SlideShare a Scribd company logo
1 of 22
Download to read offline
Washington University in Saint Louis
Undergraduate Honors Thesis
Analysis of Effective Connectivity under
Different Anesthetic Conditions Using
Resting-State fMRI Data
Author:
Wenshuai Ye
Supervisor:
Nan Lin
A thesis submitted in fulfilment of the requirements
for the degree of Bachelor of Arts
in the
Department of Mathematics
December 2013
WASHINGTON UNIVERSITY IN SAINT LOUIS
Abstract
Department of Mathematics
Bachelor of Arts
Analysis of Effective Connectivity under Different Anesthetic Conditions
Using Resting-State fMRI Data
by Wenshuai Ye
Functional Magnetic Resonance Imaging (fMRI) is a procedure that measures brain ac-
tivity by detecting associated changes in blood flow. Anesthesiologists are interested in
learning the change in brain functional activity within networks linking different cortical
areas under different sevoflurane concentrations by studying resting-state fMRI data. In
this thesis, we perform a mixed effects vector autoregressive model analysis [1] for such
data to analyze the connectivity between different regions of interest (ROIs). This model
captures the condition-specific connectivity (fixed effect), subject-specific connectivity
(random effect), and run-specific connectivity nested within conditions (random effec-
t). In our analysis, we use a simplified version of the original model in SAS based on
efficiency and feasibility. Our results show that the number of significant effective con-
nectivity effects decreases as the anesthetic concentration gets higher. And at the highest
concentration level in the experiment, no cross-ROI connectivity remains significant.
Acknowledgements
I would like to thank Professor Nan Lin for his invaluable assistance and insights leading
to the writing of this paper. My sincere thanks also go to the Professor Ronald Freiwald,
Professor Edward Spitznagel, Professor Jimin Ding, Professor Mladen Wickerhauser,
Doctor Blake Thornton for helpful comments.
ii
Contents
Abstract i
Acknowledgements ii
1 Introduction 1
1.1 Resting-State fMRI Study . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Data Collection and Description . . . . . . . . . . . . . . . . . . . . . . . 2
2 The Mixed Effects Vector Autoregressive Model 4
2.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Hypothesis testing and Effect Estimates . . . . . . . . . . . . . . . . . . . 7
3 Connectivity Analysis of fMRI Data 9
3.1 The Granger Causality Test . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Connectivity Changes Across Conditions . . . . . . . . . . . . . . . . . . . 11
4 Conclusion 14
A Significant Cross-ROI Effects 16
B Estimated Effects of the Granger Causality Testings 17
Bibliography 18
iii
Chapter 1
Introduction
Functional Magnetic Resonance Imaging (fMRI) is a procedure that measures brain
activity by detecting associated changes in blood flow. Usually volunteers are recruited
to perform a set of tasks or in rest while their brains are being scanned by a functional
Magnetic Resonance Imaging machine. It is a technique developed in the last decade and
is mainly used in the area of psychology and neuroscience, including anesthesiology [2].
In recent years there has been explosive growth in the number of neuroimaging studies
performed using fMRI [3]. The primary form of fMRI uses the blood-oxygen-level-
dependent (BOLD) contrast. As a type of specialized brain scan, it maps neural activity
in the brain or spinal cord of humans or animals by imaging the change in blood flow
related to energy use by brain cells. fMRI analysis is an interdisciplinary field involving
collaboration among people specializing in different skills. From a statistical perspective,
this thesis applies the mixed effects vector autoregressive model to an anesthetic study
based on fMRI data.
1.1 Resting-State fMRI Study
Typical fMRI research focuses on the change in blood-oxygen-level-dependent (BOLD)
signal caused by the neural response to external stimulus. Recently the attention has
been shifted to investigating the features of the baseline state of the brain, which repre-
sents the state of the human brain in the absence of goal-directed neuronal action and
external input [4]. This type of fMRI is called resting-state fMRI. As a relatively new
and powerful discipline, resting-state fMRI evaluates regional interactions that occur
when a subject is not performing explicit tasks. Because brain activity is present even
in the absence of an external prompted task, any brain region will have spontaneous
1
Chapter 1. Introduction 2
fluctuations in BOLD signal. By studying the BOLD signal, scientists are usually in-
terested in two types of connectivity: functional connectivity and effective connectivity.
Functional connectivity is defined as statistical dependencies among remote neurophys-
iological events, whereas effective connectivity refers explicitly to the influence that one
neural system exerts over another [5]. A more distinguishable difference is that func-
tional connectivity is defined at the voxel level while effective connectivity is defined for
regions of interest, which is discussed later. Driven by the rapid growth of the interest,
many programs have been created for processing resting state fMRI data accompanied
with these goals. Some common programs include Statistical Parametric Mapping (SP-
M) and Analysis of Functional NeuroImages (AFNI). To process and acquire the resting
state fMRI data with these programs, scientists have come up with many methods, yet
there are two methods that have become the most popular methods for the analysis
of functional and effective connectivity in the brain. One is independent component
analysis (ICA). The other is the method used in this thesis, which is based on regions of
interest (ROIs). Frequently a ROI is drawn on a homogenous area of an MRI image, and
the corresponding voxels are extracted from a coregistered fMRI image. ROI analysis
of the fMRI data requires scientists to pre-specify ROIs, followed by the extraction of
signal from specified ROIs. The selection of ROIs is based on structural or function
features, depending on the motivation to perform the ROI analysis [6]. However, even if
the ROIs have been determined, it is usually difficult to measure the signal within every
ROI. Many imaging experts believe that in order to obtain the information with the
best quality, fMRI and electroencephalography (EEG), a test that measures and records
the electrical activity of the brain, should be used simultaneously. In this thesis, EEG
was adopted in gathering the data.
With the preprocessed data in hand, Statisticians have come up with various com-
plex models to analyze the connectivity between pre-specified regions based on the ROI
analysis and relevant information acquired. One popular model is the mixed effects
vector autoregressive model (ME-VAR).
1.2 Data Collection and Description
Anesthesiologists are interested in learning the change in brain activity under d-
ifferent levels of anesthetic-induced unconsciousness. In this thesis, we analyze fMRI
data from a real study using a mixed model to test whether and how different doses of
