An introduction to resting state fMRI (Oxford Neuroimaging

Primers)

Visit the link below to download the full version of this book:
https://cheaptodownload.com/product/an-introduction-to-resting-state-fmri-oxford
-neuroimaging-primers-1st-edition-full-pdf-docx-download/
Oxford Neuroimaging Primers

Series Editors: Mark Jenkinson and Michael Chappell

Introduction to Neuroimaging Analysis

Mark Jenkinson
Michael Chappell

Introduction to Resting State fMRI Functional Connectivity

Janine Bijsterbosch
Stephen Smith
Christian Beckmann

Introduction to Perfusion Quantification using

Arterial Spin Labelling
Michael Chappell
Thomas Okell
Bradley MacIntosh
iiiiii

oxford neuroimaging primers

Introduction to
Resting State fMRI
Functional Connectivity
JANINE BIJSTERBOSCH
STEPHEN SMITH
CHRISTIAN BECKMANN

1
 1
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Oxford University Press 2017
The moral rights of the authors have been asserted
First Edition published in 2017
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2017938214
ISBN 978–​0–​19–​880822–​0
Printed in Great Britain by
Ashford Colour Press Ltd, Gosport, Hampshire
Oxford University Press makes no representation, express or implied, that the
drug dosages in this book are correct. Readers must therefore always check
the product information and clinical procedures with the most up-​to-​date
published product information and data sheets provided by the manufacturers
and the most recent codes of conduct and safety regulations. The authors and
the publishers do not accept responsibility or legal liability for any errors in the
text or for the misuse or misapplication of material in this work. Except where
otherwise stated, drug dosages and recommendations are for the non-​pregnant
adult who is not breast-​feeding
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
 Preface to the Series

The Oxford Neuroimaging Primers are aimed to be readily accessible texts for new researchers
or advanced undergraduates from the biological, medical or physical sciences. They are in-
tended to provide a broad understanding of the ways in which neuroimaging data can be
analyzed and interpreted. All primers in this series have been written so that they can be read
as stand-​alone books, although they have also been edited so that they “work together,” and
readers can read multiple primers in the series to build up a bigger picture of neuroimaging and
be equipped to use multiple neuroimaging methods.
Understanding the principles of the analysis of neuroimaging data is crucial for all researchers
in this field, not only because data analysis is a necessary part of any neuroimaging study,
but also because it is required in order to understand how to plan, execute, and interpret
experiments. Although MR operators, radiologists, and technicians are often available to help
with data collection, running the scanner and choosing good sequences and settings, when
it comes to analysis, researchers are often on their own. Therefore the Oxford Neuroimaging
Primers seek to provide the necessary understanding in how to do analysis at the same time
trying to show how this knowledge relates to being able to perform good acquisitions, design
good experiments and correctly interpret the results.
The series has been produced by individuals (both authors and editors) who have developed
neuroimaging analysis techniques, used these methods on real data, packaged them as software
tools for others to use, taught courses on these methods and supported people around the world
who use the software they have produced. We hope that this means everyone involved has both
the experience to instruct, but also the empathy to support the reader. It has been our aim for
these primers to not only lay out the core principles that apply in any given area of neuroimaging,
but also to help the reader avoid common pitfalls and mistakes (many of which the authors
themselves probably made first). We also hope that the series is also a good introduction to those
with a more technical background, even if they have to forgo some of the mathematical details
found in other more technical works. We make no pretence that these primers are the final word
in any given area, and the field of neuroimaging continues to develop and improve, but the
fundamentals are likely to remain the same for many years to come. Certainly some of the advice
you will find in the primers will never fail you, such as always look at your data.
Our intention with the series was always to support it with practical examples, so that the
reader can learn from working with data directly and will be equipped to use the knowledge
they gained in their own studies and on their own data. These examples, including datasets
and instructions, can be found on the associated website (www.neuroimagingprimers.org) and
directions to specific examples are placed throughout each primer. As the authors are also the
developers of various software tools within the FMRIB Software Library (FSL), the examples in
the primers mainly use tools from FSL. However, we intend these primers to be as general as
vi Preface to the Series

possible and present material that is relevant for all readers, regardless of the software they use
in practice. Such readers can still use the example data available through the primer website
with any of the major neuroimaging analysis toolboxes. We encourage all readers to interact
with these examples, as we strongly believe that a lot of the key learning is done when you ac-
tually use these tools in practice.
Mark Jenkinson
Michael Chappell
Oxford, January 2017
 Preface

