an examination of the acute effects of exercise on the ... · split into two groups, the first...
TRANSCRIPT
An Examination of the Acute Effects of Exercise on the Brain as seen by Resting State Functional MRI
by
Ahmad Rajab
A thesis submitted in conformity with the requirements for the degree of Master of Science
Department of Medical Biophysics University of Toronto
© Copyright by Ahmad Rajab 2014
ii
Thesis examination of the acute effects of exercise on the brain
as seen by resting state functional MRI
Ahmad Rajab
Master of Science
Department of Medical Biophysics
University of Toronto
2014
Abstract
Resting state fMRI (rs-fMRI) is a non-invasive imaging technique used to probe differences in
brain activity across interventions or populations. While long term exercise is known to have
cognitive, structural and neuro-protective brain effects, considerably less is known about exercise
in the context of resting state functional connectivity (rs-fc) in the brain. In this thesis I examine
the effects of moderate intensity single session exercise on rs-fc in young healthy adults, using
rs-fMRI. Data are analyzed using independent component analysis, denoising and dual
regression procedures. Analysis reveals increased activation post-exercise in three sensorimotor
areas: the post-central gyrus, secondary somatosensory area and thalamus. This thesis establishes
the feasibility of single session exercise paradigms to probe rs-fc and illustrates the importance
of analysis considerations to improve detection of session effects. These research tools are
extended to a pilot chronic stroke cohort, to establish functional measures that may relate to
stroke recovery.
iii
Acknowledgments
I would like to acknowledge the invaluable help extended to me during the course of my research
by various individuals. Firstly my supervisor Dr. Bradley J MacIntosh for his unceasing support,
belief, motivation and inspiration, and my committee Drs. Simon Graham, Christopher
Macgowan and Laura Middleton for their guidance and supportive criticism. Secondly, current
and former labmates Ilia Makedonov, Ekaterina Tchistiakova, David Crane, Dr. Andrew
Robertson, Dr. Walter Swardfager, Anna Mersov, Zarah Shirzadi and Farhang Jalilian for their
continuous support, advice and friendship. Additionally, members of the Sunnybrook branch of
the Canadian Partnership for Stroke Recovery Karen Fan, Valerie Closson, Cynthia Danells, and
Farrell Leibovitch, administrative staff in the Medical Biophysics department Merle Casci,
Donna-Marie Pow, and lastly Dr. Lothar Lilge, for convincing me to apply to Medical
Biophysics.
I would also like to acknowledge the funding sources that made this research possible, the
Natural Sciences and Engineering Research Council of Canada, the Heart and Stroke Foundation
Canadian Partnership for Stroke Recovery, the MITACS Accelerate Program and the Ontario
Graduate Scholarship in Science and Technology.
iv
Table of Contents
Acknowledgments .......................................................................................................................... iii
Table of Contents ........................................................................................................................... iv
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Chapter 1 Introduction .................................................................................................................... 1
1.1 Functional Brain Imaging ................................................................................................... 1
1.1.1 The BOLD signal and BOLD imaging ................................................................... 2
1.1.2 Task fMRI ............................................................................................................... 5
1.1.3 Resting state fMRI .................................................................................................. 5
1.2 Exercise ............................................................................................................................... 7
1.3 Changing the brain with exercise ........................................................................................ 9
1.3.1 Exercise and brain reorganization ......................................................................... 10
1.3.2 Animal studies of exercise .................................................................................... 10
1.3.3 Human studies of exercise .................................................................................... 10
1.4 Objectives of this work ..................................................................................................... 11
1.5 Analysis Techniques ......................................................................................................... 13
1.5.1 Independent Component Analysis ........................................................................ 13
1.5.2 ICA and BOLD fMRI ........................................................................................... 17
Chapter 2 Methods and Results .................................................................................................... 21
2.1 Materials and Methods ...................................................................................................... 21
2.1.1 Participants and Study Design .............................................................................. 21
2.1.2 Single Session Exercise ........................................................................................ 22
2.1.3 Magnetic Resonance Imaging ............................................................................... 23
2.1.4 MRI Analysis ........................................................................................................ 23
v
2.2 Results ............................................................................................................................... 25
2.2.1 Single Session Exercise ........................................................................................ 25
2.2.2 Exercise dataset ..................................................................................................... 25
2.2.3 No-exercise dataset ............................................................................................... 30
2.3 Discussion ......................................................................................................................... 31
Chapter 3 Future Work ................................................................................................................. 35
3.1 Acute effects of exercise on Chronic Stroke Adults ......................................................... 35
3.1.1 Participants and Study Design .............................................................................. 36
3.1.2 Preliminary Analysis and Results ......................................................................... 38
3.1.3 Discussion and Future Work ................................................................................. 44
3.2 Conclusion ........................................................................................................................ 48
Reference List ............................................................................................................................... 50
vi
List of Tables
Table 2.1: Participant Demographics ............................................................................................ 21
Table 2.2: Details of VISTAR IC removal ................................................................................... 27
Table 2.3: Cohen's D effect size comparing pre- vs. post- (exercise) and session 1 vs. 2 (no-
exercise) using the VISTAR Pipeline ........................................................................................... 30
Table 3.1: NUES participant demographics ................................................................................. 38
Table 3.2: Details of VISTAR IC removal ................................................................................... 41
Table 3.3: Cohen's D effect size comparing pre- vs. post-exercise for each exercise intensity, and
for both intensities, combined ....................................................................................................... 43
vii
List of Figures
Figure 1.1: Functional brain mapping techniques ........................................................................... 2
Figure 1.2: An illustration of the hemodynamic response .............................................................. 4
Figure 1.3: Non-Gaussian sources can be mixed together to form Gaussian signals ................... 14
Figure 1.4: a) Single subject ICA is useful for providing subject specific maps, while b) Group
ICA produces more robust spatial maps since it utilizes more information ................................. 18
Figure 1.5: ICA is capable of producing spatially similar and functionally relevant ICs: (A)
Default mode network (B) Left Lateral Frontoparietal Network (C) Right Lateral Frontoparietal
Network (D) Sensorimotor Network (E) Medial Parietal Region (F) Visual Network (G) Frontal
Lobes (H) Cerebellum (adapted from (Brookes et al., 2011), reprinted with permission) ........... 19
Figure 2.1: Data collection timeline .............................................................................................. 22
Figure 2.2: Group ICA maps overlaid on an MNI template. The Raw pipeline produced eight
spatial RSN maps and the VISTAR pipeline produced nine, with the basal ganglia being split
into two components. Both pipelines produce spatially similar maps (spatial cross correlation
(scc)) (a) Sensorimotor [MNI: 0,-14,52]: scc=0.88; (b) Auditory [54,-16,2]: 0.88; (c) DMN [0,-
56,32]: 0.79; (d) ppCG [42,-14,24]: 0.85; (e) Medial Visual [2,-58,0]: 0.88; (f) Attention [-
34,18,42]: 0.80; (g) Executive [0,14,20]: 0.89; (h) Basal Ganglia [-16,-14,2]: 0.77 [i], 0.67 [ii]. 26
Figure 2.3: The number of ICs used for the RSN group analyses in the exercise dataset from the
Raw and VISTAR pipelines. No significant differences were noted in the number of ICs Pre- vs.
Post- Raw (p=0.274), or Pre- vs. Post- VISTAR (p=0.747). ........................................................ 28
Figure 2.4: Examples of single session ICA artefacts: a) cerebral spinal fluid pulsation b) eye
motion c) head motion d) high frequency noise (>0.1Hz) e) spurious signal ............................... 28
viii
Figure 2.5: Post-exercise changes in rs-fMRI. VISTAR RSNs (yellow) overlaid with cluster
enhanced-regions of increased activation (green): (a) Auditory [MNI: -42,-24,12]; (b)
Sensorimotor [0,-40,72]; (c) Thalamic-Caudate [10,-10,-2]. ........................................................ 29
Figure 2.6: Cohen’s D values for the 9 RSNs of interest using exercise (N=15) and no-exercise
(N=15) datasets (VISTAR). SM: Sensorimotor, Aud: Auditory, DMN: Default Mode Network,
ppCG: Pre and Post-Central Gyri, Med-Vis: Medial Visual, Atten: Attention, Exec: Executive,
Put: Putamen, Th/Cau: Thalamus and Caudate ............................................................................ 31
Figure 3.1: Neurovascular Underpinnings of Exercise after Stroke trial timeline for each
participant ..................................................................................................................................... 36
Figure 3.2: Data collection timeline for visits 3 and 4. MRI sequences in green, biometrics data
in blue ............................................................................................................................................ 37
Figure 3.3: T1-weighted or FLAIR images highlighting the location of stroke lesion for each of
the participants included in the preliminary analysis for this chapter. ......................................... 39
Figure 3.4: Group ICA maps generated using 40 datasets: using 10 subjects both pre- and post-
exercise and both 50 % and 70% sessions: (a) Sensorimotor [MNI: 0,-28,50]; (b) Auditory [MNI:
-40,-14,4]; (c) DMN [-2,-62,16]; (d) ppCG [44,-12,26]; (e) Medial Visual [6,-72,0]; (f) Attention
[44,18,0]; (g) Executive [-2,22,40]; (h) Basal Ganglia [-14,-6,4] ................................................ 40
Figure 3.5: RSN masks overlaid on an MNI brain template with raw pipeline results showing
voxels of increased co-activation post-exercise: (a) Auditory [MNI: 50,-32,2]; (b) DMN [2,-
66,50] (c) ppCG [-58,-18,36]; (d) Attention [-42,70,0] ................................................................ 42
Figure 3.6: Group ICA maps generated using 20 datasets: using 10 subjects pre-exercise and both
50% and 70% sessions: (a) Sensorimotor [MNI: 2, -14, 54]; (b) Auditory [MNI: -40, -18, 6]; (c)
DMN [-2, -64, 18]; (d) ppCG [50, -20, 36]; (e) Attention [-38, 18, 36]; (f) Basal Ganglia [32, 2, -
2]. .................................................................................................................................................. 43
Figure 3.7: Results of the regression analysis to identify brain regions that are associated with
aerobic fitness level (fVO2). Brain regions in green represent areas where there is a positive or
negative association with fVO2. Voxels were considered significant if they reached p<0.05 for
ix
both the pre50 and pre70 sessions (i.e. conjunction statistic equivalent to a raw p-value=0.0025).
(a) Sensorimotor [MNI: 16, 2, 64], [0, -6, 52]; (b) Auditory [MNI: -54, -24, 16], [-52, -28, 2]; (c)
DMN [-6, -74, 28], [10,-64, 36]; (d) ppCG [-56, 0, 26], [52 , -16, 38]; (e) Attention [MNI: -32, -
10, 8], [-46, 4, 8]; (f) Basal Ganglia [-22, 4, -8], [-22, 4, -8]. ....................................................... 45
1
Chapter 1 Introduction
1.1 Functional Brain Imaging
Functional brain imaging represents a “full range of techniques by which physiological changes
accompanying brain activity are defined” (Jezzard et al., 2001). Studying the functional brain
provides a window into the ‘inner workings’ of the healthy and diseased brain. While structural
imaging provides rich information about the brain’s anatomy, it cannot tell us about the dynamic
aspects of the brain that change over the course of our day or over the course of weeks, since
structural changes are thought to occur on the timescale of months. With functional imaging, we
can ask questions like: what areas of the brain are involved in carrying out a specific behaviour;
what are the functional brain differences between a group of healthy adults and a group with a
brain disease; and how does brain function change with age and can we predict disease-risk?
Functional brain imaging modalities have a characteristic list of strengths and weaknesses, for
instance spatial and temporal resolution considerations (Figure 1.1). These modalities can be
split into two groups, the first being electro-magnetic techniques. These techniques attempt to
directly map either transient brain electrical dipoles that are generated by neuronal
depolarization, as in the case of electroencephalography or EEG, or the associated magnetic
dipoles, as in the case of magnetoencephalography or MEG. These modalities excel at measuring
neuronal signals that occur in brief periods of time, with high temporal resolution in the range of
10-100ms, but suffer from a lack of good spatial resolution (many mm-cm).
The second category of functional brain imaging modalities are those that rely on hemodynamic
techniques. Stimulation of the brain causes local increase in blood flow, as demonstrated by
Charles Roy and Charles Sherrington in 1890. This change in blood flow is needed to
accommodate changes in neuronal or synaptic activity and the associated energy demands.
Positron emission tomography (PET) is the oldest imaging modality of this type, and uses
invasive radiotracers to map functional changes in blood flow. The invasive and radioactive
nature of PET however, makes it less than ideal for studying infant, preadolescent, elderly or
diseased populations. This is especially true for exploratory studies.
2
Optical imaging techniques (like near infrared spectroscopy (NIRS)) are another hemodynamic
modality. NIRS exploits the fact that the transmission and absorption of NIR light by tissue
varies with hemoglobin concentration changes. This technique, however, lacks penetration and is
limited to probing cortical surface areas. It also suffers from poor spatial resolution.
1.1.1 The BOLD signal and BOLD imaging
Another hemodynamic based modality is functional MRI (fMRI) using the blood oxygenation
level dependent (BOLD) contrast effect, a burgeoning technique in recent decades and the
central modality used in this thesis. BOLD contrast arises from changes in the local magnetic
susceptibility of hemoglobin, a protein found in red blood cells and used for oxygen transport.
Oxygenated hemoglobin (Hb-O2) is diamagnetic (Pauling and Coryell, 1936), that is, it reduces
magnetic flux and repels an applied magnetic field. De-oxygenated hemoglobin (dHb) however,
is paramagnetic; it increases magnetic flux and attracts an applied magnetic field.
In BOLD imaging, a change in the Hb-O2/dHb ratio gives rise to BOLD contrast since this
induces a change in local distortions of the applied magnetic field (the main magnetic field of the
MRI scanner). This effect can be pronounced near venules and veins because venous blood is
Figure 1.1: Functional brain mapping techniques
3
more deoxygenated. Changes in the Hb-O2/dHb ratio follow a change in local neuronal
activation (Ogawa et al., 1993). This activation demands a local increase in oxygen supply to
meet the higher energy demands of synaptic activity. Local blood oxygenation levels are thought
to dip in the very brief moments after neuronal activation increase, as oxygen supplies are used.
Signaling leads to a local increase in blood flow, which not only restores the oxygenation status
of the tissue, but leads to a regional over abundance, giving rise to the resulting increase in the
Hb-O2/dHb ratio and an increase in the BOLD signal. The BOLD signal reaches a peak
amplitude followed by a return to baseline, but not before first undershooting. Collectively, these
temporal features are known as the hemodynamic response and is approximated by
hemodynamic curves (see Figure 1.2). For a brief stimulus, this overall response can take
between 12-18 seconds.
The relationship between local neuronal activity and cerebral blood flow (CBF) is referred to as
neurovascular coupling. One of the main advantages of BOLD imaging is that the BOLD signal
reflects synaptic activity, and it is therefore an indirect measure of neuronal activity.
Additionally, this increased energy use is demanded by synaptic activity, and synapses are found
in grey matter. Grey matter is responsible for various functions including motor activity, visual
information processing, executive control, auditory processing and speech, to name a few. Thus
BOLD imaging allows us to map the cortex and study brain function.
A third advantage of BOLD contrast is that using blood as a contrast agent negates the need for
an invasive contrast agent. This reduces the risk of complications, and allows BOLD imaging to
be used more widely as an exploratory modality in both healthy and unhealthy populations.