anesthesia affect brain functional connectivity. Washington University medical school
acquired and preprocessed the fMRI data by conducting the following experiment.
Chapter 1. Introduction 3
Twenty five human volunteers were initially selected for the study. For each volun-
teer, a facemask connected to a semi-closed anesthetic gas machine circuit was placed
over the nose and mouth. Sevoflurane with various concentrations used for induction
and maintenance of general anesthesia was delivered through this machine. Different
sevoflurane inspired concentrations were implemented for corresponding targeted con-
centrations. During these periods, BOLD-MRI signals were recorded. Meanwhile, EEG
signals were recorded continuously to obtain information with higher quality. The re-
sponse was designed as followed.
Every 30 seconds, volunteers were instructed to respond to block-randomized record-
ed auditory commands “Left” or “Right” by squeezing a pressure transducer in the cor-
responding hand. The motor response was recorded and synchronized to vital signs,
EEG waveforms, and MRI images. Responsiveness was defined as the appropriate re-
sponse to command within the allotted 10 seconds of an auditory command, with the
clinical endpoints of interest being loss of responsiveness (LOR) and return of respon-
siveness (ROR). The focus of the study is on the neural signals recorded at anesthetic
concentrations flanking LOR.
The fMRI data have been preprocessed and twenty two subjects for analysis were
selected from the results of a step that regressed out the global average MR signal
during functional connectivity processing. This step was used to ”clean up” the data by
removing putative physiologic sources of common noise in the signals.
We set two specific goals for this study. The first goal is to estimate and test the
significance of effective pairwise connectivity under different levels of anesthetic-induced
unconsciousness. The second goal is to examine the changes in brain connectivity as the
sevoflurane concentration increases.
Chapter 2
The Mixed Effects Vector
Autoregressive Model
This section presents an introduction and discussion of the mixed effects vector au-
toregressive model (ME-VAR). The model captures (i.) the expected condition-specific
connectivity matrix (fixed effect) over different runs, (ii.) a subject-specific component
(random effect) to model between-subject variation in connectivity, and (iii.) a run-
specific component (random effect) to model the random effects of different runs nested
within conditions. However, along with the comprehensive and complicated structure
of the model come its shortcomings such as the large number of parameters needed
to be estimated. When fitting the model in SAS, several issues are likely to appear.
Some common problems include “insufficient memory” or “convergence criteria does
not meet”. We performed a simplified version of the original model in SAS to overcome
these problems based on feasibility and efficiency.
2.1 Setup
Consider a brain network with R ROIs. This model is proposed to describe the
BOLD response E(s)(t) (dimension R × 1) under different anaesthetic conditions of the
resting state, that is, conditions with different sevoflurane concentrations. Our focus lies
on the neural signals recorded at anesthetic concentrations flanking loss of responsive-
ness (LOR), a concentration denoted as the clinical endpoint of interest. Five conditions
were initially selected (labeled as condition 1 - condition 5). However, condition 4 and
condition 5 have the same sevoflurane concentrations as condition 2 and condition 1 re-
spectively. As such, we only focus on the first three conditions with different sevoflurane
4
Chapter 2. The Mixed Effects Vector Autoregressive Model 5
concentrations. For a specific subject s, the model is written as
E(s)
(t) =
P
k=1
[β
(s)
1j,kW
(s)
1 + β
(s)
2j,kW
(s)
2 + β
(s)
3j,kW
(s)
3 ]E(s)
(t − k) + (s)
(t), (2.1)
where t = P +1, ..., T are the time points up to time T and s = 1, ..., S denotes the subject
with S subjects in total. The vector (s)(t) (dimension R × 1) is a white noise process
with E( (s)(t)) = 0 and Cov( (s)) = Γ (dimension R×R). We assume that Cov( (s)) = Γ
does not change over time and is constant across all participants. k = 1, ..., P denotes
the time lag with maximum order P, and j represents the experimental run, which is
nested within conditions. Other basic elements are more complicated and important.
To understand the model in a more comprehensive way, we enumerate other detailed
elements in the following.
The subject-specific indicator parameter for condition i, denoted as W
(s)
i
This parameter is used as a tool to examine brain connectivity across different
conditions. Suppose that at time t0 a stimulus, a change of sevoflurane concentrations
for condition i was presented; followed by a different sevoflurane concentrations, denoted
as condition i at the time t1. We specify the function W
(s)
i as a time window on which
the brain connectivity present is attributed to condition i in subject s. In this case, this
time window takes a value of 1 in the interval [t0, t1] and takes the value 0 elsewhere.
For example, if the anesthetic concentration in the interval [t0, t1] remains unchanged,
then the corresponding W
(s)
i takes a value of 1.
Optimum for the maximum lag order P
The optimal order can be objectively selected using information theoretic criteria:
Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC), and AIC
with a correction for infinite small sample sizes (AICC). BIC penalizes model complexity
more severely compared with the other two approaches. We take BIC as the desirable
criterion to simplify the model in SAS for the complexity of the model in our analysis.
We first use P = 3. If every parameter from every region turns out estimable with this
maximum order, we choose both higher and lower maximum orders and compare their
BIC. On the other hand, if some of the parameters cannot be estimated with P = 3,
only lower maximum orders should be used. To sum up, the first priority for choosing
the model is to guarantee that every single parameter we are interested can be estimated
in SAS. Then, from these models, we select the one with the lowest BIC.
The subject-specific direct connectivity matrix at lag k, denoted as β
(s)
ij,k
The subject-specific connectivity matrix at lag k, denoted β
(s)
ij,k, includes a subject-
specific random effect to describe the variability in the connectivity pattern across dif-
ferent subjects. This R × R matrix quantifies exactly how E(s)(t − k) directly predicts
Chapter 2. The Mixed Effects Vector Autoregressive Model 6
E(s)(t) under experimental condition i. It also includes an indicator j denoting repli-
cates (runs) of the experiment nested in conditions. As a result, β
(s)
ij,k contains both
random components and fixed components. To correctly specify the model in SAS,
it is important to separate them, that is, to decompose β
(s)
ij,k into random and fixed
components.
Before decomposition, we demonstrate the parameters of the model in vector and
matrix forms. Below is an example to illustrate the matrices and the vectors in the
model under condition 1.