Welcome to this primer on an Introduction to Resting State fMRI Functional Connectivity, which
is part of a series of Oxford NeuroImaging Primers. This primer is aimed at people who are new
to the field of resting state fMRI and functional connectivity. For many people who fit into this
category, the large variety of approaches to functional connectivity analysis is highly confusing.
This primer aims to provide an overview of the concepts and analysis decisions that need to be
made at every step of the analysis pipeline for resting state fMRI data, including issues of data
acquisition and the interpretation of findings. Our intention was to introduce a wide range of
analysis approaches without relying on any particular experience or prerequisite knowledge,
making this primer accessible to people from a broad range of different backgrounds. This
primer does assume some general background knowledge on functional magnetic resonance
imaging (fMRI), and if you are entirely new to fMRI you may benefit from reading the introduc-
tory primer for this series first.
This primer contains several different types of boxes in the text that are aimed to help you
navigate the material or find out more information for yourself. To get the most out of this
primer, you might find the descriptions of each type of box below helpful.

Example box These boxes direct you to the Oxford

NeuroImaging Primers website: www.neuroimagingprimers.org,
where you will find examples that allow you to directly interact
with data and perform practical exercises to apply the theory
described in this primer. These examples are intended to be a
useful way to prepare you for applying these methods to your
own data, but you do not need to carry out these examples as
you read through the primer. The examples are placed at the
relevant places in the text so that you know when you can get a
more hands-​on approach to the information being presented.

Box These boxes contain more technical or advanced

descriptions of some topics covered in this primer or informa-
tion on related topics or methods. None of the material in the
rest of the primer assumes that you have read these boxes, and
they are not essential for understanding and applying any of the
methods. If you are new to the field and are reading this primer
for the first time, you may therefore prefer to skip the material
in these boxes and come back to them later.
viii Preface

General statistics box This primer includes a couple of longer

boxes that describe some general statistics material that is needed
to prepare you for running your own analyses (i.e., these boxes
should not be skipped). These boxes contain material that is rele-
vant to many different topics covered in different chapters in this
primer, and it is therefore written in relatively general terms.

Summary Each chapter contains a box of summary points

towards the end, which provide a very brief overview, emphasizing
the most important topics discussed in each chapter. You may
like to use these to check that you have understood the key
points in each chapter.

Further reading Each chapter ends with a list of suggestions for

further reading, including both articles and books. A brief sum-
mary of the contents of each suggestion is included, so that you
can choose which references are most relevant to you. None of
the material in this primer assumes that you have read any of
the further reading material. Rather, these lists suggest a starting
point for diving deeper into the existing literature. These boxes
are not intended to provide a full review of the relevant litera-
ture; should you go further into the field you will find a wealth
of other sources that are helpful or important for the specific
research you are doing.

Resting state functional connectivity is a research field that is still at a relatively early stage, that
is evolving rapidly, and will most likely continue to do so over the coming years. Therefore, the
aim of this primer is to provide a comprehensive and introductory overview of the most com-
monly used approaches. We hope that this will provide good preparation for delving into the
vast literature on resting state fMRI.
Janine Bijsterbosch
Stephen Smith
Christian Beckmann
 Acknowledgments