Furthermore, BOLD imaging has benefited from advancements in MRI hardware development.
Since its creation, the spatial resolution of BOLD images has gone from 4mm isotropic to 2mm
isotropic, and the temporal resolution to capture a whole brain volume of images from greater
than 3s to less than 1s (Ugurbil et al., 2013).
Recently, much of the interest in fMRI and BOLD imaging has led to massive collaborative
initiatives in a bid to use neuroimaging to better understand the human brain. One such initiative
is the Human Connectome Project, which is on course to collect fMRI data, including BOLD
data, on 1200 healthy adults which will be freely available. This project has the potential to
inform us on the relationship between brain connectivity and behavior, genetics and environmen-
4
-tal factors. Tools developed and insight gained will no doubt influence future research using
fMRI and BOLD imaging.
A notable limitation of BOLD imaging however, is the hemodynamic response timescale. While
BOLD contrast images can be acquired at ever increasing speed, the underlying hemodynamic
response is inherently slow. So while the brain may respond to a visual cue within 15-150ms, or
to a mental reasoning cue over the course of a second, the associated hemodynamic response will
take much longer. This limits the kinds of questions we can answer. It becomes a much better
idea to map low frequency activation patterns, and look for correlations in BOLD response
during block designed tasks or extended periods of rest. Another limitation of BOLD signals
relates to instrumental and participant issues that can influence the time series data that we
measure. As is the case for many types of MR imaging, head motion on the order of a few
millimeters during the BOLD fMRI acquisition can compromise our ability to detect brain
activation.
Figure 1.2: An illustration of the hemodynamic response
5
1.1.2 Task fMRI
As mentioned previously, there exists indirect coupling between neuronal activity and the BOLD
signal. This led to the first major experimental uses of BOLD fMRI: mapping brain function by
administering an in-scanner task (task-fMRI). There are three techniques for administering task-
fMRI experiments. The first involves the isolation of variables, a method similar to that used in
brain activation PET studies. Images are collected at rest, and during a task. The rest image is
then subtracted from the task activation image and tested for significant changes, ideally giving
the region(s) responsible for reacting to the task.
A second approach involves analysis by conjunction, where multiple tasks are administered, but
which all share a common processing element. Images are analyzed to compute region(s) that are
activated during all the tasks. The third and most common approach is a block design task. Here,
task blocks, usually around 30s long are interleaved with rest blocks, where the subject
disengages from the task. Roughly speaking, regions that exhibit task-related activation will have
a BOLD response that is approximately equal to a convolution of the block paradigm and the
hemodynamic response function. Suitable statistical analysis will reveal task related brain
regions.
1.1.3 Resting state fMRI
From the above discussion of task-fMRI, it is not inconceivable that during a task, we would
expect there to be a strong correlation between spatially distant brain regions that are part of the
same functional network. For example, the left and right motor cortices are expected to show
highly correlated time series data during a bilateral finger tapping task. What is not so obvious,
however, is that these motor cortices may show strong low frequency time-locked correlation
during rest as well. This phenomenon is exactly what Biswal et al. (1995) unintentionally
discovered in their now seminal paper, inadvertently giving rise to a sub-field within human
brain mapping known as resting state fMRI (rs-fMRI). This work popularized the concept of
resting-state functional connectivity (rs-fc), a cornerstone of this thesis. Two brain regions are
said to exhibit functional connectivity if their spontaneous BOLD signal time-courses' display
strong temporal correlations. These correlated fluctuations are typically observed at low
frequency across time (0.01-0.1Hz). Brain regions that exhibit this BOLD correlation are also us-
6
-ually from functionally related networks, like visual, auditory or executive networks, or in the
case of Biswal et al. (1995), a motor network. A fundamental attraction of rs-fMRI is that it
requires no active participation on the part of the subject, except that they must keep their head
still. This, coupled with BOLD fMRI's non-invasive nature, make BOLD rs-fMRI ideal for
studying diseased and elderly populations, where issues of subject task-compliance or
behavioural performance may be more of an issue.
The concept of temporal correlation is key to how functional connectivity is defined. The two
most popular analysis techniques used to uncover functionally relevant networks, or resting state
networks (RSNs) as they are better known, are seed-based correlation analysis (SBCA), the
method used by Biswal et al., and independent component analysis (ICA). In SBCA, a (typically
small) region of interest (ROI) is selected and its mean BOLD time series extracted. This time
series is then correlated against the time series of all other voxels. Regions of high correlation
constitute the RSN. Since we are regressing against tens of thousands of time series, there needs
to be a correction for multiple comparisons. This can be done by various ways, such as
Bonferroni correction, False Discovery Rate, Permutation testing and Random Field Theory.
While SBCA is a relatively simple approach to generating RSNs, it requires an a priori
hypothesis about where to place an ROI. Such an a priori hypothesis may, however, not be
known in some experiments, hence the need for more exploratory approaches. ICA is one such
approach that requires no a priori hypothesis. It is a data-driven pattern recognition method that
is capable of identifying particular patterns (“components” that are statistically independent) that
are consistent across the data, and explain significant amounts of the data’s variance. In rs-fMRI
these patterns give rise to multiple RSNs that can be identified from a single analysis. Here,
multiple comparisons must be treated differently than for SBCA, since ICA is effectively
reducing the number of observations. I defer further discussion of ICA to section 1.5, since it
forms the backbone of the analysis in this thesis.
The discovery of RSNs raised many fundamental questions, and research since has answered
many of those questions: they exist across species (Vincent et al., 2007), populations
(Damoiseaux et al., 2006) and disease states (Greicius et al., 2004), as well as in different
behavioural or functional states, including sleep (Fukunaga et al., 2006). RSNs may have clinical
utility, with previous work establishing differences in functional connectivity across disease
7
states (Fox and Greicius, 2010), genetic predispositions (Glahn et al., 2010) and after
pharmacological interventions (Khalili-Mahani et al., 2012). Massive data-sharing initiatives
have also helped to advance our understanding of fundamental biological factors, like age and
sex (Biswal et al., 2010), that affect RSNs in healthy people.
RSNs play a central role in this thesis. They give us a framework from which we can understand
functional changes in the brain after exercise, since these networks often have distinct functions
and often overlap well with neuro-anatomical landmarks. Using rs-fMRI and RSNs to
understand brain function, and changes in brain function are not new. Changes in resting state
connectivity are thought to reflect altered brain function, since synchronous activity between
spatially distinct brain regions is thought to be integral to brain function (Schnitzler and Gross,
2005; Uhlhaas and Singer, 2010). Some neurological pathologies are thought to stem from
abnormal communication between brain regions. Changes in rs-BOLD connectivity can be
transient, like after a pharmacological intervention (Khalili-Mahani et al., 2012), or long-term,
like after a stroke (Carter et al., 2010). It is this sensitivity to external interventions that prompted
me to adopt this technique in this thesis. Here, the intervention is aerobic exercise, an
intervention that has been shown to produce positive cognitive and structural brain effects, but
whose impact on rs-BOLD connectivity is not well known.
1.2 Exercise
The US Surgeon General defines physical activity as "bodily movement produced by the
contraction of skeletal muscles that increases energy expenditure above the basal level" (US
Department of Health and Human Services, Centers for Disease Control and Prevention,
National Center for Chronic Disease Prevention and Fitness, President’s Council on Physical
Fitness, 1998). Physical activity is categorized by type, intensity and purpose. Type is defined
with respect to the two properties of a muscle contraction: mechanic and metabolic properties.
Mechanically, a muscle contraction may produce movement of the limb, or it may not, leading to
isotonic (dynamic) or isometric (static) exercises, respectively. Metabolically, oxygen
availability determines classification of the muscle contraction, that is, aerobic (oxygen
available) or anaerobic (oxygen unavailable) processes. Practically speaking, however, most
activities involve a mixture of all four of these properties, and as such, activities are classified by
their most dominant feature. Intensity is defined with respect to a person’s maximal heart rate,
8
and often determines whether exercise is aerobic or anaerobic. Lastly, purpose can range from
warming up, to stretching, to improving physical fitness, to rehabilitation.
Exercise is defined as "physical activity that is planned, structured, repetitive and purposive in
the sense that improvement or maintenance of one or more components of physical fitness is the
objective" (Caspersen et al., 1985). The components of physical fitness includes speed, agility,
reaction time, balance, body composition (relative amounts of muscle, fat, bone etc.), flexibility,
endurance and skeletal muscular strength, power and endurance.
It is now well accepted that regular physical activity and exercise are essential in achieving and
maintaining health and wellbeing, with higher levels of activity being associated with better
health outcomes. The American College of Sports Medicine recommends that most healthy
adults should engage in moderate intensity aerobic exercise at least 5 days per week, or vigorous
intensity aerobic exercise at least 3 days per week. Moderate intensity exercise should be
performed for at least 150 minutes per week, while vigorous intensity exercise, for 75 minutes
per week, or a combination of both (http://www.who.int/dietphysicalactivity/factsheet_adults/en/).
Additionally, adults are recommended to engage in resistance exercise involving each major
muscle group, flexibility training and neuromotor exercise training (gait, balance, coordination
and agility), as often as 2-3 days per week.
Long term and regular moderate intensity exercise results in physiological adaptations, which
collectively can result in improved cardio-respiratory fitness. Fitness can be measured by a
person’s maximal oxygen consumption (VO2max), which is the maximal rate that oxygen can be
taken up by the lungs (Astrand et al., 2003). There are factors that play a role in determining a
person’s VO2max, such as age and gender. In addition, changes in VO2max after training are
mediated by base level fitness and intensity, frequency and duration of training. Interestingly,
sedentary individuals are often able to make the largest increase in VO2max after training.
Population studies have documented and firmly reinforced the positive health benefits of
physical activity (Pedersen and Saltin, 2006). These health benefits transcend different chronic
diseases, such as cancer (Fong et al., 2012; Knols et al., 2005), diabetes (Sigal et al., 2006), heart
disease (Fletcher et al., 1996), arthritis (Roddy et al., 2005) and respiratory illness (Garcia-
Aymerich et al., 2006). Additionally, multiple studies (Byberg et al., 2009; Samitz et al., 2011;
Wen et al., 2011) have established a clear association between exercise and all-cause mortality at
9
a large scale observational-study level. So strong is the link between exercise and positive health
benefits, that the Global Burden of Disease Study (Lim et al., 2012) has ranked physical
inactivity as the fifth leading cause of disease burden in western Europe.
Despite clearly established benefits, and well defined activity level recommendations, population
level physical activity levels remain disappointing. In the United States, 74% of adults fail to
meet the recommended prescription of 30 minutes of moderate intensity physical activity at least
5 days a week. Similarly, only 14% of adults in the UK exercise regularly, while in England
approximately one third of adults meet the recommended levels of activity.
This massive discrepancy between the recommended activity level, potential health benefits, and
actual activity levels, has an enormous socio-economic cost. In the US, sedentary lifestyles
accounted for $76 billion in medical costs in 2000 (Pratt et al., 2000), and in Canada $2.1 billion
in 1999 (Katzmarzyk et al., 2000).
1.3 Changing the brain with exercise
Exercise research has traditionally focused on the effects of physical activity on the heart and
muscles. Over the last decade, however, there has been a burgeoning body of literature on the
positive effects of exercise on the brain (Thomas et al., 2012). Animal literature provides the
framework to study the neurobiology of exercise, while human studies show that regular exercise
can produce structural brain changes, reverse age related decline and contribute cognitive
changes. Many of these results help to dispel the notion that the brain's structure is relatively
unalterable after a critical developmental period during childhood (Amedi et al., 2003; Amedi et
al., 2004; Bedny et al., 2011; Merabet et al., 2008).
How these long term effects arise, however, is not well understood, and this is where functional
brain imaging has the potential to contribute to our understanding of the exercise-brain effect.
Brain function can be temporarily altered after a perturbation, be it a cognitive task,
pharmacological agent, or physical activity. Differences in brain structure usually take months to
manifest, but if we can better understand the short term effects of a single perturbation, this may
help us understand the long term effects associated with a lifestyle change, like an exercise
intervention. This is thus part of the rationale for the work in this thesis.
10
1.3.1 Exercise and brain reorganization
Changes in the brain due to exercise can take different forms. There can be angiogenesis, the
growth of new blood vessels, neurogenesis, the growth of new neurons as well as
synaptogenesis, the growth of new synapses. While it may not be practical to distinguish
between these biological phenomena at a microstructural / biological level, it is possible with
MRI of humans to look at more macroscopic changes, such as the measurement of cerebral blood
flow, cerebral blood volume, grey matter volume and white matter volume.
The growing exercise and neuroimaging literature has to date shown that there are fairly specific
effects that occur after adopting various types of exercise programs, such as anaerobic (Winter et
al., 2007) or resistance training (Lachman et al., 2006). In terms of eliciting a brain effect, the
notion is that moderate intensity aerobic exercise is likely to produce some of the most
compelling effects, thus this is the type of exercise that we explore in this thesis.
1.3.2 Animal studies of exercise
Early rodent work (Black et al., 1990) demonstrated that 30 days of aerobic exercise (wheel
running) produced an increase in the capillary density in the cerebellum. Later, work by using
histological imaging (Kleim et al., 2002) and using MRI (Swain et al., 2003) found increases in
motor cortex blood vessel density after prolonged motor activity.
Exercise changes in rodents, however, are not only confined to motor related cortical areas. The
hippocampus of mice shows a significant increase in the number of new neurons after several
weeks of wheel running (Pereira et al., 2007; van Praag et al., 1999). Pereira et al. (2007) also
found that an increase in cerebral blood volume correlated with the number of new neurons. This
increase in motor cortex vascular density has also been shown in macaque monkeys that
underwent a 5 month program (Rhyu et al., 2010).
1.3.3 Human studies of exercise
Although insightful, many of the animal experiments in the literature involved invasive
measurements, and thus are not conducive for human studies. Despite this, there are still many
“tools” available for human exercise research. Some of the first evidence that long term exercise
has a long lasting effect on cognition (Spirduso, 1975) employed behavioural assessment tools
11
and found that older men with a history of participating in physical activities had superior
psycho-motor skills compared to non-active men of similar age, even though participants were
no longer active.
The positive effects of long term exercise have since been stratified into various areas, such as
improved cognitive function (Colcombe and Kramer, 2003; Hillman et al., 2008) and a reduced
risk of stroke (Lee et al., 2003). Recent data on the latter issue suggests that exercise may be
more effective at aiding stroke rehabilitation relative to pharmacological agents (Naci and
Ioannidis, 2013). Technical advancements in imaging technologies have allowed us to further
probe the mechanisms that underlie some of these changes, as well as uncover other brain-
exercise effects, such as structural changes. One year of moderate intensity aerobic exercise in
older adults produced a 2% growth in the hippocampus (Erickson et al., 2011). This is thought to
reflect a reversal of an age-related shrinkage, and thus exercise is thought to help protect against
the development of cognitive impairment (Jack, Jr. et al., 2010). Similar aerobic exercise work in
older adults (Colcombe et al., 2006) showed an increase in brain volume in multiple regions:
anterior cingulate, supplementary motor area, right inferior frontal gyrus and the left superior
temporal gyrus. Additionally, Pereira et al. (2007) reported an increase in cerebral blood volume
in the dentate gyrus after aerobic exercise training.