E(s)(t)(1)
E(s)(t)(2)
...
E(s)(t)(R)







=
P
k=1







β
(s)
1j,k(1, 1) β
(s)
1j,k(1, 2) · · · β
(s)
1j,k(1, R)
β
(s)
1j,k(2, 1) β
(s)
1j,k(2, 2) · · · β
(s)
1j,k(2, R)
...
...
...
...
β
(s)
1j,k(R, 1) β
(s)
1j,k(R, 2) · · · β
(s)
1j,k(R, R)







×







E(s)(t − 1)(1)
E(s)(t − 1)(2)
...
E(s)(t − 1)(R)







+ (s)
(t) (2.2)
2.2 Decomposition
Modeling variation across replicates
One important element in the connectivity matrix β
(s)
ij,k is the subscript j indicating
that brain connectivity is allowed to vary over runs. Here, we decompose the connectivity
matrix into two components.
β
(s)
ij,k = β
(s)
i,k + φ
(s)
i(j),k, (2.3)
where β
(s)
i,k is the condition-subject-specific connectivity matrix and φ
(s)
i(j),k is the run-
specific effect nested within conditions or subjects. It models the deviation of each run
from the common condition-subject-specific connectivity matrix associated to lag k. The
elements of the matrix φ
(s)
i(j),k are assumed to be mutually independent normal random
variables.
φ
(s)
i(j),k(r, r ) ∼ N(0, σ2
k(r, r ) + σ 2
k (r, r )) (2.4)
where σ2
k(r, r ) comes from the subject effect and σ 2
k (r, r ) comes from the run effect.
Modeling variation across subjects
Another important element in the connectivity matrix β
(s)
ij,k is the superscript (s)
indicating that brain connectivity is allowed to vary over subjects. This random com-
ponent can be separated from the connectivity. It allows us to further decompose β
(s)
i,k
into a fixed component and a random component.
β
(s)
ij,k = βi,k + b
(s)
k + φ
(s)
i(j),k, (2.5)
Chapter 2. The Mixed Effects Vector Autoregressive Model 7
where βi,k is the expected condition-specific connectivity matrix and b
(s)
k is the subject-
specific effect which models an individual subject’s deviation from the common population-
specific connectivity matrix associated to lag k. We model the elements of the matrix
b
(s)
k as identically distributed independent variables following normal distribution.
b
(s)
k (r, r ) ∼ N(0, σ2
k(r, r )). (2.6)
Decomposition of the ME-VAR model
To illustrate the features of the model, we assume lag order to be P = 1 for simplicity
so that conditional mean of the observed BOLD signal at time t depends only on the
observations at the previous time point t − 1. The model for subject s is specified as
E(s)
(t) = [β
(s)
1j,1W
(s)
1 + β
(s)
2j,1W
(s)
2 + β
(s)
3j,1W
(s)
3 ]E(s)
(t − 1) + e(s)
(t). (2.7)
From the decomposition derived, we can decompose the connectivity matrix into a ran-
dom and two fixed components. The model for subject s becomes
E(s)
(t) =[(β1,1 + b
(s)
1 + φ
(s)
1(j),1)W
(s)
1 + (β2,1 + b
(s)
1 + φ
(s)
2(j),1)W
(s)
2
+ (β3,1 + b
(s)
1 + φ
(s)
3(j),1)W
(s)
3 ]E(s)
(t − 1) + (s)
(t).
(2.8)
However, in real study, even though common variance have been assumed across
subjects and runs, the combined effects of all the factors may still paralyze the software.
In this case, we may have to simplify the model by ignoring the run-specific connectivity
matrix (random effect) and analyze the data for each replicate separately. For each run,
the model then becomes
E(s)
(t) =
P
k=1
[β
(s)
1,kW
(s)
1 + β
(s)
2,kW
(s)
2 + β
(s)
3,kW
(s)
3 ]E(s)
(t − k) + (s)
(t). (2.9)
Without the run-specific connectivity component, β
(s)
1,k can only be decomposed into a
random component and a fixed component;
β
(s)
i,k = βi,k + b
(s)
k . (2.10)
2.3 Hypothesis testing and Effect Estimates
As mentioned, there are two goals for our analysis. The first one is to estimate and
test the significance of effective pairwise connectivity. This process contains 50 × 50 × 3
Chapter 2. The Mixed Effects Vector Autoregressive Model 8
hypothesis testings. The other goal is to examine the changes in brain connectivity
across conditions. The process also contains 50 × 50 × 3 hypothesis testings.
Testing for significance of connectivity
Our interest is to determine whether the past activity at ROI r during condition
i can significantly explain the current activity at ROI r. Given For the model with
maximum order P, the null and the alternative hypothesis can be written as
H0 : βi,1(r, r ) = βi,2(r, r ) = ... = βi,P (r, r ) = 0,
and H1 : At least one of the coefficients parameters are not 0.
(2.11)
βi,k(r, r ) denotes the expected value of β
(s)
ij,k(r, r ), which is the expectation of the random
effect run indicated by the subscript j and the random effect subject indicated by the