Many people have contributed to the writing of this primer. In particular, I am grateful to the
editors for giving me the opportunity to write this primer, and specifically to Mark Jenkinson
for his continued support and enthusiasm during the writing process. I am also grateful to my
coauthors, Christian and Steve, for their insightful comments and the interesting discussions
that developed while writing this primer. Furthermore, I would like to thank Zobair Arya, Jon
Campbell, Michael Chappell, Falk Eippert, Stefania Evangelisti, Olivia Faull, Ludovica Griffanti,
Sam Harrison, Saad Jbabdi, Mark Jenkinson, Fred Lenz, Paul McCarthy, Thomas Nichols,
Michael Sanders, Charlotte Stagg, and Anderson Winkler for proofreading and commenting on
earlier versions of this primer. I would also like to thank Paul McCarthy for his help with creating
the cover image. Finally, I would like to thank my husband, Maarten Vooijs, for proofreading,
and for his continual encouragement, support, and patience.
Janine Bijsterbosch
 Contents

1 Introduction 1
1.1 From neural activity to functional connectivity 1
Box 1.1: Neuronal activity and local field potential 3
1.2 What is a resting state network (RSN)? 4
1.3 What can be gained from investigating the resting brain? 6
1.4 Resting state fMRI signal properties 8
1.5 Mapping the human connectome 10
Summary 10
Further reading 11

2 Data Acquisition 13
2.1 Repetition time, voxel size, and coverage 13
2.2 Multiband EPI sequence 15
2.3 Multi-​echo EPI sequence 17
2.4 Distortion, shimming, and fieldmaps 18
2.5 Scan duration 20
2.6 Eyes open versus eyes closed 20
2.7 Motion and physiological confounds 21
Summary 22
Further reading 22

3 Data Preprocessing 25
3.1 Sources of structured noise 27
3.2 Conventional preprocessing steps 28
3.3 Low-​pass temporal filtering 34
3.4 Nuisance regression 34
3.5 Global signal regression 35
3.6 Physiological noise regression 36
3.7 Volume censoring 37
xii Contents

3.8 Independent component analysis 39

Example box: Single subject ICA 42
General statistics box: Multiple linear regression analysis (with the GLM) 44
Summary 49
Further reading 49

4 Voxel-​based Connectivity Analyses 51

4.1 Seed-​based correlation analysis 52
Example box: Seed-based correlation analysis 54
4.2 Independent component analysis 55
Box 4.1: The ICA model 55
Box 4.2: How does the ICA “unmixing” work? 57
Example box: Group-ICA networks from different datasets 60
4.3 Obtaining subject-​wise ICA estimates with dual regression 62
Example box: Visualizing dual regression group analysis results 66
4.4 Amplitude of low-frequency fluctuations 68
4.5 Regional homogeneity 69
4.6 Group-level analysis for voxel-​based methods 70
General statistics box: Multiple comparisons correction 72
Summary 80
Further reading 80

5 Node-​based Connectivity Analyses 81

5.1 What is a node? 82
5.2 Node definition 85
Example box: Examples of node parcellations 89
5.3 Timecourse extraction 89
5.4 Edge definition 90
5.5 Network modeling analysis 94
Example box: Calculating subject and group network matrices 95
5.6 Graph theory analysis 97
5.7 Dynamic causal modeling 100
5.8 Dynamic and non-​stationary methods 101
5.9 When to use voxel-​based versus node-​based approaches 105
Summary 106
Further reading 106
 Contents xiii

6 Interpretation 109
6.1 The impact of psychology 110
6.2 The effects of BOLD physiology 113
6.3 The effects of methodological choices 115
6.4 Complementary types of connectivity research 125
6.5 Conclusions 130
Summary 131
Further reading 131

Index133
 1
Introduction

T he human brain represents about 2% of our overall body weight, yet it is thought to
consume approximately 20% of the total amount of energy produced by our body, even
when it is not performing any particular cognitive task. This fact has intrigued researchers for
decades. What does the brain do when we are apparently not doing anything? How does
this intrinsic activity relate to cognition, personality, disease, and consciousness? Interest in
intrinsic activity (as opposed to extrinsic activity, which occurs in response to external stimuli)
has steadily increased over the years, and with advances in functional magnetic resonance im-
aging (fMRI), studying the brain’s activity level during rest using so called “resting state fMRI,” has
bloomed into a research field in its own right.
Aimed at researchers that are new to the field of resting state fMRI, this book introduces the
major concepts and analysis approaches. In the interest of brevity, this primer does not cover
all the technical details, or every possible approach. Rather, an overview is presented, and it
is hoped that this will provide good preparation for delving into the vast literature on resting
state fMRI. Please note that this book specifically does not explain functional MRI itself more
generally, and people who are new to fMRI in general may find it helpful to read the primer
Introduction to Neuroimaging Analysis that is also part of this series.