Advances in imaging also allow us to investigate functional brain changes. Early work
(Colcombe et al., 2004) demonstrated that there was a correlation between cardiovascular fitness
and functional activation in cortical areas that are involved in attentional control, in both cross-
sectional and longitudinal studies. More recent work (Voss et al., 2010) found a correlation
between fitness and functional connectivity within the default mode network. To date, however,
the rs-fMRI literature is relatively limited when it comes to studying the effects of exercise
(Takenobu et al., 2013).
1.4 Objectives of this work
To date, the literature shows the importance of regular exercise in maintaining overall
cardiovascular health, as well as the neurological benefits of exercise that result from positive
brain changes, sometimes referred to as “neuroplasticity”. While the evidence is often viewed as
very compelling, most of the focus of exercise-brain research has been conducted using
interventions that last months, or even years. By contrast, our understanding of the short term
12
effects of exercise on the brain is relatively undeveloped. I argue that to better appreciate the
relationship between long term cognitive, functional and structural changes, it is important to
establish the feasibility of measuring single session exercise-brain effects. Progress on this front,
as described in a meta-analysis (Chang et al., 2012), for example, shows a small positive effect
on cognitive performance, and this is well represented by EEG studies (Kamijo et al., 2007;
Kamijo et al., 2009; Moraes et al., 2007; Schneider et al., 2009). Here, I seek to establish the
utility of rs-fMRI in quantifying changes in the brain's rs-fc after a single session of exercise.
This will give researchers a new perspective from which to investigate the brain-exercise
relationship.
BOLD rs-fMRI is a desirable technique for various reasons. Firstly it is non-invasive. Previously,
studying cortical and subcortical brain activity meant using SPECT and PET imaging in
conjunction with radioactive tracers. Secondly, a resting condition makes no demands of the
subject to respond to a task or stimulus. Participants are asked to simply stay awake and remain
still, which are considered undemanding “tasks” while in the MRI scanner. The rs-fMRI
approach improves subject compliance, which helps to explain why rs-fMRI is being used to
study a range of populations, including children (Oldehinkel et al., 2013) and Alzheimer's
patients (Schwindt et al., 2013). Thirdly, rs-fMRI analysis is an active area of research currently
and there exist a plethora of advanced statistical analysis tools that can be used to improved
detection of rs-fc signals.
Additionally, one of the long term goals of this work is to aid in our understanding of stroke
recovery and prevention. Regular aerobic exercise significantly reduces the risk of stroke (Lee et
al., 2003; Reimers et al., 2009). These studies, however, provide little in the way of advancing
the neuro-protection mechanisms. Thus there is a need to develop additional neuroimaging
markers to advance this field of research.
One of the main challenges of this work will be to establish an optimal analysis pipeline for
detecting an exercise session-effect on rs-fc. Here I propose the use of ICA as an unsupervised
pattern recognition technique to attempt to demonstrate an exercise session effect. It is possible
to study multiple RSNs simultaneously. These RSNs are very robust, and have been derived
using small (Damoiseaux et al., 2006) and large population datasets (Biswal et al., 2010). Many
have a well-defined functional significance, especially when they overlap with anatomical
13
landmarks. ICA has been used to good effect previously in detecting session (Niesters et al.,
2012) and population differences (Filippini et al., 2009). In addition, once an analysis pipeline is
established, I will attempt to validate findings by analyzing a second “no-exercise” dataset as an
experimental control. This means acquiring two sets of rs-fMRI data without exercise being
performed between.
1.5 Analysis Techniques
1.5.1 Independent Component Analysis
Independent component analysis (ICA) is a computational and statistical blind source separation
(BSS) technique used to uncover the latent factors that underlie sets of random variables, signals
or measurements. ICA defines a generative model: unknown latent variables are linearly
combined by a mixing system, also unknown, to give the measured multivariable data matrix.
The columns of either the coefficient matrix (latent variables) or the mixture matrix are assumed
to be statistically independent, and the distribution of the variables in each column non-Gaussian.
Non-Gaussianity, as the name suggests, is the extent to which a measurement/variable is not
Gaussian. The simplest way to measure the non-Gaussianity of some variable, Y, is to calculate
its kurtosis, calculated as such:
kurt(Y) = E{Y4} - 3(E{Y
2})
2
Most non-Gaussian random variables have a non-zero kurtosis. Alternatively, entropy is another
way of measuring non-Gaussianity. Entropy is a measure of how random or unstructured a
random variable is, and it is related to information that observing that variable gives. Generally
speaking, a Gaussian variable has large entropy, while a non-Gaussian variable has less entropy,
sometimes defined as positive negentropy. A detailed discussion of entropy is omitted here, but
broadly speaking, it is a more robust measure of non-Gaussianity than kurtosis, but is too
complicated to calculate exactly and thus must be approximated.
Non-Gaussian signals can be mixed to generate Gaussian signals by virtue of the central limit
theorem (see Figure 1.3), and ICA is suitable for uncovering these non-Gaussian source signals.
It is not, however, suitable for uncovering Gaussian source signals. It is therefore the goal of ICA
to compute both the latent variable matrix and the mixing system, while imposing that one of th-
14
-ese estimates exhibit statistical independence and non-Gaussianity. These are the only
assumptions made in ICA, and are considered quite modest. No assumptions are made about the
structure of the data, leading to the apt description of ICA as a data driven pattern recognition
algorithm.
An example where ICA can be used for BSS is the classic cocktail party problem. Consider a
situation in which two people are speaking in a room simultaneously and we record their voices
with two microphones. For simplicity, we will ignore the effects of echoes or time delays. The
signal recorded by each microphone will be a linear combination of the amplitudes of both
voices. The coefficients of these linear combinations are constant but unknown to us. Each voice
is a source, and these constants define how these sources are mixed. We would like to record the
two mixtures (one from each microphone) and uncover both the sources and mixing coefficients.
This can be solved using ICA. This problem can be generalized to K sources (voices), but
requires at least K sensors (microphones) to be solvable. This reduction of observed data into
source vectors allows us to better understand the data. ICA builds upon the foundation of two
other popular techniques, principal component analysis and factor analysis. ICA is considered
more robust at uncovering latent variables compared to the PCA approach of producing
Figure 1.3: Non-Gaussian sources can be mixed together to form Gaussian signals
15
uncorrelated “sources”, but instead produces independent “sources” and represents more
rigorous statistical criteria.
I will now present a theoretical description of ICA adapted from Jenkinson and Beckmann (,
2001). For simplicity, I assume square matrices when possible, and ignore explicitly modelling
Gaussian noise.
Given a data matrix YT×N, we would like to express Y in terms of a product of two matrices,
AT×T, the source matrix, and ST×N, the mixing matrix:
Y = AS (1)
Here I assume that each column in Y has had its mean subtracted ( i.e. each column has a mean
of zero), and has been scaled to have unit variance. We would like to impose independence and
non-Gaussianity on S, the mixing matrix, so we assume T < N.
The ICA problem in its simplest terms boils down to: find a matrix A, which optimizes the
function f(S), where Y = AS. Here f(·) is a measure of non-Gaussianity. The general principle
here is that we now have an optimization problem, which needs to be solved iteratively.
Before we solve this iteration problem, however, we use PCA to derive one more constraint. We
compute the singular value decomposition of Y:
Y = UDVT (2)
where U and V are orthogonal (UUT = I, VV
T = I, I is the identity matrix). Now let AP = UD, SP
= VT and list the singular values of D in decreasing order. Now express Y as:
Y = APSP (3)
Since PCA uses correlation as an objective function, the columns of SP are uncorrelated, and also
have unit variance.
One of the strengths of ICA is its ability to perform dimensionality reduction. We do this by
computing the variance (power) associated with matrix AP:
16
(AP)TAP = D
TU
TUD = D
2 (4)
D2 is a matrix of the square of the singular values. Thresholding these power values at some
arbitrary value leave M values, and we keep only the first M components. Thresholding is
especially useful when attempting to keep components that are not noise related. This reduces
the size of both AP and SPT:
AP = [A1 : Areject] and SPT = [S1
T : Sreject
T] (5)
Now we define Y1 as:
Y1 = A1S1 (6)
Even though the dimension of S1 is reduced, it is still orthogonal. Now, re-express S1 as:
S1 = QQ-1
S1 = QS2 (7)
where Q is a new matrix, and S2 = Q-1
S1. Observe that the correlation of matrix of S2 is:
S2S2T = Q
-1S1S1
T Q
-T = Q
-1Q
-T = (Q
TQ)
-1 (8)
We would like to keep the vectors of S2 independent, so we need the above expression to equate
to the identity. This requires that Q be an orthogonal matrix. This is the last constraint we need
for our optimization problem. Now we have derived the equation:
Y1 = (A1Q)S2 (9)
where Y1 and A1 are known, Q is required to be orthogonal, and f(S2) must be optimized for non-
Gaussianity.
Returning to the example of the cocktail party problem, solving for S2 would give the
coefficients that represent how much a speaker contributes to each microphones signal, and A1Q
would give the source signals. When ICA is used on BOLD data, the S2 matrix turns out to be
17
statistical spatial maps of functional brain networks, and the A1Q matrix represents the
associated time series of these networks. As I explain in the next section, these spatial maps
overlap nicely with our previous knowledge of brain networks, making ICA a favorable choice
for BOLD-fMRI image analysis.
1.5.2 ICA and BOLD fMRI
Over the last 10 years, ICA has proven to be a powerful technique for analyzing both task-fMRI
and rs-fMRI data. A BOLD-fMRI image is a four dimensional grayscale image, the fourth
dimension being time. To make this image amenable to ICA, we take a 3D coordinate, or voxel,
and vectorize its time course into a column. We do this for all voxels, and form a 2D data matrix,
our Y matrix. This matrix has dimensions of time × space (e.g. the matrix could have 200 time
points and 20,000 voxels). Thus, when ICA solves for the 2 matrices, we solve for temporal and
spatial information. We have the choice of imposing our independence and non-Gaussianity
constraints on the temporal dimension or the spatial dimension. Since BOLD-fMRI images have
spatial information in the order of tens of thousands of voxels, but temporal information in the
order of a few hundred time points, we are restricted to performing spatial ICA.
Thus, when applied to BOLD data, ICA produces a set of components, each with a spatial map
and an associated time series. In task-fMRI, the time series are usually time locked to the task. In
rs-fMRI however, because the BOLD signal is spontaneous, these time series show little to no
observable temporal pattern. An exception to this rule is in the case of ICs that reflect signal
ofnon-interest in the data, like eye motion or cerebral spinal fluid (CSF) signal which reflect
physiological noise (cardiac cycle, breathing). This can be used to our advantage if we would
like to remove (de-noise) data. Nevertheless, in rs-fMRI, the spatial maps are of greater interest.
One of the earliest works to successfully employ ICA to rs-BOLD fMRI data (Beckmann et al.,
2005) used it to generate spatial maps that overlap well with neuro-anatomical as well as
functional parcellations of the brain. These parcellations included a medial visual area, a lateral
visual area, an auditory system, a sensorimotor system, an executive control system and two
dorsal visual stream networks. ICA is also capable of generating the default mode network
(DMN), a network that exhibits a reduction in activity during tasks that require external
attention, and an increase in activity during periods of rest and introspection. It was first
18
discovered in work using PET imaging (Raichle et al., 2001). Additionally, as with the temporal
signals, spatial maps can be used to identify signals of non-interest. Head motion, cerebral spinal
fluid pulsation noise, and eye motion artifact have distinct spatial patterns. These facts
demonstrate just how robust and powerful ICA is, especially when we consider that it requires no
a priori hypothesis about the data. This makes ICA ideal for exploratory analysis. This is
especially advantageous for rs-fMRI, since the BOLD signal is spontaneous, and we are not able
to model its behaviour.
One significant issue related to ICA, which is considered a disadvantage of the analytical
technique, is the challenge of interpreting the components. Some ICs will have no obvious
anatomical or functional meaning, leaving the user with the challenge of trying to reconcile the
relevance of each component. Nonetheless, ICA has been used to great success at reproducing a
set of anatomically and functionally relevant networks, across various populations, ages, genders
and scan parameters (Biswal et al., 2010; Smith et al., 2009). Beckmann's implementation of
ICA is contained in the MELODIC software package (Multivariate Exploratory Linear
Figure 1.4: a) Single subject ICA is useful for providing subject specific maps, while b) Group ICA
produces more robust spatial maps since it utilizes more information
19
Optimized Decomposition into Independent Components,
http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/MELODIC). This package provides two versions of ICA: single
subject ICA (iICA), and group ICA (gICA). iICA works as ICA generally would, by vectorizing
the image data. gICA however, vectorizes multiple images, and then stacks them in the temporal
dimension before computing components (see Figure 1.4). gICA produces much more robust
spatial maps than iICA, since it uses more data to compute components. iICA is useful in
detecting signals of non-interest at the subject level. These signals can then be removed, acting
as a data driven and brute force denoising method. A limitation of this, however, is that one must
be judicious, as to not remove signals of interest, i.e. those related to neuronal activity.
The uses of ICA in biomedical applications extend well beyond BOLD imaging, and extends to
EEG and MEG research. Quite remarkably, a study (Brookes et al., 2011) used ICA on resting
state MEG data to generate some of the canonical rs-BOLD networks obtained with ICA (see
Figure 1.5). As mentioned previously, MEG uses not the hemodynamic response, but magnetic
fields modulated by current flow in neuronal assemblies (Cohen, 1972). This work further
validated the use of the hemodynamic response as an indirect measure of brain activity, and as a
way of generating functionally relevant networks worthy of study. It is also a testament to the
power of ICA.
Figure 1.5: ICA is capable of producing spatially similar and functionally relevant ICs: (A) Default mode
network (B) Left Lateral Frontoparietal Network (C) Right Lateral Frontoparietal Network (D) Sensorimotor
Network (E) Medial Parietal Region (F) Visual Network (G) Frontal Lobes (H) Cerebellum (adapted from
(Brookes et al., 2011), reprinted with permission)
20
Overall, ICA is a powerful analysis technique that has proven itself to be robust and reliable. Its
exploratory nature makes it an indispensable tool for rs-fMRI BOLD imaging studies, and when
used judiciously, it can be used to remove structured noise in BOLD data. While various
techniques exist, the robust and automated removal of noise in BOLD data remains an elusive
goal (Murphy et al., 2013). For these reasons, I've chosen ICA as the primary technique in
generating RSNs and denoising rs-BOLD data. Comparing RSNs between sessions requires
further analysis. For this work, I use dual regression (Filippini et al., 2009), an established
technique, as well as Cohen's D, which is also well established but rarely used (if ever) in rs-
BOLD image analysis. These methods are outlined in more detail in Chapter 2.
The principle aim of this thesis is to develop a method for extracting and contrasting salient
features from rs-BOLD data, for the purpose of detecting an effect of a single session of exercise
on brain connectivity. This method will first be established in Chapter 2, using a young healthy
cohort, before being extended and refined in Chapter 3 to a clinical population of stroke patients.
21
Chapter 2 Methods and Results
2
In Chapter 1, I introduced the central modality, themes and techniques of this thesis work. These
include resting-state BOLD-fMRI, exercise, resting state networks and independent component
analysis. Moreover, I established that while long term exercise has been shown to have positive
brain effects, our understanding of the short term effects is limited. I then postulated the used rs-
BOLD fMRI as a non-invasive tool for studying functional brain networks, and detecting an
effect of a single session of aerobic exercise on these networks’ activity.
