, S.L., Tkach, J.A., He, L., Holland, S.K., 2018. Altered functional network connectivity in preterm infants: antecedents of cognitive and motor impairments? Brain Structure and Function, 1-16.

Gozzo, Y., Vohr, B., Lacadie, C., Hampson, M., Katz, K.H., Maller-Kesselman, J., Schneider,

K.C., Peterson, B.S., Rajeevan, N., Makuch, R.W., 2009. Alterations in neural connectivity in preterm children at school age. Neuroimage 48, 458-463.

Gray, J.R., Chabris, C.F., Braver, T.S., 2003. Neural mechanisms of general fluid intelligence.

Nature neuroscience 6, 316.

Hack, M., 2009. Adult outcomes of preterm children. Journal of Developmental & Behavioral

Pediatrics 30, 460-470.

101

He, L., Parikh, N.A., 2016. Brain functional network connectivity development in very preterm infants: the first six months. Early human development 98, 29-35.

Hebb, D.O., 1949. The organization of behavior: A neurophysiological approach. Wiley.[JH].

Holland, S., Choo, D., Ret, J., Hilbert, L., Dunn, R., Schmithorst, V., 2004. fMRI of severe to profoundly hearing-impaired infants and toddlers under sedation. International Congress Series.

Elsevier, pp. 383-386.

Holland, S., Dunn, R., Ball Jr, W., 1998. Performing fMRI in the pediatric age group: Technical and practical considerations. 1st International Pediatric Neuroradiology Symposium.

Philadelphia: American Society of Neuroradiology. p, pp. 15-19.

Holland, S.K., Plante, E., Byars, A.W., Strawsburg, R.H., Schmithorst, V.J., Ball Jr, W.S., 2001.

Normal fMRI brain activation patterns in children performing a verb generation task.

Neuroimage 14, 837-843.

Holland, S.K., Vannest, J., Mecoli, M., Jacola, L.M., Tillema, J.-M., Karunanayaka, P.R.,

Schmithorst, V.J., Yuan, W., Plante, E., Byars, A.W., 2007. Functional MRI of language lateralization during development in children. International journal of audiology 46, 533-551.

102

Horovitz, S.G., Fukunaga, M., de Zwart, J.A., van Gelderen, P., Fulton, S.C., Balkin, T.J., Duyn,

J.H., 2008. Low frequency BOLD fluctuations during resting wakefulness and light sleep: A simultaneous EEG-fMRI study. Human brain mapping 29, 671-682.

Hwang, K., Hallquist, M.N., Luna, B., 2012. The development of hub architecture in the human functional brain network. Cerebral Cortex 23, 2380-2393.

Karunanayaka, P.R., Holland, S.K., Schmithorst, V.J., Solodkin, A., Chen, E.E., Szaflarski, J.P.,

Plante, E., 2007. Age-related connectivity changes in fMRI data from children listening to stories. Neuroimage 34, 349-360.

Keunen, K., Counsell, S.J., Benders, M.J., 2017. The emergence of functional architecture during early brain development. Neuroimage 160, 2-14.

Kerr-Wilson, C., Mackay, D., Smith, G., Pell, J., 2012. Meta-analysis of the association between preterm delivery and intelligence. Journal of public health 34, 209-216.

Kozberg, M.G., Chen, B.R., DeLeo, S.E., Bouchard, M.B., Hillman, E.M., 2013. Resolving the transition from negative to positive blood oxygen level-dependent responses in the developing brain. Proceedings of the National Academy of Sciences, 201212785.

Kringelbach, M.L., 2005. The human orbitofrontal cortex: linking reward to hedonic experience.

Nature Reviews Neuroscience 6, 691.

103

Kwon, S.H., Scheinost, D., Lacadie, C., Sze, G., Schneider, K.C., Dai, F., Constable, R.T., Ment,

L.R., 2015. Adaptive mechanisms of developing brain: Cerebral lateralization in the prematurely-born. Neuroimage 108, 144-150.

Lancaster, J.L., Woldorff, M.G., Parsons, L.M., Liotti, M., Freitas, C.S., Rainey, L., Kochunov,

P.V., Nickerson, D., Mikiten, S.A., Fox, P.T., 2000. Automated Talairach atlas labels for functional brain mapping. Human brain mapping 10, 120-131.

Larson-Prior, L.J., Zempel, J.M., Nolan, T.S., Prior, F.W., Snyder, A.Z., Raichle, M.E., 2009.

Cortical network functional connectivity in the descent to sleep. Proceedings of the National

Academy of Sciences 106, 4489-4494.