1.1 From neural activity to functional connectivity

On a general level, there are two overarching concepts in the field of neuroimaging that can
inform us about how the brain works. The first of these is localization, which aims to assign
functions to specific regions of the brain. Many researchers use carefully designed behavioral
tasks that subjects perform in the MRI scanner in order to localize functionally specialized
regions of the brain that activate in response to a specific aspect of behavior. Tasks typically
include multiple different conditions (including baseline periods), and task-​induced activation
is measured and localized by comparing the blood oxygen level dependent (BOLD) signal
2 Introduction

between different conditions. The second general concept is to investigate connectivity, or the
way in which brain regions communicate with one another and information is passed from
one brain area to the next. In order to investigate connectivity, we measure the similarity of
the BOLD signals from different brain regions, because if the signals are similar, this is likely to
mean that the regions are passing on information from one region to the other (i.e., there is
connectivity). In order to study connectivity, we often look at spontaneous fluctuations in the
signal, when there are no specific cognitive demands for the subject (so-​called resting state
scans). Using spontaneous fluctuations allows us to investigate similarity between regions when
it is not biased by any specific task. As such, resting state fMRI has emerged as a valuable way
to study brain connectivity. For the purpose of this book, “rest” or “resting state” are defined as
being awake, but not performing any specific task (unless otherwise stated).
The previous paragraph describes the ideas and concepts that people refer to when they
talk about localization and connectivity. It is useful to understand how these concepts relate to
physiological processes in the brain both at the neuronal level and at the macroscopic level
that we measure in fMRI (Figure 1.1). At the microscopic level, a neuron consists of a cell body
(soma) that receives input through dendrites and passes action potentials through axonal tracts
to other cells. These microscopic processes in turn result in a localized increase in blood flow
that far exceeds the oxygen demands of the neural activity, leading to a local increase in blood
oxygenation level (Figure 1.1). It is crucial to appreciate that BOLD fMRI measures this increase
in blood oxygenation, which is a secondary and indirect measure of neuronal activity. The
hemodynamic response to neural activity which is measured in fMRI (i.e., blood oxygenation
levels) is a relatively slow process that only reaches its peak approximately 5–​6 seconds after
the start of the neural activity. Using concurrent fMRI and electrophysiological recording,
previous research has shown a strong link between spontaneous fluctuations in resting state

(a)
Post-synaptic Neurovascular Blood flow Blood BOLD
activity coupling oxygenation

(b) Timing of the hemodynamic response function

BOLD activation

Baseline BOLD

Stimulus 5 10 15 20 25 30 35 40
onset Time in seconds

Figure 1.1: The BOLD signal is an indirect measure of neuronal activity that is mediated by a slow
increase in local oxygenated blood flow that takes several seconds to peak. (a) Several complex
biological processes such as neurovascular coupling take place, which together result in the localized
increases in blood oxygenation that are measured in BOLD fMRI. (b) The standard form of the
hemodynamic response function is shown. From stimulus onset, the BOLD signal takes approximately
5 seconds to reach its maximum.
  From neural activity to functional connectivity 3

BOLD data and slow fluctuations in the local field potential (LFP). Therefore, the BOLD signal
is thought to primarily reflect the excitatory inputs to the neural population (synchronized
post-​synaptic activity). The relationship between resting state functional connectivity and local
neurophysiology is discussed in more detail in Chapter 6.