In this Chapter, I present the main study of this thesis and its findings. A young healthy adult
cohort was scanned before and after engaging in a single session of moderate intensity aerobic
exercise. Stationary cycling was chosen as the form of exercise because it is simple, requires
minimal motor learning, and is conducive to the monitoring of subject performance and
compliance.
A manuscript version of the work contained in this chapter has been submitted for publication
and is currently under peer review.
2.1 Materials and Methods
2.1.1 Participants and Study Design
Table 2.1: Participant Demographics
Characteristics Exercise No-exercise
N 15 15
Ages (years) 26.1 ± 4.3 27.0 ± 6.5
Sex (M/F) 6/9 8/7
Handedness (L/R) 3/12 0/15
Exercise dataset: Sixteen young healthy adults (10 women) between the ages of 20 and 35 years
were recruited for this study. One subject’s MRI data were corrupted and consequently this
22
parti
cipant was excluded from analysis (see Table 2.1). Exclusion criteria included contraindications
to MRI or inability to complete the exercise session. The Sunnybrook Health Sciences Centre
Research Ethics Board approved this study and all participants provided signed informed
consent. Detailed below, the study protocol entailed: a baseline pre-exercise MRI scan, a 25-min
exercise session, a 10-min no-exercise cool down period and a post-exercise MRI scan.
No-exercise dataset: A second dataset was included in this study to act as an exercise null
scenario to help validate exercise findings (see Table 2.1). MRI data were accessed from the
Neuroimaging Informatics Tools and Resources Clearinghouse (NITRC) database
(http://www.nitrc.org/frs/?group_id=296; dataset NewHaven_b). This dataset included multi-
session rs-fMRI in a cohort of 16 healthy (7 women) adults between the ages of 18 and 42, with
comparable MRI scan parameters to the exercise dataset. One dataset was randomly removed so
that this group matched the exercise group. Although 4 rs-fMRI sessions were available for each
subject, I restricted my analysis to sessions 1 and 2, as to more closely match the exercise data
since they preceded an actual task in the scanner. Session 2 followed immediately after session 1.
2.1.2 Single Session Exercise
Exercise was performed on a semi-recumbent cycle ergometer. Participants spent 2 minutes
cycling to warm up, 20 minutes exercising at 70% (moderate intensity) of their age-predicted
maximum heart rate (HR), and finally 3 minutes cooling down at a self-determined pace (see
Figure 2.1). The age-predicted maximum HR, defined as 220 beats-per-minute (bpm) minus age
Figure 2.1: Data collection timeline
23
in years (Londeree and Moeschberger, 1984), was used to determine the exercise target HR. HR
was monitored in real-time using a Polar heart rate monitor and participants were given verbal
feedback if they deviated from their target HR. During exercise, power output (Watts) and
pedaling frequency (rpm) were recorded every minute, and ratings of perceived exertion (RPE;
self-reported using a 10-point Borg scale) were recorded every ten minutes. Systolic and
diastolic blood pressure (mmHg) were recorded before and immediately following exercise.
2.1.3 Magnetic Resonance Imaging
2.1.3.1 Exercise dataset
MRI data were acquired on a Philips 3T Achieva MRI scanner equipped with an 8-channel head
coil. Rs-fMRI BOLD data were acquired pre- and post-exercise. A T1 anatomical image was
also acquired post-exercise for registration purposes. The imaging parameters were: T2*-
weighted echo planar imaging (TR/TE = 1500/30 ms, FA = 70°, 80×80×28 matrix, voxel =
3.0×3.0×5.0 mm, 28 slices, 230 volumes, acquisition time = 6:00 min:s), T1-weighted structural
(TR/TE/TI = 9.5/2.3/1400 ms, FA = 8°, 256×164×140 matrix, voxel = 1.0×1.2×1.2 mm,
acquisition time = 8:38 min:s). Participants were instructed to refrain from drinking caffeine an
hour before scanning and to keep their eyes open during the resting state scans.
2.1.3.2 No-exercise dataset
MRI data were acquired on a Siemens 3T scanner, intra-sessionally without an exercise
intervention. Scan parameters were: EPI BOLD (TR/TE = 1500/25 ms, FA = 80°, 64×64×22
matrix, voxel = 3.4×3.4×5.5 mm, 22 slices, 181 volumes, acquisition time = 4:31 min:s), T1-
weighted structural (TR/TE/TI = 2000/3.67/1100 ms, FA = 7°, 256×256×160 matrix, voxel =
1.0x1.0x1.0 mm, acquisition time = 6:24 min:s). Participants were instructed to keep their eyes
open during the rs-fMRI scans.
2.1.4 MRI Analysis
Analysis was conducted using FMRIB Software Library (FSL) tools (Jenkinson et al., 2012) and
specially developed scripts. Structural images were processed using FSL’s Brain Extraction
Tool (BET) (Smith, 2002), and visually inspected for optimal extraction. Functional images
were analysed using FSL’s Multivariate Exploratory Linear Optimized Decomposition into
Independent Components (MELODIC) tool (Beckmann et al., 2005). Automatic dimensionality
24
estimation (Laplace) was used for all MELODIC analyses. Data were analysed using two
pipelines: 1) an unsupervised analysis approach (Raw), and 2) a Visually Inspected STructured
Artefact Removal (VISTAR) noise suppression approach.
Raw pipeline: Data were pre-processed: motion corrected, high pass filtered, slice time
corrected, spatially smoothed (5mm FWHM using a Gaussian kernel), linearly registered to each
individual’s structural image and then transformed to a 3mm Montreal Neurological Institute
(MNI) template. Group-level MELODIC analysis identified several RSNs. Given the paucity of
rs-fMRI exercise literature, I focused on RSNs that have well established functions, or those
potentially related to motor activity. Additionally, two other RSNs were selected, the default
mode network (DMN) RSN, for its perceived clinical utility (Schwindt et al., 2013), and the
medial visual RSN, as a control.
A dual regression technique (Filippini et al., 2009) was used to produce participant/session-
specific RSNs. First, for a given RSN, the spatial map was used as a regressor against a subject’s
pre-processed BOLD data. This produces the time-series associated with the RSN for that
subject. This time series was then regressed against his pre-processed BOLD data, producing a
subject specific spatial map image for this RSN. The value at each voxel in this image
represented regression coefficients. This was repeated for each subject.
To investigate resting-state activation differences in an RSN between sessions, these resulting
subject-specific regression coefficient images were compared with a paired t-test across the
group in a voxel-wise analysis and corrected for multiple comparisons using FSL’s Randomise
(5000 permutations, threshold-free cluster enhancement; (Nichols and Holmes, 2002)). To boost
the statistical power of this analysis, the t-test was restricted to voxels contained in the RSN, i.e.
an RSN mask was used. This voxel-wise analysis was performed for each of the RSNs under
investigation.
Furthermore, to verify this voxel-wise approach using a less computationally intensive ROI
based approach, these participant-level RSNs were compared using a second method: calculating
the Cohen’s D effect size of session-related changes in RSN activation, and testing for a
significant non-zero effect across the group mean. Analysis was again limited to voxels
circumscribed by the group RSN volume.
25
VISTAR pipeline: Pre-processing steps were carried out in the same manner as in the Raw
pipeline, but with the addition of a subject-level MELODIC. These subject-specific MELODIC
outputs were inspected manually and ICs of non-interest (ICnoise) were identified and removed
from the pre-processed data (see Figure 2.4 for examples if ICnoise components). ICnoise were
components consisting of head motion, eye motion, CSF signal, high frequency signals and other
spurious signals as described in a previous MELODIC study (Kelly, Jr. et al., 2010). The
VISTAR datasets were then used in a group-level MELODIC, and RSNs of interest were again
identified. Dual regression and permutation testing, as well as Cohen’s D analysis were repeated.
This analysis was also limited to group level RSN masks. A spatial correlation coefficient was
used to identify comparable RSNs between Raw and VISTAR pipelines.
2.2 Results
2.2.1 Single Session Exercise
The average age and body mass index of the 15 exercising participants (9 women) were 26.1 ±
4.1 years and 22.5 ± 2.1 kg/m2, respectively. HR during the moderate intensity exercise session
was 70 ± 3.5 % of age-predicted maximum. The average RPE was 3.9 ± 1.4, after 10 minutes of
exercise and 4.5 ± 1.8 after 20 minutes; both of which fall in the descriptive range of “somewhat
strong” to “strong”. Systolic blood pressure remained significantly higher than baseline at 10
minutes post-exercise (baseline: 112 ± 12 mmHg, post: 118 ± 13 mmHg, p=0.007). Diastolic
blood pressure, however, was not significantly different (baseline: 74 ± 9 mmHg, post: 73 ± 8
mmHg, p=0.544). The mean time between the end of exercise and the second rs-fMRI scan was
16:51 ± 3:33 min:s. Finally, HR was significantly higher at the start of the post-exercise rs-fMRI
scan (baseline: 66 ± 10 bpm, post: 76 ± 10 bpm, p<0.001).
2.2.2 Exercise dataset
Raw pipeline: Using group level ICA, I identified the following eight RSNs: a) Sensorimotor, b)
Auditory, c) Default Mode (DMN), d) Pre/Post-Central Gyri (ppCG), e) Medial Visual, f)
Attention, g) Executive and h) Basal Ganglia (see Figure 2.2). After cluster-enhancement and c-
26
Figure 2.2: Group ICA maps overlaid on an MNI template. The Raw pipeline produced eight spatial
RSN maps and the VISTAR pipeline produced nine, with the basal ganglia being split into two
components. Both pipelines produce spatially similar maps (spatial cross correlation (scc)) (a)
Sensorimotor [MNI: 0,-14,52]: scc=0.88; (b) Auditory [54,-16,2]: 0.88; (c) DMN [0,-56,32]: 0.79; (d)
ppCG [42,-14,24]: 0.85; (e) Medial Visual [2,-58,0]: 0.88; (f) Attention [-34,18,42]: 0.80; (g) Executive
[0,14,20]: 0.89; (h) Basal Ganglia [-16,-14,2]: 0.77 [i], 0.67 [ii].
R L
27
-orrection for multiple comparisons, voxel-wise analysis showed no effect of session. Cohen's D
analysis obtained borderline insignificance in the auditory, ppCG and basal ganglia RSNs
(p=0.075, p=0.081, p=0.117 respectively), while the other 5 RSNs showed no significant effect
size (p>0.264).
VISTAR pipeline: VISTAR removed 36% of a total 1714 ICs (see Table 2.2 and Figure 2.3).
There was no significant difference in the number of ICs produced using single-session ICA pre-
vs. post-exercise (p=0.274), or the number removed pre- vs. post-exercise (p=0.223) by session
(see Figure 2.3). Group level ICA produced nine RSNs - the same previous eight identified by
the Raw pipeline, with the exception of the basal ganglia, which was split in two components:
Putamen and Thalamic/Caudate RSNs (see Figure 2.2). There was good spatial overlap between
Raw and VISTAR RSNs (spatial overlap range: 0.67 - 0.88; see Figure 2.2). Three RSNs were
identified as showing significant voxel-wise session differences (pcorrected<0.05), with clusters in
the following regions (see Figure 2.5): 1) the auditory RSN showed increased activation in the
central and parietal operculum cortices, 2) the sensorimotor RSN showed increased activation in
the medial post-central gyrus, and 3) the thalamic-caudate RSN shows increased activity in the
right thalamus. Additionally, a cluster of voxels in the ppCG exhibited a trend of increased post-
exercise activity (pcorrected<0.07), centred at MNI [-52, -22, 34]. The results of Cohen’s D RSN-
based (ROI) analysis revealed that the auditory, sensorimotor and thalamic-caudate networks
exhibited a significant session effect (p=0.014, p=0.049, p=0.023 respectively; see Table 2.3).
The ppCG RSN showed a trend (p=0.059), while all other RSNs were not significant (p>0.428).
Table 2.2: Details of VISTAR IC removal
Characteristics Exercise No-exercise
Total number of ICs 1714 2032
Total number of artefact ICs removed 624 625
Mean number of ICs per participant per session (SD) 57 ± 4 68 ± 3
Mean number of artefact ICs per participant per session
(SD)
21 ± 4 21 ± 5
Min/Max % of removed artefact ICs per session 23/50 18/48
28
Figure 2.3: The number of ICs used for the RSN group analyses in the exercise dataset from the Raw
and VISTAR pipelines. No significant differences were noted in the number of ICs Pre- vs. Post- Raw
(p=0.274), or Pre- vs. Post- VISTAR (p=0.747).
Figure 2.4: Examples of single session ICA artefacts: a) cerebral spinal fluid pulsation b) eye motion c)
head motion d) high frequency noise (>0.1Hz) e) spurious signal
29
Figure 2.5: Post-exercise changes in rs-fMRI. VISTAR RSNs (yellow) overlaid with cluster
enhanced-regions of increased activation (green): (a) Auditory [MNI: -42,-24,12]; (b) Sensorimotor
[0,-40,72]; (c) Thalamic-Caudate [10,-10,-2].
R L
30
Table 2.3: Cohen's D effect size comparing pre- vs. post- (exercise) and session 1 vs. 2 (no-
exercise) using the VISTAR Pipeline
RSN
Exercise dataset No-exercise dataset
mean ± SD p-value mean ± SD p-value
Sensorimotor 0.18 ± 0.33 0.049 * 0.04 ± 0.27 0.591
Auditory 0.18 ± 0.25 0.014 * 0.07 ± 0.41 0.545
DMN 0.04 ± 0.19 0.480 0.02 ± 0.22 0.756
ppCG 0.13 ± 0.25 0.059 0.02 ± 0.55 0.875
Medial Visual 0.05 ± 0.26 0.428 0.05 ± 0.38 0.651
Attention 0.00 ± 0.21 0.957 -0.03 ± 0.32 0.765
Executive -0.02 ± 0.14 0.607 0.02 ± 0.41 0.823
Putamen 0.04 ± 0.36 0.661 0.03 ± 0.40 0.796
Caudate/Thalamus 0.22 ± 0.34 0.023 * 0.11 ± 0.39 0.308
* => p < 0.05
2.2.3 No-exercise dataset
The average age of the 15 participants (7 women) was 27.0 ± 6.5 years. There was no significant
difference in ages between the exercise and no-exercise subjects (p=0.791).
Raw pipeline: Group level ICA produced the same nine RSNs as the exercise dataset when
analysed with the VISTAR pipeline. Voxel-wise analysis showed no effect of session after
cluster-enhancement and correction for multiple comparisons, while Cohen’s D analysis showed
a session-effect in the executive RSN (p=0.009).
VISTAR pipeline: VISTAR removed 31% of a total number of 2032 ICs (see Table 2.2). There
was no significant difference in the number of ICs produced using single-session ICA session 1
vs. 2 (p=0.595), or the number removed from session 1 vs. 2 (p=0.529; see Figure 2.3). Group
level ICA produced the same nine RSNs as in the Raw pipeline. No significant session
differences were identified (p>0.329). Cohen’s D analysis found no significant session effect size
(p>0.244; see Figure 2.6 and Table 2.3 for a comparison between these results and the exercise
results).