Latora, V., Marchiori, M., 2001. Efficient behavior of small-world networks. Physical review letters 87, 198701.

Laumann, T.O., Snyder, A.Z., Mitra, A., Gordon, E.M., Gratton, C., Adeyemo, B., Gilmore,

A.W., Nelson, S.M., Berg, J.J., Greene, D.J., 2016. On the stability of BOLD fMRI correlations.

Cerebral cortex 27, 4719-4732.

Lauterbur, P.C., 1973. Image formation by induced local interactions: examples employing nuclear magnetic resonance.

104

Lawrence, E.J., McGuire, P.K., Allin, M., Walshe, M., Giampietro, V., Murray, R.M., Rifkin, L.,

Nosarti, C., 2010. The very preterm brain in young adulthood: the neural correlates of verbal paired associate learning. J Pediatr 156, 889-895.

Lee, J.D., Park, H.-J., Park, E.S., Oh, M.-K., Park, B., Rha, D.-W., Cho, S.-R., Kim, E.Y., Park,

J.Y., Kim, C.H., 2011. Motor pathway injury in patients with periventricular leucomalacia and spastic diplegia. Brain 134, 1199-1210.

Lee, W., Donner, E.J., NOSSIN-MANOR, R., Whyte, H.E., Sled, J.G., Taylor, M.J., 2012.

Visual functional magnetic resonance imaging of preterm infants. Developmental Medicine &

Child Neurology 54, 724-729.

Lehman, S.S., 2012. Cortical visual impairment in children: identification, evaluation and diagnosis. Current opinion in ophthalmology 23, 384-387.

Liang, X., Zou, Q., He, Y., Yang, Y., 2013. Coupling of functional connectivity and regional cerebral blood flow reveals a physiological basis for network hubs of the human brain.

Proceedings of the National Academy of Sciences 110, 1929-1934.

Liu, Y., Balériaux, D., Kavec, M., Metens, T., Absil, J., Denolin, V., Pardou, A., Avni, F., Van

Bogaert, P., Aeby, A., 2010. Structural asymmetries in motor and language networks in a population of healthy preterm neonates at term equivalent age: a diffusion tensor imaging and probabilistic tractography study. Neuroimage 51, 783-788.

105

Lo, C.-Y., Wang, P.-N., Chou, K.-H., Wang, J., He, Y., Lin, C.-P., 2010. Diffusion tensor tractography reveals abnormal topological organization in structural cortical networks in

Alzheimer's disease. Journal of Neuroscience 30, 16876-16885.

Lubsen, J., Vohr, B., Myers, E., Hampson, M., Lacadie, C., Schneider, K.C., Katz, K.H.,

Constable, R.T., Ment, L.R., 2011. Microstructural and functional connectivity in the developing preterm brain. Seminars in perinatology. Elsevier, pp. 34-43.

Malviya, S., Voepel-Lewis, T., Tait, A.R., 2006. A comparison of observational and objective measures to differentiate depth of sedation in children from birth to 18 years of age. Anesthesia

& Analgesia 102, 389-394.

Malviya, S., Voepel-Lewis, T., Tait, A.R., Watcha, M.F., Sadhasivam, S., Friesen, R.H., 2007.

Effect of age and sedative agent on the accuracy of bispectral index in detecting depth of sedation in children. Pediatrics 120, e461-e470.

Mansfield, P., 1977. Multi-planar image formation using NMR spin echoes. Journal of Physics

C: Solid State Physics 10, L55.

Mansfield, P., Grannell, P.K., 1973. NMR'diffraction'in solids? Journal of Physics C: Solid State

Physics 6, L422.

106

Månsson, J., Stjernqvist, K., 2014. Children born extremely preterm show significant lower cognitive, language and motor function levels compared with children born at term, as measured by the Bayley-III at 2.5 years. Acta Paediatrica 103, 504-511.

Martin, E., Joeri, P., Loenneker, T., Ekatodramis, D., Vitacco, D., Hennig, J., Marcar, V.L.,

1999. Visual processing in infants and children studied using functional MRI. Pediatric research

46, 135.

Mateo, C., Knutsen, P.M., Tsai, P.S., Shih, A.Y., Kleinfeld, D., 2017. Entrainment of arteriole vasomotor fluctuations by neural activity is a basis of blood-oxygenation-level-dependent

“resting-state” connectivity. Neuron 96, 936-948. e933.