Box 1.1: Neuronal activity and local field potential

Inputs are transferred from one neuron’s axon terminal to the dendrites of another neuron
via the release of chemicals called neurotransmitters (such as glutamate) into the synaptic
cleft that exists between axon terminals and dendrites. Activity in a population of neurons
can therefore be measured by either looking at the firing rate (the overall pattern of spikes or
action potentials), or by measuring the LFP, which reflects a summation of the synchronized
post-​synaptic activity. The relationship between the firing rate and the LFP is complex
because an increase in post-​synaptic activity may affect different groups of neurons in
different ways and does therefore not simply result in an increase in population firing rate.
The physiological basis of BOLD is discussed in more detail in Chapter 6.

Functional connectivity is typically defined as: “the observed temporal correlation (or other
statistical dependencies) between two electro-​or neurophysiological measurements from
different parts of the brain.” For resting state fMRI this definition means that functional connec-
tivity can inform us about the relationship between BOLD signals obtained from two separate
regions of the brain. The underlying assumption is that if two regions show similarities in their
BOLD signals over time, they are functionally connected.
Many different methods exist to look at such similarities, and several are covered in more
detail in this book. The simplest way to investigate similarity between two signals is by looking
at their timeseries correlation using Pearson’s correlation coefficient. Correlation ranges from
–​1 (perfect negative correlation) to +1 (perfect positive correlation), where 0 indicates no
relationship on average between two signals. In 1995, Biswal and colleagues compared task
activation maps during finger tapping with a map of correlation coefficients of BOLD data
obtained during a scan when the subject was resting. The resting state correlation map was
created by taking all voxels that were activated by the motor task and using only the resting
state data to calculate the correlation of each voxel in the brain with those “activated” voxels.
The task activation map and resulting resting state correlation map showed strong spatial
similarities. This work is now often cited as the first study to show that intrinsic fluctuations
measured in the brain at rest by functional MRI hold information about the inherent functional
organization of the human brain. The spatial structure of functionally connected regions, which
is consistently and reliably found in resting state fMRI data, forms the foundation for resting
state fMRI research. Therefore, while functional connectivity is defined in terms of temporal
similarity between signals, the spatial patterns that emerge when looking at connectivity are
often of primary interest in functional connectivity research.
It is important to distinguish functional connectivity from other types of connectivity. First,
while functional connectivity describes the relationship between two regions, it is not typically
used to describe directionality (also called causality, which refers to a situation where the signal
4 Introduction

from one region is responsible for driving the signal in a second region). This type of directed
connectivity is described by effective connectivity. Directionality and causality are challenging
topics to study in BOLD fMRI data, and this is discussed further in Chapter 6. Secondly, it is
tempting to interpret functional connectivity as a direct physical connection (e.g., an axonal
white matter tract) between two brain regions. However, this type of anatomical connectivity
cannot be inferred from functional connectivity results alone. These different types of connec-
tivity will be discussed in more detail in Chapter 6, but the majority of this book will focus on
functional connectivity estimated from BOLD fMRI recordings.
Many different methodological approaches to studying functional connectivity are available
and several of these will be discussed in more detail in Chapters 4 and 5. While some of these
analysis approaches may appear very different from one another at first glance, it is important
to realize that all methods discussed in this work are based on the definition of functional con-
nectivity described above. Therefore, the field of resting state functional connectivity revolves
around detecting similarities among distinct regions of the brain. Specifically, in the case of
functional connectivity MRI, we detect similarities between the BOLD signal measured in
different parts of the brain.