31
2.3 Discussion
This study demonstrates the feasibility of using rs-fMRI to detect and localize acute effects of
exercise on brain connectivity. I observed a change in the resting state BOLD functional
connectivity of young healthy adults in three RSNs, predominantly localized to cortical areas
involved in sensorimotor activity. The exercise effect was sensitive to voxel-wise and Cohen’s D
analyses, but only after the data were processed with a de-noising pipeline. These methods also
demonstrated good test-retest reliability using the no-exercise control dataset.
The central and parietal operculum cortices, commonly referred to as the secondary
somatosensory region (S2), showed increased activation after exercise. The S2 region has been
shown to modulate tactile attention (Burton et al., 1999). Increased activation observed in the
current study was primarily lateralized to the left S2. In addition, a cortical area responsible for
processing tactile sensation from the lower limbs also exhibited an increase in activation. This
area forms part of the sensory homunculus, and the sensorimotor RSN generated by ICA. A
Figure 2.1: Cohen’s D values for the 9 RSNs of interest using exercise (N=15) and no-exercise (N=15) datasets
(VISTAR). SM: Sensorimotor, Aud: Auditory, DMN: Default Mode Network, ppCG: Pre and Post-Central Gyri,
Med-Vis: Medial Visual, Atten: Attention, Exec: Executive, Put: Putamen, Th/Cau: Thalamus and Caudate
32
previous PET study demonstrated that an analogous region of the motor homunculus exhibited
increased activity during cycling (Christensen et al., 2000).
Finally, I observed an exercise-related effect in a basal ganglia RSN, specifically in the thalamus.
This RSN is of interest given the role that the basal ganglia play in motor learning and reward
(Alexander et al., 1990), but to date has not been reported in structural MRI exercise-brain
studies that utilize long-tern exercise programs (Colcombe et al., 2006; Erickson et al., 2011).
Previous studies have shown positive changes in cognition (Yanagisawa et al., 2010) and
attention (Kamijo et al., 2004) after acute exercise; however the results of this chapter were
restricted to sensorimotor related networks.
The areas of the ppCG RSN that exhibit a trend of increased activation post-exercise is situated
in the sensory homunculus. This area was left lateralized, and corresponds to a cortical region
responsible for processing sensation from the face and hands. While it is unclear why these areas
would be up regulated post-exercise, this is further evidence that the post-central gyrus
experiences an increase in connectivity post-exercise
Exercise participants were compliant with the experimental protocol. During the 15-minute
transition period between the end of exercise and the start of the repeat rs-fMRI scan, heart rate
and arterial blood pressure returned close to their baseline steady state levels. The no-exercise
dataset was selected because it closely matched the age and scan parameters of the exercise
dataset. Performing the comparable analyses on both datasets helped contextualize my findings.
To my knowledge, this is the first study that examined changes in rs-BOLD functional
connectivity after a single session of exercise and helps to establish the feasibility of probing
acute brain effects of exercise. A novel aspect of this study was the use of the Cohen’s D to
assess the session differences within an entire RSN. This procedure detected an effect of session
in the same three aforementioned RSNs. While Cohen's D does not localize sub-regions within
an RSN, it is much less computationally intensive than permutation testing and proved to be
sensitive to a session effect. This approach may be conducive to compare longitudinal changes in
RSNs of interest, like the DMN (Greicius et al., 2004; Schwindt et al., 2013; Tanabe et al.,
2011).
33
Another important contribution of this study is the added value of “denoising” rs-fMRI data as a
means to increase the sensitivity of these analyses to detect exercise effects. Artefact ICs (i.e. ICs
of non-interest) are well-established; however, no robust tool exists for automatically detecting
ICs of non-interest (see Murphy et al. (2013) for a review of denoising techniques). For this
study, each dataset was decomposed using MELODIC into ICs, and each IC's spatial and
temporal signatures were manually inspected. The manual inspection approach is time-
consuming (i.e. approximately 30 minutes to evaluate ICs from one dataset); however, visual
inspection is still the gold standard for artefact detection when using ICA.
This study however, is not without its limitations. Systolic blood pressure was elevated post-
exercise, while breathing rate and arterial blood O2 and CO2 levels were not monitored. Although
I cannot rule out the possibility that differences in global vascular regulation during the post-
exercise scans contributed to the altered functional connectivity patterns, the VISTAR approach
was designed to remove physiological noise and thus minimize these potential mitigating
physiological effects. However, the specificity of the finding to motor areas of the brain aligns
with previous work in the lab on cerebral blood flow changes (MacIntosh et al., 2014), in support
of the notion that exercise can lead to a neuronal carryover effect.
Furthermore, the no-exercise dataset represents an intra-session scanning protocol, as opposed to
the inter-session design of the exercise protocol. The no-exercise cohort was a post hoc dataset,
used to approximate a test-retest scenario. This study was designed to focus on the exercise
session effect, and so for reasons of limited resources, it was not possible to collect no-exercise
data on the exercise cohort. Thus, I decided to take advantage of a freely available imaging
resource. Nonetheless, I believe that this secondary dataset helped contextualize my findings.
Additionally, I contacted the principle investigator responsible for collecting this data, in order to
obtain the scan protocol, as well as verify the scan parameters.
Future work is needed to investigate the effect that exercise parameters, like type, duration and
intensity, as well as baseline fitness levels, as assessed by VO2peak, have on acute exercise
responses. Additionally, although outside the scope of the current study due to sample size
considerations, age, sex and genetics have been shown to affect baseline functional connectivity
(Biswal et al., 2010; Glahn et al., 2010), and investigating their influence on exercise effects will
further our understanding of the scope of the exercise-brain relationship. Further work is also
34
required to establish the optimal time post-exercise to best observe functional connectivity
changes. Lastly, while this study established within-RSN session differences additional work is
required to develop methods sensitive to inter-RSN session changes. In conclusion, these results
suggest that rs-fMRI can be used to assess changes in functional connectivity that may relate to
an acute exercise session.
35
Chapter 3 Future Work
3
3.1 Acute effects of exercise on Chronic Stroke Adults
The focus of Chapter 2 was to establish rs-fMRI as a viable modality to detect effects of acute
exercise on brain connectivity. A young healthy adult cohort was used because this group is ideal
for studying baseline rs-fc brain-exercise effects and presents minimal challenges to recruitment.
Now that I have demonstrated rs-fMRI as a viable tool for studying the exercise session effects,
future research will extend this experimental paradigm and analysis to a population that may
garner significant brain benefits, stroke survivors. As mentioned previously, studies show that
regular exercise has neuro-protective effects against stroke (Lee et al., 2003), thus I would like to
understand the session effect of exercise on stroke survivors. In this Chapter, I will present
preliminary analysis in a study of 10 such subjects.
The experimental protocol for this experiment builds upon the scan-exercise-scan structure of the
previous Chapter, with a few refinements aimed at: 1) assessing the test re-test reliability of this
rs-fMRI dataset, 2) exploring the effect of exercise intensity “dose” and 3) exploring the effect of
fitness on rs-fc. First, to better understand the test re-test reliability of the data, I scanned
volunteers on two separate days, before exercise, and use a similar analysis procedure described
in Chapter 2 to investigate inter-session differences. Second, to characterize the effect of exercise
intensity, i.e. the dose effect, volunteers engaged in exercise on two different days, at two
different intensities, 50% and 70% of their peak exertion (VO2peak). Third, peak VO2 was
assessed during a supervised cardiopulmonary exercise test as a measure of fitness. My
preliminary hypotheses were: 1) a VISTAR pipeline analysis of stroke survivors will produce
RSNs similar to those identified in Chapter 2, namely the sensorimotor, auditory, default mode,
ppCG, medial visual, attention, executive and basal ganglia RSNs, 2) it will be possible to detect
a session effect in select RSNs, namely the auditory, sensorimotor and basal ganglia RSNs, and
3) VO2peak will be associated with RSN activity in the sensorimotor, auditory, default mode,
ppCG, attention and basal ganglia networks.
36
This exploratory work, however, will present unique challenges to recruitment. Criteria include:
volunteers must have suffered an ischemic or hemorrhagic stroke, be more than six months post-
stroke at the time of enrollment, be able to exercise on a stationary bicycle for twenty minutes
and walk 10 meters with an assistive device if needed but without physical assistance from
another person, must be older than 30 years of age, must have a Montreal Cognitive Assessment
(MoCA) score greater than 22 and be able to consent (e.g. no substantial aphasia communication
deficit). Exclusion criteria include contraindications to MRI scanning or a history of depression
or psychiatric illness. Analysis of rs-BOLD data will be comparable to the steps listed in Chapter
2.
3.1.1 Participants and Study Design
Collection of these data commenced April 2012, and is ongoing. The trial is called
“Neurovascular Underpinnings of Exercise after Stroke” (NUES). The Sunnybrook Health
Sciences Centre Ethics Board approved this study and all participants provided signed informed
consent and thus far 17 volunteers have been enrolled.
The study protocol involves four visits (see Figure 3.1). The first visit to Sunnybrook Health
Sciences Centre (SHSC) was used to collect patient medical history information, psychometric
testing, informed consent and MRI screening. Any participants deemed unsuitable at this stage,
for example due to an MRI-incompatible stent, were excluded. Second, participants visit the
Toronto Rehabilitation Institute Rumsey Centre, where they are administered a cardiopulmonary
exercise fitness test to determine whether or not they would be able to complete the exercise
intervention as well as to obtain VO2rest and VO2peak exertion values, and the differential,
VO2peak – VO2rest (Δ). These values were used to calculate each volunteer's target heart rate for
Figure 3.1: Neurovascular Underpinnings of Exercise after Stroke trial timeline for each participant
37
their two exercise sessions, a low intensity session (VO2rest + Δ×50%) and a moderate intensity
session (VO2rest + Δ×70%). Electrocardiogram data are taken during this visit and used to
exclude volunteers for whom the exercise session may pose a health risk. Visits three and four
were at SHSC and involved exercise and MRI scanning. Similar to the previous study, the
protocol on each visit entailed: a pre-exercise MRI scan, a 25 minute exercise session, and a
post-exercise MRI scan.
Exercise was performed on a semi-recumbent cycle ergometer after a baseline MRI session.
Participants spent 2 minutes cycling to warm up, 20 minutes exercising at low or moderate
intensity, and finally 3 minutes cooling down at a self-determined pace (see Figure 3.2). HR was
monitored in real-time using a ZephyrMD heart rate monitor and participants were given verbal
feedback if they deviated from their target HR. During exercise, power output (watts) and
pedaling frequency (rpm) were recorded every minute, and ratings of perceived exertion (RPE;
self-reported using a 10-point Borg scale) were recorded every ten minutes. Systolic and
diastolic blood pressure (mmHg) were recorded before and within 5 minutes after exercise. Each
volunteer had one low-intensity and one moderate-intensity session, and these were counter-
balanced by visit. Participants were wheeled to and from the MRI scanner to ensure that they
engaged in minimal physical exertion outside of the exercise session.
Figure 3.2: Data collection timeline for visits 3 and 4. MRI sequences in green, biometrics data in blue
38
Table 3.1: NUES participant demographics
Characteristics All subjects Full-data subjects
N 17 10
Ages (years) 62.1 ± 15.0 62.0 ± 16.1
Sex (M/F) 12/5 6/4
Handedness (L/R/?) 2/4/11 2/3/5
Stroke Hemisphere (L/R) 12/5 7/3
Time since stroke (months) 24.5 ± 17.0 22.3 ± 17.3
Education (years) 16 ± 4 15 ± 4
fVO2 - 0.7 ± 0.2
Stroke: Haemorrhagic/Ischemic/? 4/12/1 2/8/0
MoCA 22 ± 3 23 ± 4
MRI data were acquired on a Philips 3T Achieva MRI scanner using an 8-channel head coil, and
scan parameters were identical to those described in Chapter 2. The MRI protocol included rs-
BOLD data collection before and after the exercise session. Of the seventeen stroke volunteers
recruited thus far, full imaging data has been collected on eleven, ten of whom also completed
the exercise test, four were not MRI compatible and two were not exercise compatible (see
Table 3.1 for detailed demographic information).
3.1.2 Preliminary Analysis and Results
The preliminary results presented in this section are based on complete datasets available from
ten participants. Figure 3.3 shows the approximate stroke lesion location for each participant.
Participants had relatively modest to minor sensorimotor stroke deficits.
Exercise: For the low intensity exercise session, the average HR was within 4% of the target.
The average RPE was 3.4 ± 0.8 after 10 minutes of exercise and 3.7 ± 0.9 after 20 minutes; both
of which fall in the descriptive range of “moderate” to “somewhat strong” exertion. Systolic
blood pressure was not significantly different post exercise than at baseline (baseline: 130 ± 19
mmHg, post: 135 ± 13 mmHg, p=0.509), while diastolic blood pressure showed a trend
(baseline: 78 ± 8 mmHg, post: 82 ± 8 mmHg, p=0.083). Prior to the post-exercise MRI, HR was
not significantly different versus baseline (baseline: 64 ± 6 bpm, post: 66 ± 4 bpm, p=0.267). The
mean time between the end of exercise and the post-exercise rs-fMRI scan was 42:00 ± 6:06
min:s.
39
For the moderate intensity exercise session, the average HR was within 5% of the target. The
average RPE was 4.1 ± 1.0 after 10 minutes of exercise and 4.4 ± 1.4 after 20 minutes; both of
which fall in the descriptive range of “somewhat strong” to “strong” exertion. Systolic blood
pressure was significantly different immediately after exercise than at baseline (baseline: 131 ±
11 mmHg, post: 141 ± 16 mmHg, p=0.028), as was diastolic blood pressure (baseline: 77 ± 7
mmHg, post: 81 ± 5 mmHg, p=0.013). HR prior to post-exercise MRI was significantly different
versus baseline (baseline: 65 ± 7 bpm, post: 70 ± 4 bpm, p=0.007). The mean time between the
end of exercise and the post-exercise rs-fMRI scan was 41:42 ± 5:54 min:s.
VISTAR pipeline: Similar to the previous chapter, structural images were processed using FSL’s
BET (Smith, 2002), and visually inspected for optimal extraction, while functional images were
analysed using MELODIC (Beckmann et al., 2005). Automatic dimensionality estimation (Lapl-
Figure 3.3: T1-weighted or FLAIR images highlighting the location of stroke lesion for each of the
participants included in the preliminary analysis for this chapter.
40
-ace) was used for all MELODIC analysis. Subject-level MELODIC analysis was used to
removes ICs of non-interest. Preliminary analysis focused on: 1) obtaining RSNs consistent with
the previous study, 2) investigating the effect of exercise and 3) investigating the effect of fitness
on the activity of these RSNs.
To test hypotheses 1 and 2, a VISTAR pipeline with 40 datasets was used: both pre- and post-
exercise data from both low (pre50, post50) and moderate (pre70, post70) intensity sessions from 10
subjects (2×2×10). Rs-fMRI Data were pre-processed and de-noised under the same criteria used
in Chapter 2. Group-level MELODIC analysis identified eight RSNs spatially similar to those in
Chapter 2: Sensorimotor, Auditory, DMN, ppCG, Medial Visual, Attention, Executive and Basal
Ganglia (see Figure 3.4). These results provide support for hypothesis 1.