Mazaika, P.K., Hoeft, F., Glover, G.H., Reiss, A.L., 2009. Methods and software for fMRI analysis of clinical subjects. Neuroimage 47, S58.

Ment, L.R., Peterson, B.S., Vohr, B., Allan, W., Schneider, K.C., Lacadie, C., Katz, K.H.,

Maller-Kesselman, J., Pugh, K., Duncan, C.C., 2006. Cortical recruitment patterns in children born prematurely compared with control subjects during a passive listening functional magnetic resonance imaging task. J Pediatr 149, 490-498. e492.

Mercuri, E., Atkinson, J., Braddick, O., Anker, S., Cowan, F., Rutherford, M., Pennock, J.,

Dubowitz, L., 1997a. Visual function in full-term infants with hypoxic-ischaemic encephalopathy. Neuropediatrics 28, 155-161.

107

Mercuri, E., Atkinson, J., Braddick, O., Anker, S., Cowan, F., Rutherford, M., Pennock, J.,

Dubowitz, L., 1997b. Basal ganglia damage and impaired visual function in the newborn infant.

Archives of Disease in Childhood-Fetal and Neonatal Edition 77, F111-F114.

Merhar, S.L., Gozdas, E., Tkach, J.A., Harpster, K.L., Schwartz, T.L., Yuan, W., Kline-Fath,

B.M., Leach, J.L., Altaye, M., Holland, S.K., 2016. Functional and structural connectivity of the visual system in infants with perinatal brain injury. Pediatric research 80, 43.

Meunier, D., Achard, S., Morcom, A., Bullmore, E., 2009. Age-related changes in modular organization of human brain functional networks. Neuroimage 44, 715-723.

Moore, T., Hennessy, E.M., Myles, J., Johnson, S.J., Draper, E.S., Costeloe, K.L., Marlow, N.,

2012. Neurological and developmental outcome in extremely preterm children born in England in 1995 and 2006: the EPICure studies. Bmj 345, e7961.

Murphy, K., Birn, R.M., Handwerker, D.A., Jones, T.B., Bandettini, P.A., 2009. The impact of global signal regression on resting state correlations: are anti-correlated networks introduced?

Neuroimage 44, 893-905.

Myers, E.H., Hampson, M., Vohr, B., Lacadie, C., Frost, S.J., Pugh, K.R., Katz, K.H., Schneider,

K.C., Makuch, R.W., Constable, R.T., 2010. Functional connectivity to a right hemisphere language center in prematurely born adolescents. Neuroimage 51, 1445-1452.

108

Newman, M.E., 2002. Assortative mixing in networks. Physical review letters 89, 208701.

Newman, M.E., 2006. Modularity and community structure in networks. Proceedings of the national academy of sciences 103, 8577-8582.

Ogawa, S., Lee, T.-M., Kay, A.R., Tank, D.W., 1990. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proceedings of the National Academy of Sciences 87,

9868-9872.

Ogawa, S., Tank, D.W., Menon, R., Ellermann, J.M., Kim, S.G., Merkle, H., Ugurbil, K., 1992.

Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging. Proceedings of the National Academy of Sciences 89, 5951-5955.

Parikh, N.A., 2016. Advanced neuroimaging and its role in predicting neurodevelopmental outcomes in very preterm infants. Seminars in perinatology. Elsevier, pp. 530-541.

Patel, A.M., Cahill, L.D., Ret, J., Schmithorst, V., Choo, D., Holland, S., 2007. Functional magnetic resonance imaging of hearing-impaired children under sedation before cochlear implantation. Archives of Otolaryngology–Head & Neck Surgery 133, 677-683.

Penfield, W., Jasper, H., 1954. Epilepsy and the functional anatomy of the human brain.

109

Picard, N., Strick, P.L., 2001. Imaging the premotor areas. Current opinion in neurobiology 11,

663-672.

Pietrasanta, M., Restani, L., Caleo, M., 2012. The corpus callosum and the visual cortex: plasticity is a game for two. Neural plasticity 2012.

Plante, E., Schmithorst, V.J., Holland, S.K., Byars, A.W., 2006. Sex differences in the activation of language cortex during childhood. Neuropsychologia 44, 1210-1221.

Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L., Petersen, S.E., 2012. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion.

Neuroimage 59, 2142-2154.

Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L., Petersen, S.E., 2013. Steps toward optimizing motion artifact removal in functional connectivity MRI; a reply to Carp. Neuroimage

76.