1.2 What is a resting state network (RSN)?

A group or system of interconnected people or things is often called a network. Think, for
example, about your social media network or a computer network. Given the definition of
functional connectivity described above, a resting state network is simply a set of brain regions
that show similarities in their BOLD timeseries obtained during rest. At present, we do not have
a complete understanding of the network structure of the resting brain. Nonetheless, several
networks can be reproducibly found using a variety of analysis approaches.
Different resting state networks have been identified and “named” mostly on the basis of
the spatial similarity between the resting state networks and activation patterns seen in task
fMRI experiments. This naming convention is most accurate for areas associated with sensory
processing, where it has been established that a correspondence exists between areas that can
be mapped in response to sensory stimulation and areas that have strong resting state BOLD
similarities. Other parts of the brain, for example within multimodal association cortex, are
more ambiguously related to task experiments.
Perhaps the best-​known resting state network of all is the Default Mode Network (DMN;
Figure 1.2). The DMN contains regions in the brain that consistently show decreases in activity
when the brain is performing any type of task compared with rest (deactivations), as shown by
early task-​based imaging studies using both fMRI and positron emission tomography (PET). Key
regions of the DMN are the posterior cingulate cortex, precuneus, medial prefrontal cortex,
inferior parietal lobule, and lateral temporal cortex.
The dorsal attention network (DAN; also called the task-​positive network; Figure 1.2) is
another commonly described network made up of regions that are commonly activated
during various types of goal-​directed behavior. Regions that are included in the DAN are the
inferior parietal cortex, frontal eye fields, supplementary motor area, insula and dorsolateral
prefrontal cortices. Some findings suggest that the DMN and the DAN may be anticorrelated,
 What is a resting state network (RSN)? 5

Default Mode Network

Dorsal Attention Network

Figure 1.2: The default mode and dorsal attention networks shown on a surface view of the brain (left)
and on a volumetric view (right). In order to create the surface view, the cortical gray matter ribbon is
inflated so that all of the cortical surface (including sulci) can be viewed at once. The top two surface views
in each panel show the brain viewed from the left and right side, and the bottom two surface views show a
medial view of the brain. For the volumetric images on the right, the most representative sagittal, coronal,
and horizontal planes are shown. This figure is showing networks identified using independent component
analysis performed on data from the Human Connectome Project (discussed further in Section 1.5).
6 Introduction

although these results may, in part, be driven by preprocessing choices. This is discussed in
more detail in Chapter 3.
Other commonly described networks include multiple distinguishable visual networks
(including dorsal and ventral visual networks), auditory networks, and sensorimotor networks.
In addition to the DMN and DAN, additional cognitive networks include salience, executive
control, and fronto-​parietal networks.
It is important to note that this nomenclature describes a categorization of the brain at a
single and somewhat arbitrarily chosen level of granularity. This is to say that these resting state
networks form a hierarchy, where networks can be broken down further into yet finer-​grained
systems (i.e., form “networks within networks”). As such, it is not the case that every area in
the brain can be uniquely assigned to one of a set of resting state networks. Indeed, brain
regions that are known to have extensive connectivity with many other brain regions, also show
functional connectivity with multiple resting state networks.

1.3 What can be gained from investigating

the resting brain?
There are multiple reasons for choosing to investigate the brain using resting state fMRI. Some
of these reasons relate to the types of knowledge that can be gained from studying the brain at
rest, whereas others are more pragmatic in nature.
First, resting state fMRI can be used to inform us about the inherent organization and
functioning of the brain. Gaining a better understanding of the brain’s intrinsic architecture and
the level of communication this supports is an important basic neuroscience aim in its own
right, and may help us understand how the brain enables complex information processing and
rich sets of behaviors, thoughts, and motivations. Likewise, understanding communication in
the brain may also help us understand how things go wrong in a variety of different disorders.
For example, disorders like attention deficit hyperactivity disorder are thought to be associated
with aberrant communication between different regions of the brain.
Additionally, gaining a better understanding of the brain in its basal resting state may be
helpful in order to better understand how the brain activates in response to task demands.
Currently, task fMRI activation studies commonly rely on the notion that any cognitive process
is simply added to anything else that the brain is doing, without affecting any of the other
processes (based on the principle of pure insertion). This assumption is probably violated in
many situations, due to complex interaction between cognitive processes and demands. For
example, processes like attention and working memory are involved in even the simplest of
tasks. Studying the brain at rest can help to gain a perspective on the variability of spontaneous
fluctuations, and how these may be influenced by recent experiences, and by the current cog-
nitive and emotional state of the subject.
In addition to the improved understanding of the basic neuroscience of the brain, resting
state fMRI also has great potential to serve as a biomarker for mental disorders. A biomarker is
something that can be measured accurately and reproducibly, and can therefore serve as an
objective indication of the medical state of a person. Biomarkers can be used: (i) as indicators
 What can be gained from investigating the resting brain? 7