superscript (s). In this case, we can say that activity at ROI r “Granger causes” the
activity at ROI r if the null hypothesis is rejected. That is, the activity at ROI r
provides statistically significant information about future activity at ROI r.
Testing for changes of connectivity across conditions
Our second interest is to investigate the differences of brain connectivity across
different conditions. As the sevoflurane concentration changes, we need to determine
if the connectivity within networks linking the thalamus, frontal and parietal cortical
areas change too. The null and the alternative hypothesis can be written as
H0 : βi,k(r, r ) = βi ,k(r, r ),
and H1 : βi,k(r, r ) = βi ,k(r, r ).
(2.12)
This hypothesis testing measures the difference between the expected causality effect of
the brain activity at ROI r and lag order k on ROI r under condition i and that under
condition i . If the null hypothesis is rejected, we can say that the brain connectivity
between ROI r and ROI r under condition i is different from that under condition i .
Both hypothesis testings contain multiple correlated testings. Therefore, it is nec-
essary to control the expected proportion of incorrectly rejected null hypotheses. False
Discovery Rate (FDR) is used to correct for multiple comparisons. Compared with Fam-
ilywise Error Rate (FWER), FDR increases the power at the cost of increasing the rate
of type 1 errors. It is particularly useful in the analysis using datasets with relative-
ly small sample sizes and large numbers of variables being measured per sample. The
threshold q-value for the FDR is 0.1.
Chapter 3
Connectivity Analysis of fMRI
Data
This chapter presents the result of our study. We use statistical packages R and
SAS 9.2 to perform the analysis. R is mainly used for structuring the dataset and graph-
ing relevant plots, whereas SAS is the primary tool to analyze the dataset under proc
MIXED. Under models that incorporate the run-specific connectivity matrix, SAS shows
insufficient memory of the computer. Therefore, to guarantee that every parameter is
estimable in SAS, we simplify the model by using the data from the first run. That is,
we only fitted the model to the first replicate of the data. Moreover, models with 2 lags
and 3 lags do not guarantee that every parameter is estimable. In this situation, we pick
the model with 1 lag eventually when specifying it in SAS, regardless of the BIC.
3.1 The Granger Causality Test
We test the significance of each element in the connectivity matrices βi,k to investi-
gate the lag cross-dependence structure in each condition. k equals 1 here since we fitted
the model with only 1 lag. To identify the effective connectivity in pairs of brain regions,
we perform a Granger causality test, as discussed in Chapter 2. For each condition, a
heatmap (Figure 3.1) containing information on the effect of all regions of interest(ROIs)
is presented. Column ROIs are response ROIs. Row ROIs represent regions that help
us predict the influence on the response ROIs. In other words, Column regions are the
response in our model, whereas row regions are regions with a time lag (lag 1).
Each heatmap in Figure 3.1 represents the connectivity between different ROIs under
each condition. Black here denotes that the Granger causality is significant after FDR
9
Chapter 3. Connectivity Analysis of fMRI Data 10
Figure 3.1: Granger Causality Heatmaps
correction. As for those insignificant effects, we use grey and white colors. The color
gets deeper as the connectivity gets stronger. We notice that the lag dependence of a
ROI on its own region is always significant under any condition. Besides, there are 19
significant cross-dependence effects under condition 1, as shown in the heatmap. Under
condition 2 with the increase of sevoflurane concentration, the number of significant
cross-dependence effects reduces to 9, all of which are not the same ROIs as those shown
in condition 1. As the sevoflurane concentration increases further, none of the cross-
dependence effects becomes significant. The only significant lag dependence of a ROI
is its own region. The details of the significant cross-dependence effects are listed in
Appendix A. The overall pattern implies that as the concentration of sevoflurane goes
Chapter 3. Connectivity Analysis of fMRI Data 11
up, the overall connectivity between two ROIs decreases. To improve the accuracy of