Power, J.D., Cohen, A.L., Nelson, S.M., Wig, G.S., Barnes, K.A., Church, J.A., Vogel, A.C.,

Laumann, T.O., Miezin, F.M., Schlaggar, B.L., 2011. Functional network organization of the human brain. Neuron 72, 665-678.

Power, J.D., Schlaggar, B.L., Petersen, S.E., 2015. Recent progress and outstanding issues in motion correction in resting state fMRI. Neuroimage 105, 536-551.

110

Purcell, E.M., Torrey, H.C., Pound, R.V., 1946. Resonance absorption by nuclear magnetic moments in a solid. Physical review 69, 37.

Redcay, E., Kennedy, D.P., Courchesne, E., 2007. fMRI during natural sleep as a method to study brain function during early childhood. Neuroimage 38, 696-707.

Rubinov, M., Sporns, O., 2010. Complex network measures of brain connectivity: uses and interpretations. Neuroimage 52, 1059-1069.

Rogers, C.E., Sylvester, C.M., Mintz, C., Kenley, J.K., Shimony, J.S., Barch, D.M., Smyser,

C.D., 2017. Neonatal amygdala functional connectivity at rest in healthy and preterm infants and early internalizing symptoms. Journal of the American Academy of Child & Adolescent

Psychiatry 56, 157-166.

Rowlands, M.A., Scheinost, D., Lacadie, C., Vohr, B., Li, F., Schneider, K.C., Constable, R.T.,

Ment, L.R., 2016. Language at rest: A longitudinal study of intrinsic functional connectivity in preterm children. NeuroImage: Clinical 11, 149-157.

Sakurai, J., 1994. In Tuan, SF, Ed.; Modern Quantum Mechanics, Revised Edition. Reading:

Addison-Wesley Publishing Company, Inc.

111

Salvador, R., Suckling, J., Coleman, M.R., Pickard, J.D., Menon, D., Bullmore, E., 2005.

Neurophysiological architecture of functional magnetic resonance images of human brain.

Cerebral cortex 15, 1332-1342.

Sarfeld, A.S., Diekhoff, S., Wang, L.E., Liuzzi, G., Uludağ, K., Eickhoff, S.B., Fink, G.R.,

Grefkes, C., 2012. Convergence of human brain mapping tools: neuronavigated TMS parameters and fMRI activity in the hand motor area. Human brain mapping 33, 1107-1123.

Schafer, R.J., Lacadie, C., Vohr, B., Kesler, S.R., Katz, K.H., Schneider, K.C., Pugh, K.R.,

Makuch, R.W., Reiss, A.L., Constable, R.T., 2009. Alterations in functional connectivity for language in prematurely born adolescents. Brain 132, 661-670.

Scheinost, D., Kwon, S.H., Shen, X., Lacadie, C., Schneider, K.C., Dai, F., Ment, L.R.,

Constable, R.T., 2016. Preterm birth alters neonatal, functional rich club organization. Brain

Structure and Function 221, 3211-3222.

Schmithorst, V.J., Holland, S.K., 2004. Comparison of three methods for generating group statistical inferences from independent component analysis of functional magnetic resonance imaging data. Journal of Magnetic Resonance Imaging: An Official

Journal of the International Society for Magnetic Resonance in Medicine 19, 365-368.

Schmithorst, V.J., Holland, S.K., 2006. Functional MRI evidence for disparate developmental processes underlying intelligence in boys and girls. Neuroimage 31, 1366-1379.

112

Schmithorst, V.J., Holland, S.K., 2007. Sex differences in the development of neuroanatomical functional connectivity underlying intelligence found using Bayesian connectivity analysis.

Neuroimage 35, 406-419.

Schmithorst, V.J., Holland, S.K., Plante, E., 2006. Cognitive modules utilized for narrative comprehension in children: a functional magnetic resonance imaging study. Neuroimage 29,

254-266.

Schmithorst, V.J., Vannest, J., Lee, G., Hernandez-Garcia, L., Plante, E., Rajagopal, A., Holland,

S.K., Consortium, C.A., 2015. Evidence that neurovascular coupling underlying the BOLD effect increases with age during childhood. Human brain mapping 36, 1-15.

Shi, F., Fan, Y., Tang, S., Gilmore, J.H., Lin, W., Shen, D., 2010. Neonatal brain image segmentation in longitudinal MRI studies. Neuroimage 49, 391-400.

Shi, F., Yap, P.-T., Fan, Y., Gilmore, J.H., Lin, W., Shen, D., 2010. Construction of multi- region-multi-reference atlases for neonatal brain MRI segmentation. Neuroimage 51, 684-693.