Figure 3.4: Group ICA maps generated using 40 datasets: using 10 subjects both pre- and post-exercise and both
50 % and 70% sessions: (a) Sensorimotor [MNI: 0,-28,50]; (b) Auditory [MNI: -40,-14,4]; (c) DMN [-2,-62,16]; (d)
ppCG [44,-12,26]; (e) Medial Visual [6,-72,0]; (f) Attention [44,18,0]; (g) Executive [-2,22,40]; (h) Basal Ganglia
[-14,-6,4]
41
Table 3.2: Details of VISTAR IC removal
Characteristics Pre-Exercise Post-exercise
Total number of ICs 1224 1279
Total number of artefact ICs removed 423 423
Mean number of ICs per participant per session (SD) 61 ± 8 64 ± 9
Mean number of non-artefact ICs per participant per
session (SD)
40 ± 6 43 ± 8
Mean number of artefact ICs per participant per session
(SD)
21 ± 6 21 ± 3
Min/Max % of removed artefact ICs per session 22/47 25/42
VISTAR removed 34% of a total of 2503 ICs (see Table 3.2). There was a significant difference
in the number of ICs produced using single-session ICA pre- vs. post-exercise (p=0.008). There
was no significant difference, however, in the number of ICs removed pre- vs. post-exercise
(p=1.000), or in the number of ICs identified as non-artefact (p =0.067).
To explore hypothesis 2, dual regression and permutation testing were used (5000 permutations,
threshold-free cluster enhancement), similar to Chapter 2, to conduct a voxel-wise analysis to
investigate a session effect that was consistent for both low and moderate sessions. Four RSNs
showed an effect (see Figure 3.5): the Auditory, DMN, ppCG and Attention RSNs showed a
significant increase in co-activation after exercise (pcorrected<0.05). Cohen’s D values were
calculated pre- vs post-exercise. For analysis, these values were grouped three different ways,
low, moderate, and low and moderate session data, and tested for non-zero significance (see
Table 3.2). This stratification of the data allowed me to investigate for a dose effect.
Additionally, a four-way repeated measures ANOVA (5000 permutations, threshold-free cluster
enhancement) was used to test for a difference between the four sessions.
To test hypothesis 3, 20 de-noised datasets were used: pre50 and pre70 datasets 10 subjects
(1×2×10). Group-level MELODIC analysis identified all six RSNs of interest: Sensorimotor,
Auditory, DMN, ppCG, Attention and Basal Ganglia (see Figure 3.6).
42
Figure 3.5: RSN masks overlaid on an MNI brain template with raw pipeline results showing voxels of increased
co-activation post-exercise: (a) Auditory [MNI: 50,-32,2]; (b) DMN [2,-66,50] (c) ppCG [-58,-18,36]; (d) Attention
[-42,70,0]
43
Table 3.3: Cohen's D effect size comparing pre- vs. post-exercise for each exercise intensity, and
for both intensities, combined
RSN
Low and Moderate
intensity (N=40)
Low intensity
(N=20)
Moderate intensity
(N=20)
mean ± SD p-value mean ± SD p-value mean ± SD p-value
Sensorimotor 0.05 ± 0.25 0.430 -0.01 ± 0.20 0.836 0.10 ± 0.29 0.288
Auditory 0.19 ± 0.36 0.032* 0.25 ± 0.45 0.113 0.13 ± 0.26 0.159
DMN 0.12 ± 0.25 0.043* 0.10 ± 0.26 0.229 0.14 ± 0.26 0.122
ppCG 0.17 ± 0.25 0.007* 0.20 ± 0.17 0.005* 0.14 ± 0.32 0.204
Medial Visual 0.09 ± 0.27 0.142 0.13 ± 0.30 0.220 0.06 ± 0.25 0.465
Attention 0.23 ± 0.28 0.001* 0.34 ± 0.22 0.001* 0.12 ± 0.30 0.223
Executive 0.02 ± 0.16 0.492 0.03 ± 0.17 0.608 0.02 ± 0.15 0.679
Basal Ganglia 0.10 ± 0.28 0.132 0.23 ± 0.19 0.004* -0.04 ± 0.29 0.709
* => p < 0.05
Figure 3.6: Group ICA maps generated using 20 datasets: using 10 subjects pre-exercise and both 50% and 70%
sessions: (a) Sensorimotor [MNI: 2, -14, 54]; (b) Auditory [MNI: -40, -18, 6]; (c) DMN [-2, -64, 18]; (d) ppCG
[50, -20, 36]; (e) Attention [-38, 18, 36]; (f) Basal Ganglia [32, 2, -2].
44
To investigate the effect of fitness on these RSNs, age and sex corrected VO2peak values, or
fractional VO2 (fVO2), were used as a variable of interest in a regression analysis. I used a
conjugate analysis to establish voxels that showed a repeated effect for each of the individual
sessions. First, dual regression was used to develop subject/session specific RSNs, similar to
Chapter 2. Next, for the pre50 images of a given RSN, voxel-wise activation was regressed
against fVO2 and the resulting image was thresholded (p<0.05). This was repeated for the pre70
images, and the two resulting images were overlapped, preserving only voxels that were p<0.05
significant in both cases. This is repeated for all six RSNs, twice each, for voxels positively and
negatively correlated with fitness (see Figure 3.7).
3.1.3 Discussion and Future Work
The preliminary results from minor stroke participants in this Chapter are promising. Firstly,
group ICA produced all eight RSNs that hypothesis 1 predicted. Three of these networks,
however, displayed slightly different spatial patterns to those from Chapter 2: the sensorimotor
RSN is smaller and the attention RSN is smaller and contains predominantly anterior areas. This
may be influenced by the relatively small sample size, or post-stroke reorganization of these
networks (Corbetta and Shulman, 2011; Grefkes and Fink, 2011; Park et al., 2011). Secondly, a
VISTAR analysis uncovered an effect of session in four networks. The Auditory RSN showed a
significant bi-lateralized increase in activation post-exercise in brain regions, central and parietal
operculum cortices, similar to those found using VISTAR on the exercise data in Chapter 2. The
area of increase activity in the ppCG is left lateralized and lies in the part of the sensory
homunculus. This area is associated with processing sensation from the face, and it is unclear
why such an area would show an up-regulation in activation after exercise. Increased activation
in a sensorimotor related RSN, however, is not surprising. Additionally, the centre of this
increased activation (MNI [-58, -18, 36]) is close to that of the trend shown by the ppCG in
Chapter 2 (MNI [-52, -22, 34]).
Furthermore, two cognition-related RSNs showed increased activity post-exercise. The DMN
showed increased activity in the precuneous cortex, while the Attention RSN showed a bi-lateral
increase in the insular cortex and the frontal operculum cortex. The precuneous cortex is known
to be involved in self-awareness and self-consciousness, episodic memory and visuospatial
imagery. The insula is involved in regulating autonomic function, including heart rate and blood
45
pressure, as well as motor learning and salience. These results are distinct from those in Chapter
2, since that study found no effect of exercise on RSNs involved in cognitive processing.
The Cohen’s D analysis to characterize an entire RSN produced results consistent with the voxel-
wise analysis. This RSN-based (ROI) analysis revealed that the Auditory, DMN, ppCG and
Attention networks all showed a significant effect of session. Moreover, when the Cohen's D
Figure 3.7: Results of the regression analysis to identify brain regions that are associated with aerobic fitness
level (fVO2). Brain regions in green represent areas where there is a positive or negative association with fVO2.
Voxels were considered significant if they reached p<0.05 for both the pre50 and pre70 sessions (i.e. conjunction
statistic equivalent to a raw p-value=0.0025). (a) Sensorimotor [MNI: 16, 2, 64], [0, -6, 52]; (b) Auditory [MNI:
-54, -24, 16], [-52, -28, 2]; (c) DMN [-6, -74, 28], [10,-64, 36]; (d) ppCG [-56, 0, 26], [52 , -16, 38]; (e) Attention
[MNI: -32, -10, 8], [-46, 4, 8]; (f) Basal Ganglia [-22, 4, -8], [-22, 4, -8].
46
values were stratified by dose, a different story emerged. The ppCG, Attention and Basal
Ganglia RSNs showed an effect of session for the 50% session, while no RSNs showed an effect
for the 70% session. This may suggest an effect of dose on the Cohen's D values. It should be
noted, however, that ten Cohen's D values were used per RSN for these dose-effect analyses.
An ANOVA found no voxel-wise session by intensity interaction between the four (pre50, post50,
pre70, post70) datasets when corrected for multiple comparisons. The lack of findings on this
ANOVA test may reflect the stringent correction for multiple comparisons, since the uncorrected
F-statistic map did identify regions with highly significant uncorrected p-values (i.e. p<0.0001).
The replication of the result found in the Auditory RSN from Chapter 2 further strengthens the
notion that S2 is up-regulated after exercise. Additionally, increased activity in the post-central
gyrus, albeit a different area, is further evidence that sensorimotor cortical areas show increased
rs-fc after exercise. The discovery of a session effect in cognition-related RSNs provides further
evidence that exercise has an effect on cognitive networks.
The regression analysis that used aerobic fitness level on RSN activity is unique relative to the
current neuroimaging and exercise literature. A conjunction statistic was used for this
exploratory analysis. In this case, a significant result is reached when a correlation with fitness is
achieved for both pre50 and pre70 datasets. This technique produced mixed results. All six RSNs
studies showed an effect of fitness on their activity. The Sensorimotor, DMN, ppCG and
Attention RSNs showed areas, non-overlapping, whose activity were positively correlated or
negatively correlated with fitness. The Auditory and Basal Ganglia RSNs showed areas of
activity positively correlated with fitness. Higher baseline co-activation in an RSN may be an
indication that the RSN requires more cortical resources to carry out its baseline functions. One
would suspect that that would be more apparent in less fit individuals. The paucity of literature
on the effects of fitness on rs-fc makes it difficult to interpret these results. Nonetheless, this is
interesting result that could be further investigated with a larger sample size.
It should be noted that, for contrast, analysis on non-denoised data was also performed. This
analysis produced eight group level ICA maps, similar to the VISTAR analysis, while a voxel-
wise analysis produced similar, albeit weaker, results in the Auditory and ppCG RSNs. Cohen's
D analysis was inconsistent with voxel-wise analysis. That the Raw pipeline was able to detect
an exercise-related session effect may suggest that exercise has a stronger effect on the stroke
47
population. Alternatively, it may be due to a reduced level of structural noise present in the data.
Thirdly, the NUES Raw pipeline used 40 datasets, as opposed to the 30 used in Chapter 2,
producing better statistical power. This may further explain its superior sensitivity to detecting an
effect of session on resting state connectivity.
While VISTAR has proven to be successful at improving the sensitivity of session-effect
analyses, its prohibitively time consuming nature makes it impractical for large datasets. De-
noising a single dataset takes roughly 30 minutes. However, de-noising rs-fMRI data using ICA
is not new, and fortunately, researchers at FMRIB, where MELODIC was developed, have
recently published on an ICA-based Xnoiseifier (FIX) supervised-learning classification
algorithm that may contend with my VISTAR approach (Salimi-Khorshidi et al., 2014). FIX is
similar to VISTAR in the sense that manual inspection is required on single subject ICA results
to identify good vs. bad components, albeit on a subset of the entire rs-BOLD dataset. FIX then
uses a classifier to work on the rest of the data. This new procedure is more automated and time
efficient compared to VISTAR. FIX differs from VISTAR, however, in that classification is
performed on data that has not been pre-processed. While this may be a 'purer' form of de-
noising, since no data has been removed by filtering or smoothing, it is inherently more difficult
since it makes ICnoise maps more difficult to identify. Whilst implementation of the beta FIX
software and comparisons between FIX and VISTAR are beyond the scope of the thesis, future
work will involve applying a FIX to the stroke data. FIX may further improve the sensitivity of
the analysis, bolstering the current results, and also uncovering other RSNs that exhibit a session
effect.
Future work will also include additional recruitment to reach a target of 15 completed stroke
participant datasets. I believe this will be sufficient to investigate the effects of dose and fitness
on rs-fc and the exercise session effect since our sample size will be comparable to the cohorts
used in Chapter 2. Fifteen datasets would also allow for 5000 permutations when correcting for
multiple comparisons, which is the standard in the resting-state fMRI literature (Filippini et al.,
2009; Schwindt et al., 2013). A larger sample size would also allow a comparison between
Randomise and conjunction analysis approaches, as well as a direct comparison between the
young healthy cohort and the stroke cohort.
R L
R L
48
Furthermore, additional datasets would allow me to better investigate the intersession reliability
of the resting state data. Previous work (Zuo et al., 2010) has used dual regression and template
matching to analyze session reliability in healthy adults. This approach can be extended to the
stroke patient data. Raw, VISTAR and FIX pipelines could be used to contrast the effects of
denoising on intersession reliability.
Lastly, lesion location and seed based analysis may be used to probe lesion-location effects on
RSN activity and the inter-hemispheric connectivity effects of exercise, respectively. A recent
study demonstrated there is an effect of lesion location on RSN maps (Ovadia-Caro et al., 2013),
and another study used SBCA to uncover inter-hemispheric motor-RSN connectivity differences
in a stroke population (Chen and Schlaug, 2013). Lesion location may influence exercise-related
connectivity changes. Furthermore, I have shown that the effect of exercise in the stroke cohort
appears to be left-lateralized in a sensorimotor RSN. This may be an indication of inter-
hemispheric connectivity after exercise.
3.2 Conclusion
This thesis develops the use of resting state fMRI as a means to characterize session related
changes in functional connectivity in grey matter regions in the brain. Rs-fMRI data are
inherently complex, given the multiple voxels and time series data present in a single 4D image.
Including a visual inspection step in the single-subject ICA procedure (i.e. VISTAR)
significantly improved the sensitivity of my analysis by effectively reducing the amount of
spurious signals in the rs-fMRI data, a common problem in rs-fMRI analysis. In Chapter 2, I
observed that this procedure resulted in the detection of brain regions that showed a significant
session-effect, using both voxel-wise and region of interest analyses. Chapter 3 extended this
research on functional connectivity by exploratory associations of fitness on rs-fc in a smaller
cohort of chronic stroke participants. Preliminary results in Chapter 3 help to support the
findings from Chapter 2. Additional analysis procedures were developed in Chapter 3 to help
advance this field of research.
This thesis helps to develop the use of rs-fc as a tool to study the dynamic changes in the brain’s
connections that are related to environmental / physiological changes. Acute exercise leads to not
49
only physiological changes that are part of the body’s recovery from the cardiovascular
challenge or stimuli, but also alterations in the brains functional connectivity. Previous work has
shown that, after exercise, blood pressure may be reduced for several hours among hypertensive
adults (MacDonald et al., 2001), and that brain function may be altered due to an acute exercise
carry-over effect by ‘thinking about’ the exercise experiences (MacIntosh et al., 2014). This
research may lead to improving our understanding of the neurobiological effects of exercise and
may eventually be useful to clinicians and rehabilitation specialists that use aerobic exercise as
therapy to help survivors in their recovery after stroke.
Many of the analysis tools used in this thesis are consistent with those used by the Human
Connectome Project (Van Essen et al., 2013), including MELODIC, Dual Regression and
Randomize. FIX is also being developed to denoise BOLD-fMRI data in the Human
Connectome Project. This initiative is massive in scale, which includes resting state and task
fMRI, behavioral, genetic and environmental data being collected on 1200 healthy adults,
including twin pairs and their families (amount to over 10GB of data per subject). Huge data
collection and data sharing initiatives like this are proof of the massive potential of fMRI to map
the brains structural and functional connectivity, as well as the collaborative effort needed to do
so. The Human Connectome Project seeks to uncover the inter-individual variability in rs-fc, and
will also help to standardize analysis techniques for rs- and task-fMRI (Marcus et al., 2011).