of normal biological processes, (ii) for early detection of for mental disorders, (iii) as markers of
disease progression, (iv) as markers of response to treatment, and (v) for creating personalized
treatment strategies that are optimized for the patient. No reliable and objective biomarkers are
currently available for many mental disorders, but it is hoped that functional connectivity may
be used as a biomarker in several of these (such as, for example, depression). Resting state is a
particularly promising target for biomarker research because it has many pragmatic benefits (see
below), which greatly improve the likelihood of resting state fMRI being used in clinical settings.
Additionally, these advantages also mean that it is feasible to obtain enough data to develop
normative look-​up graphs needed for personalized medicine (similar to the graphs that indicate
the distribution of weight for a healthy baby of age 12 months). These advantages are, therefore,
very important for what can potentially be gained from studying the resting brain.
The first pragmatic benefit of resting state fMRI is that it does not require a lot of extra
equipment beyond the MRI scanner, because there is no need to present the subjects with
any information or record any behavioral responses (such as button presses) during the scan.
Additionally, the need for expertise of the person acquiring the scan is also reduced, given
that no stimulus presentation programming is required and instructions for the participant
are minimal. As a result of reduced dependence on equipment and expertise, resting state
fMRI is relatively easy to acquire. It also means that resting state fMRI is attractive in the light
of data sharing and big data efforts. Additionally, the relative ease of resting state acquisition
encourages researchers to replicate their own and other people’s findings.
A further pragmatic advantage of investigating the resting brain is the absence of cogni-
tive demands for the subject. Studying subjects at rest is feasible for a large set of subject
populations. For example, many clinical populations are unable to perform tasks for a variety of
reasons, but may be studied at rest. As a consequence, comparing data across the full lifespan
from infancy (or even during prenatal development) to old age is made possible with resting
state fMRI. Therefore, the potential of resting state fMRI as a biomarker for early detection and/​
or personalized treatment of disorders is vast. Importantly, the relative ease of acquisition and
potentially short scan duration means that routine acquisition of resting state fMRI would be
feasible in a clinical setting (scan duration is discussed more in Chapter 2). While resting state
fMRI is not currently used in clinical practice, this biomarker potential has been a primary driver
of the exponential increase of resting state research over the last decade.
While there is much to be gained from studying the resting brain, one common criticism should
also be discussed briefly here. Everyone knows intuitively that the brain is not doing “nothing”
when we are awake but not engaged in any specific task, and more importantly, we know that the
brain can be engaged in a large variety of different processes when “at rest.” When the subject is
lying in the scanner listening to the scanner sounds during a resting state scan he/​she might, for
example, be thinking about what to cook for dinner later that night, might be internally rehearsing
a skill such as playing the piano, or could be worrying or daydreaming. Studying the brain while it is
in this uncontrolled state, that may involve a large variety of processes, is seen as highly problematic
by some, because it can be difficult to see how we could learn something about the brain when
we are unsure what it is doing in that period. Nevertheless, functional connectivity networks during
rest have been shown to be remarkably consistent and can be reliably found across subjects and
across studies. These consistent and reliable networks convincingly address one aspect of the
criticism outlined above, namely that there appears to be some information represented in the
resting brain that can be observed and studied regardless of the subject’s cognitive state during
8 Introduction

rest. However, an important issue that arises from the fact that “rest” is an uncharacterized state
(together with inherent limitations of the BOLD signal) is the interpretation of functional connec-
tivity. For example, when we see a change in a resting state network between patients and healthy
controls, how can and should we be interpreting this change? Interpretation is a major challenge
in resting state fMRI and is discussed in more detail in Chapter 6.