Shi, F., Yap, P.-T., Wu, G., Jia, H., Gilmore, J.H., Lin, W., Shen, D., 2011. Infant brain atlases from neonates to 1-and 2-year-olds. PloS one 6, e18746.

113

Smith, D.V., Utevsky, A.V., Bland, A.R., Clement, N., Clithero, J.A., Harsch, A.E., Carter,

R.M., Huettel, S.A., 2014. Characterizing individual differences in functional connectivity using dual-regression and seed-based approaches. Neuroimage 95, 1-12.

Smith, S.M., Jenkinson, M., Woolrich, M.W., Beckmann, C.F., Behrens, T.E., Johansen-Berg,

H., Bannister, P.R., De Luca, M., Drobnjak, I., Flitney, D.E., 2004. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 23, S208-S219.

Smyser, C.D., Inder, T.E., Shimony, J.S., Hill, J.E., Degnan, A.J., Snyder, A.Z., Neil, J.J., 2010.

Longitudinal analysis of neural network development in preterm infants. Cerebral cortex 20,

2852-2862.

Smyser, C.D., Snyder, A.Z., Shimony, J.S., Mitra, A., Inder, T.E., Neil, J.J., 2014. Resting-state network complexity and magnitude are reduced in prematurely born infants. Cerebral cortex, bhu251.

Song, M., Zhou, Y., Li, J., Liu, Y., Tian, L., Yu, C., Jiang, T., 2008. Brain spontaneous functional connectivity and intelligence. Neuroimage 41, 1168-1176.

Sroka, M.C., Vannest, J., Maloney, T.C., Horowitz-Kraus, T., Byars, A.W., Holland, S.K.,

Consortium, C.A., 2015. Relationship between receptive vocabulary and the neural substrates for story processing in preschoolers. Brain imaging and behavior 9, 43-55.

114

Synnes, A., Anderson, P.J., Grunau, R.E., Dewey, D., Moddemann, D., Tin, W., Davis, P.G.,

Doyle, L.W., Foster, G., Khairy, M., 2015. Predicting severe motor impairment in preterm children at age 5 years. Archives of disease in childhood, archdischild-2014-307695.

Szaflarski, J.P., Altaye, M., Rajagopal, A., Eaton, K., Meng, X., Plante, E., Holland, S.K., 2012.

A 10-year longitudinal fMRI study of narrative comprehension in children and adolescents.

Neuroimage 63, 1188-1195.

Szaflarski, J.P., Holland, S.K., Schmithorst, V.J., Byars, A.W., 2006. fMRI study of language lateralization in children and adults. Human brain mapping 27, 202-212.

Szaflarski, J.P., Schmithorst, V.J., Altaye, M., Byars, A.W., Ret, J., Plante, E., Holland, S.K.,

2006. A longitudinal functional magnetic resonance imaging study of language development in children 5 to 11 years old. Annals of neurology 59, 796-807.

Tam, E.W., Widjaja, E., Blaser, S.I., MacGregor, D.L., Satodia, P., Moore, A.M., 2008.

Occipital lobe injury and cortical visual outcomes after neonatal hypoglycemia. Pediatrics 122,

507-512.

Tan, L., Chen, Y., Maloney, T.C., Caré, M.M., Holland, S.K., Lu, L.J., 2013. Combined analysis of sMRI and fMRI imaging data provides accurate disease markers for hearing impairment.

NeuroImage: Clinical 3, 416-428.

115

Thomason, M.E., Grove, L.E., Lozon Jr, T.A., Vila, A.M., Ye, Y., Nye, M.J., Manning, J.H.,

Pappas, A., Hernandez-Andrade, E., Yeo, L., 2015. Age-related increases in long-range connectivity in fetal functional neural connectivity networks in utero. Developmental cognitive neuroscience 11, 96-104.

Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N.,

Mazoyer, B., Joliot, M., 2002. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 15, 273-

289.

van den Heuvel, M.P., Kahn, R.S., Goñi, J., Sporns, O., 2012. High-cost, high-capacity backbone for global brain communication. Proceedings of the National Academy of Sciences 109, 11372-

11377.

Van Den Heuvel, M.P., Pol, H.E.H., 2010. Exploring the brain network: a review on resting-state fMRI functional connectivity. European neuropsychopharmacology 20, 519-534.

Van Den Heuvel, M.P., Sporns, O., 2011. Rich-club organization of the human connectome.