In conclusion, BOLD-fMRI has proven to be a rapidly maturing field over the last decade, being
used to probe cognitive, behavioral, pharmacological, psychological, environmental, and now
exercise, brain-effects, and has as long way to go still. Advancements are continually being made
in the acquisition and analysis of MRI data. The brain is an inherently inter-connected organ,
relying on functional connectivity to carry out its tasks. One of the ultimate goals of brain
research is to help ameliorate our understanding of its behavior, and ultimately, its ability to
change.
50
4 Reference List
Alexander, G.E., Crutcher, M.D., DeLong, M.R., 1990. Basal ganglia-thalamocortical circuits:
parallel substrates for motor, oculomotor, "prefrontal" and "limbic" functions. Prog.Brain
Res. 85, 119-146.
Amedi, A., Floel, A., Knecht, S., Zohary, E., Cohen, L.G., 2004. Transcranial magnetic
stimulation of the occipital pole interferes with verbal processing in blind subjects.
Nat.Neurosci. 7, 1266-1270.
Amedi, A., Raz, N., Pianka, P., Malach, R., Zohary, E., 2003. Early 'visual' cortex activation
correlates with superior verbal memory performance in the blind. Nat.Neurosci. 6, 758-
766.
Astrand, K., Rodahl, K., Dahl, H., Stromme, S., 2003. Textbook of Work Physiology-4th:
Physiological Bases of Exercise.
Beckmann, C.F., DeLuca, M., Devlin, J.T., Smith, S.M., 2005. Investigations into resting-state
connectivity using independent component analysis. Philos.Trans.R.Soc.Lond B Biol.Sci.
360, 1001-1013.
Bedny, M., Pascual-Leone, A., Dodell-Feder, D., Fedorenko, E., Saxe, R., 2011. Language
processing in the occipital cortex of congenitally blind adults. Proc.Natl.Acad.Sci.U.S.A
108, 4429-4434.
Biswal, B., Yetkin, F.Z., Haughton, V.M., Hyde, J.S., 1995. Functional connectivity in the motor
cortex of resting human brain using echo-planar MRI. Magn Reson.Med. 34, 537-541.
Biswal, B.B., Mennes, M., Zuo, X.N., Gohel, S., Kelly, C., Smith, S.M., Beckmann, C.F.,
Adelstein, J.S., Buckner, R.L., Colcombe, S., Dogonowski, A.M., Ernst, M., Fair, D.,
Hampson, M., Hoptman, M.J., Hyde, J.S., Kiviniemi, V.J., Kotter, R., Li, S.J., Lin, C.P.,
Lowe, M.J., Mackay, C., Madden, D.J., Madsen, K.H., Margulies, D.S., Mayberg, H.S.,
McMahon, K., Monk, C.S., Mostofsky, S.H., Nagel, B.J., Pekar, J.J., Peltier, S.J.,
Petersen, S.E., Riedl, V., Rombouts, S.A., Rypma, B., Schlaggar, B.L., Schmidt, S.,
Seidler, R.D., Siegle, G.J., Sorg, C., Teng, G.J., Veijola, J., Villringer, A., Walter, M.,
Wang, L., Weng, X.C., Whitfield-Gabrieli, S., Williamson, P., Windischberger, C., Zang,
Y.F., Zhang, H.Y., Castellanos, F.X., Milham, M.P., 2010. Toward discovery science of
human brain function. Proc.Natl.Acad.Sci.U.S.A 107, 4734-4739.
Black, J.E., Isaacs, K.R., Anderson, B.J., Alcantara, A.A., Greenough, W.T., 1990. Learning
causes synaptogenesis, whereas motor activity causes angiogenesis, in cerebellar cortex
of adult rats. Proc.Natl.Acad.Sci.U.S.A 87, 5568-5572.
Brookes, M.J., Woolrich, M., Luckhoo, H., Price, D., Hale, J.R., Stephenson, M.C., Barnes,
G.R., Smith, S.M., Morris, P.G., 2011. Investigating the electrophysiological basis of
51
resting state networks using magnetoencephalography. Proc.Natl.Acad.Sci.U.S.A 108,
16783-16788.
Burton, H., Abend, N.S., MacLeod, A.M., Sinclair, R.J., Snyder, A.Z., Raichle, M.E., 1999.
Tactile attention tasks enhance activation in somatosensory regions of parietal cortex: a
positron emission tomography study. Cereb.Cortex 9, 662-674.
Byberg, L., Melhus, H., Gedeborg, R., Sundstrom, J., Ahlbom, A., Zethelius, B., Berglund, L.G.,
Wolk, A., Michaelsson, K., 2009. Total mortality after changes in leisure time physical
activity in 50 year old men: 35 year follow-up of population based cohort. BMJ 338,
b688.
Carter, A.R., Astafiev, S.V., Lang, C.E., Connor, L.T., Rengachary, J., Strube, M.J., Pope, D.L.,
Shulman, G.L., Corbetta, M., 2010. Resting interhemispheric functional magnetic
resonance imaging connectivity predicts performance after stroke. Ann.Neurol. 67, 365-
375.
Caspersen, C.J., Powell, K.E., Christenson, G.M., 1985. Physical activity, exercise, and physical
fitness: definitions and distinctions for health-related research. Public Health Rep. 100,
126-131.
Chang, Y.K., Labban, J.D., Gapin, J.I., Etnier, J.L., 2012. The effects of acute exercise on
cognitive performance: a meta-analysis. Brain Res. 1453, 87-101.
Chen, J.L., Schlaug, G., 2013. Resting state interhemispheric motor connectivity and white
matter integrity correlate with motor impairment in chronic stroke. Front Neurol. 4, 178.
Christensen, L.O., Johannsen, P., Sinkjaer, T., Petersen, N., Pyndt, H.S., Nielsen, J.B., 2000.
Cerebral activation during bicycle movements in man. Exp.Brain Res. 135, 66-72.
Cohen, D., 1972. Magnetoencephalography: detection of the brain's electrical activity with a
superconducting magnetometer. Science 175, 664-666.
Colcombe, S., Kramer, A.F., 2003. Fitness effects on the cognitive function of older adults: a
meta-analytic study. Psychol.Sci. 14, 125-130.
Colcombe, S.J., Erickson, K.I., Scalf, P.E., Kim, J.S., Prakash, R., McAuley, E., Elavsky, S.,
Marquez, D.X., Hu, L., Kramer, A.F., 2006. Aerobic exercise training increases brain
volume in aging humans. J Gerontol.A Biol.Sci.Med.Sci. 61, 1166-1170.
Colcombe, S.J., Kramer, A.F., Erickson, K.I., Scalf, P., McAuley, E., Cohen, N.J., Webb, A.,
Jerome, G.J., Marquez, D.X., Elavsky, S., 2004. Cardiovascular fitness, cortical
plasticity, and aging. Proc.Natl.Acad.Sci.U.S.A 101, 3316-3321.
Corbetta, M., Shulman, G.L., 2011. Spatial neglect and attention networks. Annu.Rev.Neurosci.
34, 569-599.
52
Damoiseaux, J.S., Rombouts, S.A., Barkhof, F., Scheltens, P., Stam, C.J., Smith, S.M.,
Beckmann, C.F., 2006. Consistent resting-state networks across healthy subjects.
Proc.Natl.Acad.Sci.U.S.A 103, 13848-13853.
Erickson, K.I., Voss, M.W., Prakash, R.S., Basak, C., Szabo, A., Chaddock, L., Kim, J.S., Heo,
S., Alves, H., White, S.M., Wojcicki, T.R., Mailey, E., Vieira, V.J., Martin, S.A., Pence,
B.D., Woods, J.A., McAuley, E., Kramer, A.F., 2011. Exercise training increases size of
hippocampus and improves memory. Proc.Natl.Acad.Sci.U.S.A 108, 3017-3022.
Filippini, N., MacIntosh, B.J., Hough, M.G., Goodwin, G.M., Frisoni, G.B., Smith, S.M.,
Matthews, P.M., Beckmann, C.F., Mackay, C.E., 2009. Distinct patterns of brain activity
in young carriers of the APOE-epsilon4 allele. Proc.Natl.Acad.Sci.U.S.A 106, 7209-
7214.
Fletcher, G.F., Balady, G., Blair, S.N., Blumenthal, J., Caspersen, C., Chaitman, B., Epstein, S.,
Sivarajan Froelicher, E.S., Froelicher, V.F., Pina, I.L., Pollock, M.L., 1996. Statement on
exercise: benefits and recommendations for physical activity programs for all Americans.
A statement for health professionals by the Committee on Exercise and Cardiac
Rehabilitation of the Council on Clinical Cardiology, American Heart Association.
Circulation 94, 857-862.
Fong, D.Y., Ho, J.W., Hui, B.P., Lee, A.M., Macfarlane, D.J., Leung, S.S., Cerin, E., Chan,
W.Y., Leung, I.P., Lam, S.H., Taylor, A.J., Cheng, K.K., 2012. Physical activity for
cancer survivors: meta-analysis of randomised controlled trials. BMJ 344, e70.
Fox, M.D., Greicius, M., 2010. Clinical applications of resting state functional connectivity.
Front Syst.Neurosci. 4, 19.
Fukunaga, M., Horovitz, S.G., van Gelderen, P., de Zwart, J.A., Jansma, J.M., Ikonomidou,
V.N., Chu, R., Deckers, R.H., Leopold, D.A., Duyn, J.H., 2006. Large-amplitude,
spatially correlated fluctuations in BOLD fMRI signals during extended rest and early
sleep stages. Magn Reson.Imaging 24, 979-992.
Garcia-Aymerich, J., Lange, P., Benet, M., Schnohr, P., Anto, J.M., 2006. Regular physical
activity reduces hospital admission and mortality in chronic obstructive pulmonary
disease: a population based cohort study. Thorax 61, 772-778.
Glahn, D.C., Winkler, A.M., Kochunov, P., Almasy, L., Duggirala, R., Carless, M.A., Curran,
J.C., Olvera, R.L., Laird, A.R., Smith, S.M., Beckmann, C.F., Fox, P.T., Blangero, J.,
2010. Genetic control over the resting brain. Proc.Natl.Acad.Sci.U.S.A 107, 1223-1228.
Grefkes, C., Fink, G.R., 2011. Reorganization of cerebral networks after stroke: new insights
from neuroimaging with connectivity approaches. Brain 134, 1264-1276.
Greicius, M.D., Srivastava, G., Reiss, A.L., Menon, V., 2004. Default-mode network activity
distinguishes Alzheimer's disease from healthy aging: evidence from functional MRI.
Proc.Natl.Acad.Sci.U.S.A 101, 4637-4642.
53
Hillman, C.H., Erickson, K.I., Kramer, A.F., 2008. Be smart, exercise your heart: exercise
effects on brain and cognition. Nat.Rev.Neurosci. 9, 58-65.
Jack, C.R., Jr., Wiste, H.J., Vemuri, P., Weigand, S.D., Senjem, M.L., Zeng, G., Bernstein, M.A.,
Gunter, J.L., Pankratz, V.S., Aisen, P.S., Weiner, M.W., Petersen, R.C., Shaw, L.M.,
Trojanowski, J.Q., Knopman, D.S., 2010. Brain beta-amyloid measures and magnetic
resonance imaging atrophy both predict time-to-progression from mild cognitive
impairment to Alzheimer's disease. Brain 133, 3336-3348.
Jenkinson, M., Beckmann, C., 2001. TR01MJ2 : ICA in FMRI: A Basic Introduction.
Jenkinson, M., Beckmann, C.F., Behrens, T.E., Woolrich, M.W., Smith, S.M., 2012. FSL.
Neuroimage. 62, 782-790.
Jezzard, P., Matthews, P.M., Smith, S.M., 2001. Functional MRI: An Introduction to Methods,
First Edition ed. Oxford University Press.
Kamijo, K., Hayashi, Y., Sakai, T., Yahiro, T., Tanaka, K., Nishihira, Y., 2009. Acute effects of
aerobic exercise on cognitive function in older adults. J Gerontol.B Psychol.Sci.Soc.Sci.
64, 356-363.
Kamijo, K., Nishihira, Y., Hatta, A., Kaneda, T., Wasaka, T., Kida, T., Kuroiwa, K., 2004.
Differential influences of exercise intensity on information processing in the central
nervous system. Eur.J Appl.Physiol 92, 305-311.
Kamijo, K., Nishihira, Y., Higashiura, T., Kuroiwa, K., 2007. The interactive effect of exercise
intensity and task difficulty on human cognitive processing. Int.J Psychophysiol. 65, 114-
121.
Katzmarzyk, P.T., Gledhill, N., Shephard, R.J., 2000. The economic burden of physical
inactivity in Canada. CMAJ. 163, 1435-1440.
Kelly, R.E., Jr., Alexopoulos, G.S., Wang, Z., Gunning, F.M., Murphy, C.F., Morimoto, S.S.,
Kanellopoulos, D., Jia, Z., Lim, K.O., Hoptman, M.J., 2010. Visual inspection of
independent components: defining a procedure for artifact removal from fMRI data.
J.Neurosci.Methods 189, 233-245.
Khalili-Mahani, N., Zoethout, R.M., Beckmann, C.F., Baerends, E., de Kam, M.L., Soeter, R.P.,
Dahan, A., van Buchem, M.A., van Gerven, J.M., Rombouts, S.A., 2012. Effects of
morphine and alcohol on functional brain connectivity during "resting state": a placebo-
controlled crossover study in healthy young men. Hum.Brain Mapp. 33, 1003-1018.
Kleim, J.A., Cooper, N.R., VandenBerg, P.M., 2002. Exercise induces angiogenesis but does not
alter movement representations within rat motor cortex. Brain Res. 934, 1-6.
Knols, R., Aaronson, N.K., Uebelhart, D., Fransen, J., Aufdemkampe, G., 2005. Physical
exercise in cancer patients during and after medical treatment: a systematic review of
randomized and controlled clinical trials. J Clin.Oncol. 23, 3830-3842.
54
Lachman, M.E., Neupert, S.D., Bertrand, R., Jette, A.M., 2006. The effects of strength training
on memory in older adults. J Aging Phys.Act. 14, 59-73.
Lee, C.D., Folsom, A.R., Blair, S.N., 2003. Physical activity and stroke risk: a meta-analysis.
Stroke 34, 2475-2481.