1.4 Resting state fMRI signal properties

Functional MRI enables us to record in vivo data from the whole brain at a relatively good
spatial and temporal resolution. However, the BOLD signal obtained in fMRI is an indirect,
metabolic measure of neural activity mediated by the hemodynamic response function (HRF;
for an introduction to fMRI please see the primer Introduction to Neuroimaging Analysis in
this series). As a result, it is necessary to interpret the BOLD signal as a relative, rather than
quantitative, measure of activity. Additionally, the BOLD signal is extremely noisy because it is
affected by many things other than neural activity, including breathing and cardiac pulsations,
and MRI-​related artifacts. Therefore, when we show our subject a stimulus, the resulting local
increase in neural activity typically only changes the BOLD signal by 1–​3% of the baseline BOLD
signal. That means that by simply looking at the BOLD images, it is often not possible to “see the
activation,” that is, it is not possible to distinguish it by eye from the noise.
Many of the challenges faced when analyzing resting state fMRI data result from the large
amount of noise and from the indirect nature of the BOLD signal. In resting state fMRI we
try to reduce the noise as much as possible with preprocessing, before actually analyzing the
data. The reason for this is that functional connectivity methods aim to find similarities in the
BOLD signal between two different brain regions, and many types of noise may induce such
similarities. Think, for example, about a deep breath or a jolt of movement, these types of artifacts
would affect multiple areas of the brain in the same way and could therefore be detected as
“functional connectivity.” As such, making good preprocessing decisions is extremely important
in resting state studies, and many controversies and debates relate to the magnitude and nature
of the impact of these preprocessing steps on the results. Commonly adopted preprocessing
methods are described and discussed in detail in Chapter 3.
It may be helpful to know that functional connectivity is also commonly described in the
literature as correlated low-​frequency fluctuations (or oscillations) in the BOLD signal. The
reason for describing fluctuations as “low frequency” is that the power spectrum of a BOLD
timeseries has most power in the low-​frequency range (Figure 1.3). Calculating the power
spectrum using the Fourier transform describes how the variance of the signal is distributed
across frequencies; that is, it reflects how much of the changes in signal occur slowly versus
more rapidly. The dominance of the low frequencies is not specific to resting state fMRI data; it
is an inherent property of the BOLD signal. The reason for this is that the BOLD signal measures
blood oxygenation in the active area through the HRF, which takes several seconds to change.
Any neural activity is therefore only measured through the veil of the slow HRF, meaning that
most of the signal varies relatively slowly over time (which is sometimes referred to by saying
that “the HRF acts as a low pass filter”).
 RESTING STATE fMRI SIGNAL PROPERTIES 9

0.9

0.8

0.7
Normalized power
0.6

0.5

0.4

0.3

0.2

0.1

0
0.01 0.1 0.2 0.3 0.4 0.5 0.6
Frequency (Hz)

Figure 1.3: Fourier transforms can be used to measure the strength of the signal across different
frequencies, and the results are visualized as frequency spectra as shown above. The Fourier transforms
of 25 resting state networks are shown, and reveal that the resting state BOLD signal is dominated by
low-​frequency power (i.e., the power is high in the lower frequencies). Two of the networks contain a large
amount of noise (red and blue lines), which shows up in the frequency spectrum as relatively high power
in higher frequencies (i.e., the spectra have a “raised tail”).

Given that much of the power in the BOLD signal occurs at lower frequencies, early functional
connectivity studies applied bandpass filtering to resting state data to remove fluctuations over
certain frequencies (such as >0.1 Hz) in order to reduce noise (this is discussed in more detail in
Chapter 3). However, studies have since shown that information about resting state networks is
also present at higher frequencies and, as a result, the term “low-​frequency fluctuations” is not
seen as often in the recent literature.
Despite the fact that the majority of the power is in the low-​frequency range, there is
information in the higher frequencies, and it can therefore be beneficial to include the high-​
frequency data for functional connectivity analyses, instead of throwing it away with bandpass
filtering. There is no reason why group differences in functional connectivity (which are most
commonly of primary interest in resting state fMRI research) would be limited to the low-​
frequency range. At a neuronal level, differences between patients and healthy controls may
arise from some form of electrophysiological change (for example, ion channels may be altered
in patients). To our knowledge, this neuronal change may be reflected in the BOLD signal at
any frequency. Therefore, it may be preferable to consider the full range of frequencies when
studying resting state fMRI data.