Journal of Neuroscience 31, 15775-15786.

van den Heuvel, M.P., Stam, C.J., Kahn, R.S., Pol, H.E.H., 2009. Efficiency of functional brain networks and intellectual performance. Journal of Neuroscience 29, 7619-7624.

116

Vannest, J., Rajagopal, A., Cicchino, N.D., Franks-Henry, J., Simpson, S.M., Lee, G., Altaye,

M., Sroka, C., Holland, S.K., Consortium, C.A., 2014. Factors determining success of awake and asleep magnetic resonance imaging scans in nonsedated children. Neuropediatrics 45, 370-377.

Vannest, J.J., Karunanayaka, P.R., Altaye, M., Schmithorst, V.J., Plante, E.M., Eaton, K.J.,

Rasmussen, J.M., Holland, S.K., 2009. Comparison of fMRI data from passive listening and active-response story processing tasks in children. Journal of Magnetic Resonance Imaging 29,

971-976.

van’t Hooft, J., van der Lee, J.H., Opmeer, B.C., Aarnoudse-Moens, C.S., Leenders, A.G., Mol,

B.W.J., de Haan, T.R., 2015. Predicting developmental outcomes in premature infants by term equivalent MRI: systematic review and meta-analysis. Systematic reviews 4, 71.

Vigneau, M., Beaucousin, V., Herve, P.-Y., Duffau, H., Crivello, F., Houde, O., Mazoyer, B.,

Tzourio-Mazoyer, N., 2006. Meta-analyzing left hemisphere language areas: phonology, semantics, and sentence processing. Neuroimage 30, 1414-1432.

Volpe, J.J., 2009. Brain injury in premature infants: a complex amalgam of destructive and developmental disturbances. The Lancet Neurology 8, 110-124.

Watts, D.J., Strogatz, S.H., 1998. Collective dynamics of ‘small-world’networks. nature 393,

440.

117

Wechsler, D., 1981. WAIS-R manual: Wechsler adult intelligence scale-revised. Psychological

Corporation.

Wernicke, C., 1874. Der aphasische Symptomencomplex: eine psychologische Studie auf anatomischer Basis. Cohn.

Whitfield-Gabrieli, S., Nieto-Castanon, A., 2012. Conn: a functional connectivity toolbox for correlated and anticorrelated brain networks. Brain connectivity 2, 125-141.

Wilke, M., Hauser, T.K., Krägeloh-Mann, I., Lidzba, K., 2014. Specific impairment of functional connectivity between language regions in former early preterms. Human brain mapping 35,

3372-3384.

Wilke, M., Holland, S.K., Altaye, M., Gaser, C., 2008. Template-O-Matic: a toolbox for creating customized pediatric templates. Neuroimage 41, 903-913.

Wilke, M., Holland, S.K., Ball, W.S., 2003. Language processing during natural sleep in a 6- year-old boy, as assessed with functional MR imaging. American journal of neuroradiology 24,

42-44.

Wilke, M., Sohn, J.-H., Byars, A.W., Holland, S.K., 2003. Bright spots: correlations of gray matter volume with IQ in a normal pediatric population. Neuroimage 20, 202-215.

118

Williams, K.T., 2007. EVT-2: Expressive vocabulary test. Pearson Assessments.

Woodward, L.J., Moor, S., Hood, K.M., Champion, P.R., Foster-Cohen, S., Inder, T.E., Austin,

N.C., 2009. Very preterm children show impairments across multiple neurodevelopmental domains by age 4 years. Archives of Disease in Childhood-Fetal and Neonatal Edition 94, 339-

344.

Wu, K., Taki, Y., Sato, K., Hashizume, H., Sassa, Y., Takeuchi, H., Thyreau, B., He, Y., Evans,

A.C., Li, X., 2013. Topological organization of functional brain networks in healthy children: differences in relation to age, sex, and intelligence. PloS one 8, e55347.

Yamada, H., Sadato, N., Konishi, Y., Muramoto, S., Kimura, K., Tanaka, M., Yonekura, Y.,

Ishii, Y., Itoh, H., 2000. A milestone for normal development of the infantile brain detected by functional MRI. Neurology 55, 218-223.

Yu, B., Guo, Q., Fan, G., Liu, N., 2011. Assessment of cortical visual impairment in infants with periventricular leukomalacia: a pilot event-related FMRI study. Korean journal of radiology 12,

463-472.

Zuo, X.-N., Ehmke, R., Mennes, M., Imperati, D., Castellanos, F.X., Sporns, O., Milham, M.P.,

2011. Network centrality in the human functional connectome. Cerebral cortex 22, 1862-1875.

119

120