our findings, we implement a test of changes in connectivity across different conditions.
3.2 Connectivity Changes Across Conditions
To investigate whether there is a change in connectivity across conditions, we test
if the two connectivity matrix βi,k and βi ,kare equal as the sevoflurane concentration
changes. This procedure is implemented in SAS 9.2. We then used the “qvalue” package
in R to calculate the corrected threshold for the p-values by converting them to q-values.
It turns out that none of the q-values are significant in this case. As a result, none of
the connectivity changes is actually significant. However, it is still valuable to dig out
more information based on SAS outputs and heatmaps. Our focus lies primarily on the
significant connectivity based on the Granger causality test discussed in the previous
section. If the connectivity does not exist, it does not make much sense to discuss the
changes in connectivity further.
These heatmaps in Figure 3.2 represent the connectivity changes under different
conditions. Darker colors indicate that the changes are larger. However, since none of the
changes are significant, black does not mean the changes are significant. It only implies
the changes are strong. The heatmaps help us to identify patterns. As shown, there is
no explicit pattern in the first two heatmaps. However, with a more significant increase
in the sevoflurane concentration, the connectivity changes become noticeable as shown
in the third heatmap. The diagonal color gets deeper in general compared with the first
two heatmaps. From the pattern, we guess that the changes in connectivity are in one
direction - either the connectivity gets stronger when the sevoflurane concentration goes
up, or it gets weaker. Figure 3.1 shows that the connectivity weakens with the increase
in sevoflurane concentration. However, since our focus is on the significant connectivity,
it is necessary to investigate those values to see the direction of connectivity changes.
To find out the direction, we referred to the significant estimated values of the Granger
causality effect under all conditions. From Figure 3.1, we can see no cross-dependence
under condition 3. As a result, for each ROI, we only focused on the causality effects
from itself with a time lag (lag 1).
Appendix B shows the estimated effect of each region under each condition. From
condition 1 to condition 2 as the sevoflurane concentration increases, the estimated ef-
fects of the connectivity go up except for ROI36, ROI42, and ROI46. However, from
condition 2 to condition 3, as the sevoflurane concentration increases further, the con-
nectivity trend in general becomes vaguer, though the connectivity still weakens in total
as shown in Figure 3.3. Each line represents a sevoflurane concentration across ROIs.
Chapter 3. Connectivity Analysis of fMRI Data 12
Figure 3.2: Connectivity Changes Heatmaps
The horizontal axis denotes ROIs, whereas the vertical axis denotes estimated effects for
each corresponding ROIs.
Chapter 3. Connectivity Analysis of fMRI Data 13
Figure 3.3: Connectivity Plot across ROIs
Chapter 4
Conclusion
In this thesis, we explored the effect of anesthesia on brain effective connectivity. We
used a mixed effects vector autoregressive model (ME-VAR) to analyze a resting-state
fMRI dataset. This model captures the expected condition-specific component (fixed
effect), the subject specific component (random effect), and the run-specific component
(random effect). Theoretically, this is a comprehensive model for the analysis of the
resting-state FMRI data. However, when applying to large datasets with many ROIs,
it exceeds the limit of SAS and hence simplification of the model is necessary. Based
on feasibility, we removed the run-specific component from the model and only demon-
strated the result from the first replicate of the dataset. Since evaluation or testing of a
single item does not allow for item-to-item variation and may not represent the batch or
process, the model without a run-specific matrix is likely to contain bias. Moreover, even
with the simplified model and data information, the maximum lag order is restricted to
1 in SAS, which is another limit of the model in this analysis.
Based on the heatmap, the result shows that 69 Granger causality effects are signif-
icant under condition 1. 50 of them are effects within ROIs over time. Under condition