Lim, S.S., Vos, T., Flaxman, A.D., Danaei, G., Shibuya, K., Adair-Rohani, H., Amann, M.,
Anderson, H.R., Andrews, K.G., Aryee, M., Atkinson, C., Bacchus, L.J., Bahalim, A.N.,
Balakrishnan, K., Balmes, J., Barker-Collo, S., Baxter, A., Bell, M.L., Blore, J.D., Blyth,
F., Bonner, C., Borges, G., Bourne, R., Boussinesq, M., Brauer, M., Brooks, P., Bruce,
N.G., Brunekreef, B., Bryan-Hancock, C., Bucello, C., Buchbinder, R., Bull, F., Burnett,
R.T., Byers, T.E., Calabria, B., Carapetis, J., Carnahan, E., Chafe, Z., Charlson, F., Chen,
H., Chen, J.S., Cheng, A.T., Child, J.C., Cohen, A., Colson, K.E., Cowie, B.C., Darby,
S., Darling, S., Davis, A., Degenhardt, L., Dentener, F., Des Jarlais, D.C., Devries, K.,
Dherani, M., Ding, E.L., Dorsey, E.R., Driscoll, T., Edmond, K., Ali, S.E., Engell, R.E.,
Erwin, P.J., Fahimi, S., Falder, G., Farzadfar, F., Ferrari, A., Finucane, M.M., Flaxman,
S., Fowkes, F.G., Freedman, G., Freeman, M.K., Gakidou, E., Ghosh, S., Giovannucci,
E., Gmel, G., Graham, K., Grainger, R., Grant, B., Gunnell, D., Gutierrez, H.R., Hall, W.,
Hoek, H.W., Hogan, A., Hosgood, H.D., III, Hoy, D., Hu, H., Hubbell, B.J., Hutchings,
S.J., Ibeanusi, S.E., Jacklyn, G.L., Jasrasaria, R., Jonas, J.B., Kan, H., Kanis, J.A.,
Kassebaum, N., Kawakami, N., Khang, Y.H., Khatibzadeh, S., Khoo, J.P., Kok, C.,
Laden, F., Lalloo, R., Lan, Q., Lathlean, T., Leasher, J.L., Leigh, J., Li, Y., Lin, J.K.,
Lipshultz, S.E., London, S., Lozano, R., Lu, Y., Mak, J., Malekzadeh, R., Mallinger, L.,
Marcenes, W., March, L., Marks, R., Martin, R., McGale, P., McGrath, J., Mehta, S.,
Mensah, G.A., Merriman, T.R., Micha, R., Michaud, C., Mishra, V., Mohd, H.K.,
Mokdad, A.A., Morawska, L., Mozaffarian, D., Murphy, T., Naghavi, M., Neal, B.,
Nelson, P.K., Nolla, J.M., Norman, R., Olives, C., Omer, S.B., Orchard, J., Osborne, R.,
Ostro, B., Page, A., Pandey, K.D., Parry, C.D., Passmore, E., Patra, J., Pearce, N.,
Pelizzari, P.M., Petzold, M., Phillips, M.R., Pope, D., Pope, C.A., III, Powles, J., Rao,
M., Razavi, H., Rehfuess, E.A., Rehm, J.T., Ritz, B., Rivara, F.P., Roberts, T., Robinson,
C., Rodriguez-Portales, J.A., Romieu, I., Room, R., Rosenfeld, L.C., Roy, A., Rushton,
L., Salomon, J.A., Sampson, U., Sanchez-Riera, L., Sanman, E., Sapkota, A., Seedat, S.,
Shi, P., Shield, K., Shivakoti, R., Singh, G.M., Sleet, D.A., Smith, E., Smith, K.R.,
Stapelberg, N.J., Steenland, K., Stockl, H., Stovner, L.J., Straif, K., Straney, L., Thurston,
G.D., Tran, J.H., Van, D.R., van, D.A., Veerman, J.L., Vijayakumar, L., Weintraub, R.,
Weissman, M.M., White, R.A., Whiteford, H., Wiersma, S.T., Wilkinson, J.D., Williams,
H.C., Williams, W., Wilson, N., Woolf, A.D., Yip, P., Zielinski, J.M., Lopez, A.D.,
Murray, C.J., Ezzati, M., AlMazroa, M.A., Memish, Z.A., 2012. A comparative risk
assessment of burden of disease and injury attributable to 67 risk factors and risk factor
clusters in 21 regions, 1990-2010: a systematic analysis for the Global Burden of Disease
Study 2010. Lancet 380, 2224-2260.
Londeree, B.R., Moeschberger, M.L., 1984. Influence of age and other factors on maximal heart
rate. J Cardiac Rehabil 4, 44-49.
MacDonald, J.R., Hogben, C.D., Tarnopolsky, M.A., MacDougall, J.D., 2001. Post exercise
hypotension is sustained during subsequent bouts of mild exercise and simulated
activities of daily living. J Hum.Hypertens. 15, 567-571.
55
MacIntosh, B.J., Crane, D.E., Sage, M.D., Rajab, A.S., Donahue, M.J., McIlroy, W.E.,
Middleton, L.E., 2014. Impact of a single bout of aerobic exercise on regional brain
perfusion and activation responses in healthy young adults. PLoS.One. 9, e85163.
Marcus, D.S., Harwell, J., Olsen, T., Hodge, M., Glasser, M.F., Prior, F., Jenkinson, M.,
Laumann, T., Curtiss, S.W., Van Essen, D.C., 2011. Informatics and data mining tools
and strategies for the human connectome project. Front Neuroinform. 5, 4.
Merabet, L.B., Hamilton, R., Schlaug, G., Swisher, J.D., Kiriakopoulos, E.T., Pitskel, N.B.,
Kauffman, T., Pascual-Leone, A., 2008. Rapid and reversible recruitment of early visual
cortex for touch. PLoS.One. 3, e3046.
Moraes, H., Ferreira, C., Deslandes, A., Cagy, M., Pompeu, F., Ribeiro, P., Piedade, R., 2007.
Beta and alpha electroencephalographic activity changes after acute exercise. Arq
Neuropsiquiatr. 65, 637-641.
Murphy, K., Birn, R.M., Bandettini, P.A., 2013. Resting-state fMRI confounds and cleanup.
Neuroimage. 80, 349-359.
Naci, H., Ioannidis, J.P., 2013. Comparative effectiveness of exercise and drug interventions on
mortality outcomes: metaepidemiological study. BMJ 347, f5577.
Nichols, T.E., Holmes, A.P., 2002. Nonparametric permutation tests for functional
neuroimaging: a primer with examples. Hum.Brain Mapp. 15, 1-25.
Niesters, M., Khalili-Mahani, N., Martini, C., Aarts, L., van, G.J., van Buchem, M.A., Dahan,
A., Rombouts, S., 2012. Effect of subanesthetic ketamine on intrinsic functional brain
connectivity: a placebo-controlled functional magnetic resonance imaging study in
healthy male volunteers. Anesthesiology 117, 868-877.
Ogawa, S., Menon, R.S., Tank, D.W., Kim, S.G., Merkle, H., Ellermann, J.M., Ugurbil, K.,
1993. Functional brain mapping by blood oxygenation level-dependent contrast magnetic
resonance imaging. A comparison of signal characteristics with a biophysical model.
Biophys.J 64, 803-812.
Oldehinkel, M., Francx, W., Beckmann, C.F., Buitelaar, J.K., Mennes, M., 2013. Resting state
FMRI research in child psychiatric disorders. Eur.Child Adolesc.Psychiatry 22, 757-770.
Ovadia-Caro, S., Villringer, K., Fiebach, J., Jungehulsing, G.J., van der Meer, E., Margulies,
D.S., Villringer, A., 2013. Longitudinal effects of lesions on functional networks after
stroke. J Cereb.Blood Flow Metab 33, 1279-1285.
Park, C.H., Chang, W.H., Ohn, S.H., Kim, S.T., Bang, O.Y., Pascual-Leone, A., Kim, Y.H.,
2011. Longitudinal changes of resting-state functional connectivity during motor
recovery after stroke. Stroke 42, 1357-1362.
Pauling, L., Coryell, C.D., 1936. The Magnetic Properties and Structure of Hemoglobin,
Oxyhemoglobin and Carbonmonoxyhemoglobin. Proc.Natl.Acad.Sci.U.S.A 22, 210-216.
56
Pedersen, B.K., Saltin, B., 2006. Evidence for prescribing exercise as therapy in chronic disease.
Scand.J Med.Sci.Sports 16 Suppl 1, 3-63.
Pereira, A.C., Huddleston, D.E., Brickman, A.M., Sosunov, A.A., Hen, R., McKhann, G.M.,
Sloan, R., Gage, F.H., Brown, T.R., Small, S.A., 2007. An in vivo correlate of exercise-
induced neurogenesis in the adult dentate gyrus. Proc.Natl.Acad.Sci.U.S.A 104, 5638-
5643.
Pratt, M., Macera, C.A., Wang, G., 2000. Higher direct medical costs associated with physical
inactivity. Phys.Sportsmed. 28, 63-70.
Raichle, M.E., MacLeod, A.M., Snyder, A.Z., Powers, W.J., Gusnard, D.A., Shulman, G.L.,
2001. A default mode of brain function. Proc.Natl.Acad.Sci.U.S.A 98, 676-682.
Reimers, C.D., Knapp, G., Reimers, A.K., 2009. Exercise as stroke prophylaxis.
Dtsch.Arztebl.Int. 106, 715-721.
Rhyu, I.J., Bytheway, J.A., Kohler, S.J., Lange, H., Lee, K.J., Boklewski, J., McCormick, K.,
Williams, N.I., Stanton, G.B., Greenough, W.T., Cameron, J.L., 2010. Effects of aerobic
exercise training on cognitive function and cortical vascularity in monkeys. Neuroscience
167, 1239-1248.
Roddy, E., Zhang, W., Doherty, M., 2005. Aerobic walking or strengthening exercise for
osteoarthritis of the knee? A systematic review. Ann.Rheum.Dis. 64, 544-548.
Salimi-Khorshidi, G., Douaud, G., Beckmann, C.F., Glasser, M.F., Griffanti, L., Smith, S.M.,
2014. Automatic denoising of functional MRI data: Combining independent component
analysis and hierarchical fusion of classifiers. Neuroimage. 90C, 449-468.
Samitz, G., Egger, M., Zwahlen, M., 2011. Domains of physical activity and all-cause mortality:
systematic review and dose-response meta-analysis of cohort studies. Int.J Epidemiol. 40,
1382-1400.
Schneider, S., Brummer, V., Abel, T., Askew, C.D., Struder, H.K., 2009. Changes in brain
cortical activity measured by EEG are related to individual exercise preferences. Physiol
Behav. 98, 447-452.
Schnitzler, A., Gross, J., 2005. Normal and pathological oscillatory communication in the brain.
Nat.Rev.Neurosci. 6, 285-296.
Schwindt, G.C., Chaudhary, S., Crane, D., Ganda, A., Masellis, M., Grady, C.L., Stefanovic, B.,
Black, S.E., 2013. Modulation of the default-mode network between rest and task in
Alzheimer's Disease. Cereb.Cortex 23, 1685-1694.
Sigal, R.J., Kenny, G.P., Wasserman, D.H., Castaneda-Sceppa, C., White, R.D., 2006. Physical
activity/exercise and type 2 diabetes: a consensus statement from the American Diabetes
Association. Diabetes Care 29, 1433-1438.
Smith, S.M., 2002. Fast robust automated brain extraction. Hum.Brain Mapp. 17, 143-155.
57
Smith, S.M., Fox, P.T., Miller, K.L., Glahn, D.C., Fox, P.M., Mackay, C.E., Filippini, N.,
Watkins, K.E., Toro, R., Laird, A.R., Beckmann, C.F., 2009. Correspondence of the
brain's functional architecture during activation and rest. Proc.Natl.Acad.Sci.U.S.A 106,
13040-13045.
Spirduso, W.W., 1975. Reaction and movement time as a function of age and physical activity
level. J.Gerontol. 30, 435-440.
Swain, R.A., Harris, A.B., Wiener, E.C., Dutka, M.V., Morris, H.D., Theien, B.E., Konda, S.,
Engberg, K., Lauterbur, P.C., Greenough, W.T., 2003. Prolonged exercise induces
angiogenesis and increases cerebral blood volume in primary motor cortex of the rat.
Neuroscience 117, 1037-1046.
Takenobu, Y., Hayashi, T., Moriwaki, H., Nagatsuka, K., Naritomi, H., Fukuyama, H., 2013.
Motor recovery and microstructural change in rubro-spinal tract in subcortical stroke.
Neuroimage.Clin. 4, 201-208.
Tanabe, J., Nyberg, E., Martin, L.F., Martin, J., Cordes, D., Kronberg, E., Tregellas, J.R., 2011.
Nicotine effects on default mode network during resting state. Psychopharmacology
(Berl) 216, 287-295.
Thomas, A.G., Dennis, A., Bandettini, P.A., Johansen-Berg, H., 2012. The effects of aerobic
activity on brain structure. Front Psychol. 3, 86.
Ugurbil, K., Xu, J., Auerbach, E.J., Moeller, S., Vu, A.T., Duarte-Carvajalino, J.M., Lenglet, C.,
Wu, X., Schmitter, S., Van de Moortele, P.F., Strupp, J., Sapiro, G., De, M.F., Wang, D.,
Harel, N., Garwood, M., Chen, L., Feinberg, D.A., Smith, S.M., Miller, K.L.,
Sotiropoulos, S.N., Jbabdi, S., Andersson, J.L., Behrens, T.E., Glasser, M.F., Van Essen,
D.C., Yacoub, E., 2013. Pushing spatial and temporal resolution for functional and
diffusion MRI in the Human Connectome Project. Neuroimage. 80, 80-104.
Uhlhaas, P.J., Singer, W., 2010. Abnormal neural oscillations and synchrony in schizophrenia.
Nat.Rev.Neurosci. 11, 100-113.
Van Essen, D.C., Smith, S.M., Barch, D.M., Behrens, T.E., Yacoub, E., Ugurbil, K., 2013. The
WU-Minn Human Connectome Project: an overview. Neuroimage. 80, 62-79.
van Praag, H., Kempermann, G., Gage, F.H., 1999. Running increases cell proliferation and
neurogenesis in the adult mouse dentate gyrus. Nat.Neurosci. 2, 266-270.
Vincent, J.L., Patel, G.H., Fox, M.D., Snyder, A.Z., Baker, J.T., Van Essen, D.C., Zempel, J.M.,
Snyder, L.H., Corbetta, M., Raichle, M.E., 2007. Intrinsic functional architecture in the
anaesthetized monkey brain. Nature 447, 83-86.
Voss, M.W., Erickson, K.I., Prakash, R.S., Chaddock, L., Malkowski, E., Alves, H., Kim, J.S.,
Morris, K.S., White, S.M., Wojcicki, T.R., Hu, L., Szabo, A., Klamm, E., McAuley, E.,
Kramer, A.F., 2010. Functional connectivity: a source of variance in the association
between cardiorespiratory fitness and cognition? Neuropsychologia 48, 1394-1406.
58
Wen, C.P., Wai, J.P., Tsai, M.K., Yang, Y.C., Cheng, T.Y., Lee, M.C., Chan, H.T., Tsao, C.K.,
Tsai, S.P., Wu, X., 2011. Minimum amount of physical activity for reduced mortality and
extended life expectancy: a prospective cohort study. Lancet 378, 1244-1253.
Winter, B., Breitenstein, C., Mooren, F.C., Voelker, K., Fobker, M., Lechtermann, A., Krueger,
K., Fromme, A., Korsukewitz, C., Floel, A., Knecht, S., 2007. High impact running
improves learning. Neurobiol.Learn.Mem. 87, 597-609.
Yanagisawa, H., Dan, I., Tsuzuki, D., Kato, M., Okamoto, M., Kyutoku, Y., Soya, H., 2010.
Acute moderate exercise elicits increased dorsolateral prefrontal activation and improves
cognitive performance with Stroop test. Neuroimage. 50, 1702-1710.
Zuo, X.N., Kelly, C., Adelstein, J.S., Klein, D.F., Castellanos, F.X., Milham, M.P., 2010.
Reliable intrinsic connectivity networks: test-retest evaluation using ICA and dual
regression approach. Neuroimage. 49, 2163-2177.