2, 59 Granger causality effects are significant. Among them, 50 effects within ROIs over
time are still significant. As the sevoflurane concentration goes up further (condition
3), there are only 50 significant Granger causality effects, all of which are effects within
ROIs over time.
We then tested for the significance of the connectivity changes. It turns out that
none of the changes is significant. Despite the insignificance presented, we can still find a
trend that from condition 1 to condition 2 with the increase in sevoflurane concentration,
the connectivity generally becomes weaker, with few exceptions. From condition 2 to
condition 3, the direction of the changes become vaguer. However, the connectivity still
gets weaker on aggregate. In the future study, the difference of sevoflurane concentration
14
Chapter 4. Conclusion 15
among conditions can be more substantial in order for us to visualize the connectivity
changes better.
Appendix A
Significant Cross-ROI Effects
condition1 condition2 condition3
ROI01→ROI15 ROI06→ROI49 ROI19→ROI49 ROI07→ROI49 ROI10→ROI49
ROI43→ROI36 ROI28→ROI12 ROI44→ROI36 ROI12→ROI49 ROI29→ROI14
ROI30→ROI36 ROI46→ROI36 ROI31→ROI36 ROI38→ROI31 ROI41→ROI31
ROI36→ROI38 ROI37→ROI36 ROI36→ROI48 ROI42→ROI31 ROI48→ROI32
ROI38→ROI36 ROI40→ROI44 ROI39→ROI36 ROI49→ROI17
ROI37→ROI48 ROI40→ROI36 ROI38→ROI48
ROI40→ROI48
16
Appendix B
Estimated Effects of the Granger
Causality Testings
ROIs condition1 condition2 condition3 ROIs condition1 condition2 condition3
ROI1 0.7952652 0.7406001 0.7209555 ROI26 0.7709983 0.6702235 0.6885903
ROI2 0.7169583 0.6979163 0.5990974 ROI27 0.7894696 0.7809740 0.6930602
ROI3 0.8088205 0.7619170 0.6537976 ROI28 0.7304779 0.7205294 0.7228392
ROI4 0.8199955 0.7108279 0.6914172 ROI29 0.7273849 0.7055247 0.6709614
ROI5 0.6966409 0.6447304 0.6453244 ROI30 0.7810669 0.7444860 0.7055461
ROI6 0.7178230 0.7269425 0.6152923 ROI31 0.6942039 0.6709644 0.6105873
ROI7 0.7428886 0.6928639 0.7060899 ROI32 0.7216514 0.7042265 0.6398730
ROI8 0.7533285 0.7180890 0.6002316 ROI33 0.7076750 0.6858878 0.6270619
ROI9 0.7448033 0.6964896 0.6999977 ROI34 0.6417452 0.5972246 0.6158163
ROI10 0.7405824 0.7127173 0.7392774 ROI35 0.6354104 0.6075662 0.5722229
ROI11 0.7403181 0.7103553 0.6728991 ROI36 0.4412667 0.5161668 0.5357359
ROI12 0.7315507 0.7362889 0.6877589 ROI37 0.7790736 0.7009024 0.6537599
ROI13 0.7367875 0.6523549 0.6761815 ROI38 0.7876667 0.7411852 0.7356888
ROI14 0.7412977 0.6965885 0.6984047 ROI39 0.6479298 0.6074324 0.6146055
ROI15 0.7366455 0.7094993 0.6603111 ROI40 0.6345684 0.5881167 0.6609347
ROI16 0.7049970 0.6632437 0.6330410 ROI41 0.7285654 0.6526174 0.6368883
ROI17 0.7710784 0.6583128 0.5939490 ROI42 0.6411857 0.6785500 0.6836995
ROI18 0.7503926 0.7099066 0.6509170 ROI43 0.6711933 0.6094010 0.6600454
ROI19 0.7755234 0.7190092 0.7280341 ROI44 0.6931007 0.6377876 0.5954626
ROI20 0.7672113 0.7196823 0.6337438 ROI45 0.7214215 0.7022580 0.7021112
ROI21 0.6921319 0.6425770 0.5799561 ROI46 0.6991743 0.7237863 0.7231898
ROI22 0.7276675 0.6848949 0.6267191 ROI47 0.7591686 0.6389944 0.6334283
ROI23 0.7499073 0.6820988 0.6768443 ROI48 0.7615498 0.6183721 0.6661121
ROI24 0.7289385 0.6023442 0.6239701 ROI49 0.6787235 0.6522500 0.6039819
ROI25 0.7416147 0.6635005 0.6665899 ROI50 0.7133189 0.7122303 0.5711190
17
Bibliography
[1] Cristina Gorrostieta, Hernando Ombao, Patrick Bdard b, and Jerome N. Sanes.
Investigating brain connectivity using mixed effects vector autoregressive models.
NeuroImage, 59:3347 – 3355, 2012. doi: 10.1016/j.neuroimage.2011.08.115.
[2] Nicole A. Lazar, William F. Eddy, Christopher R. Genovese, and Joel Welling. Sta-
tistical issues in fmri for brain imaging. International Statistical Review, 69:105127,
2001. doi: 10.1111/j.1751-5823.2001.tb00482.x.
[3] Martin A. Lindquist. The statistical analysis of fmri data. Statistical Science, 23:
439 – 464, 2008. doi: 10.1214/09-STS282.
[4] J. S. Damoiseaux, S. A. R. B. Rombouts, F. Barkhof, P. Scheltens, C. J. Stam,
S. M. Smith, and C. F. Beckmann. Consistent resting-state networks across healthy
subjects. PNAS, 103:1384813853, 2006. doi: 10.1073/pnas.0601417103.
[5] Karl J. Friston. Functional and effective connectivity: A review. Brain Connectivity,
1:13 – 36, 2011. doi: 10.1089/brain.2011.0008.
[6] Russell A. Poldrack. Region of interest analysis for fmri. Soc Cogn Affect Neurosci,
2:6770, 2007. doi: 10.1093/scan/nsm006.
18

More Related Content

What's hot

ANALYSIS OF BRAIN COGNITIVE STATE FOR ARITHMETIC TASK AND MOTOR TASK USING EL...
ANALYSIS OF BRAIN COGNITIVE STATE FOR ARITHMETIC TASK AND MOTOR TASK USING EL...ANALYSIS OF BRAIN COGNITIVE STATE FOR ARITHMETIC TASK AND MOTOR TASK USING EL...
ANALYSIS OF BRAIN COGNITIVE STATE FOR ARITHMETIC TASK AND MOTOR TASK USING EL...sipij
 
CLASSIFICATION OF ALZHEIMER USING fMRI DATA AND BRAIN NETWORK
CLASSIFICATION OF ALZHEIMER USING fMRI DATA AND BRAIN NETWORKCLASSIFICATION OF ALZHEIMER USING fMRI DATA AND BRAIN NETWORK
CLASSIFICATION OF ALZHEIMER USING fMRI DATA AND BRAIN NETWORKcscpconf
 
IRJET- Facial Expression Recognition System using Neural Network based on...
IRJET-  	  Facial Expression Recognition System using Neural Network based on...IRJET-  	  Facial Expression Recognition System using Neural Network based on...
IRJET- Facial Expression Recognition System using Neural Network based on...IRJET Journal
 
Autism_risk_factors
Autism_risk_factorsAutism_risk_factors
Autism_risk_factorsColleen Chen
 
Aiding the Aid: Computational Early Clinical Diagnosis of Electronic Health R...
Aiding the Aid: Computational Early Clinical Diagnosis of Electronic Health R...Aiding the Aid: Computational Early Clinical Diagnosis of Electronic Health R...
Aiding the Aid: Computational Early Clinical Diagnosis of Electronic Health R...Tarun Amarnath
 
A clonal based algorithm for the reconstruction of genetic network using s sy...
A clonal based algorithm for the reconstruction of genetic network using s sy...A clonal based algorithm for the reconstruction of genetic network using s sy...
A clonal based algorithm for the reconstruction of genetic network using s sy...eSAT Journals
 
Teaching Techniques: Neurotechnologies the way of the future (Stotler, 2019)
Teaching Techniques: Neurotechnologies the way of the future (Stotler, 2019)Teaching Techniques: Neurotechnologies the way of the future (Stotler, 2019)
Teaching Techniques: Neurotechnologies the way of the future (Stotler, 2019)Jacob Stotler
 
A clonal based algorithm for the reconstruction of
A clonal based algorithm for the reconstruction ofA clonal based algorithm for the reconstruction of
A clonal based algorithm for the reconstruction ofeSAT Publishing House
 
Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison
Analysis of EEG data Using ICA and Algorithm Development for Energy ComparisonAnalysis of EEG data Using ICA and Algorithm Development for Energy Comparison
Analysis of EEG data Using ICA and Algorithm Development for Energy Comparisonijsrd.com
 
A framework for approaches to transfer of mind substrate
A framework for approaches to transfer of mind substrateA framework for approaches to transfer of mind substrate
A framework for approaches to transfer of mind substrateKarlos Svoboda
 
A Novel Approach to Study the Effects of Anesthesia on Respiratory Signals by...
A Novel Approach to Study the Effects of Anesthesia on Respiratory Signals by...A Novel Approach to Study the Effects of Anesthesia on Respiratory Signals by...
A Novel Approach to Study the Effects of Anesthesia on Respiratory Signals by...IJECEIAES
 
19 3 sep17 21may 6657 t269 revised (edit ndit)
19 3 sep17 21may 6657 t269 revised (edit ndit)19 3 sep17 21may 6657 t269 revised (edit ndit)
19 3 sep17 21may 6657 t269 revised (edit ndit)IAESIJEECS
 
Performance analysis of neural network models for oxazolines and oxazoles der...
Performance analysis of neural network models for oxazolines and oxazoles der...Performance analysis of neural network models for oxazolines and oxazoles der...
Performance analysis of neural network models for oxazolines and oxazoles der...ijistjournal
 
K-NN Classifier Performs Better Than K-Means Clustering in Missing Value Imp...
K-NN Classifier Performs Better Than K-Means Clustering in  Missing Value Imp...K-NN Classifier Performs Better Than K-Means Clustering in  Missing Value Imp...
K-NN Classifier Performs Better Than K-Means Clustering in Missing Value Imp...IOSR Journals
 
Alotaiby2014 article eeg_seizure_detectionandpredicti
Alotaiby2014 article eeg_seizure_detectionandpredictiAlotaiby2014 article eeg_seizure_detectionandpredicti
Alotaiby2014 article eeg_seizure_detectionandpredictiMuhammad Rizwan
 
AN EFFICIENT PSO BASED ENSEMBLE CLASSIFICATION MODEL ON HIGH DIMENSIONAL DATA...
AN EFFICIENT PSO BASED ENSEMBLE CLASSIFICATION MODEL ON HIGH DIMENSIONAL DATA...AN EFFICIENT PSO BASED ENSEMBLE CLASSIFICATION MODEL ON HIGH DIMENSIONAL DATA...
AN EFFICIENT PSO BASED ENSEMBLE CLASSIFICATION MODEL ON HIGH DIMENSIONAL DATA...ijsc
 

What's hot (18)

ANALYSIS OF BRAIN COGNITIVE STATE FOR ARITHMETIC TASK AND MOTOR TASK USING EL...
ANALYSIS OF BRAIN COGNITIVE STATE FOR ARITHMETIC TASK AND MOTOR TASK USING EL...ANALYSIS OF BRAIN COGNITIVE STATE FOR ARITHMETIC TASK AND MOTOR TASK USING EL...
ANALYSIS OF BRAIN COGNITIVE STATE FOR ARITHMETIC TASK AND MOTOR TASK USING EL...
 
CLASSIFICATION OF ALZHEIMER USING fMRI DATA AND BRAIN NETWORK
CLASSIFICATION OF ALZHEIMER USING fMRI DATA AND BRAIN NETWORKCLASSIFICATION OF ALZHEIMER USING fMRI DATA AND BRAIN NETWORK
CLASSIFICATION OF ALZHEIMER USING fMRI DATA AND BRAIN NETWORK
 
Ferree Resume
Ferree ResumeFerree Resume
Ferree Resume
 
IRJET- Facial Expression Recognition System using Neural Network based on...
IRJET-  	  Facial Expression Recognition System using Neural Network based on...IRJET-  	  Facial Expression Recognition System using Neural Network based on...
IRJET- Facial Expression Recognition System using Neural Network based on...
 
Autism_risk_factors
Autism_risk_factorsAutism_risk_factors
Autism_risk_factors
 
Aiding the Aid: Computational Early Clinical Diagnosis of Electronic Health R...
Aiding the Aid: Computational Early Clinical Diagnosis of Electronic Health R...Aiding the Aid: Computational Early Clinical Diagnosis of Electronic Health R...
Aiding the Aid: Computational Early Clinical Diagnosis of Electronic Health R...
 
A clonal based algorithm for the reconstruction of genetic network using s sy...
A clonal based algorithm for the reconstruction of genetic network using s sy...A clonal based algorithm for the reconstruction of genetic network using s sy...
A clonal based algorithm for the reconstruction of genetic network using s sy...
 
Teaching Techniques: Neurotechnologies the way of the future (Stotler, 2019)
Teaching Techniques: Neurotechnologies the way of the future (Stotler, 2019)Teaching Techniques: Neurotechnologies the way of the future (Stotler, 2019)
Teaching Techniques: Neurotechnologies the way of the future (Stotler, 2019)
 
A clonal based algorithm for the reconstruction of
A clonal based algorithm for the reconstruction ofA clonal based algorithm for the reconstruction of
A clonal based algorithm for the reconstruction of
 
Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison
Analysis of EEG data Using ICA and Algorithm Development for Energy ComparisonAnalysis of EEG data Using ICA and Algorithm Development for Energy Comparison
Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison
 
A framework for approaches to transfer of mind substrate
A framework for approaches to transfer of mind substrateA framework for approaches to transfer of mind substrate
A framework for approaches to transfer of mind substrate
 
A Novel Approach to Study the Effects of Anesthesia on Respiratory Signals by...
A Novel Approach to Study the Effects of Anesthesia on Respiratory Signals by...A Novel Approach to Study the Effects of Anesthesia on Respiratory Signals by...
A Novel Approach to Study the Effects of Anesthesia on Respiratory Signals by...
 
19 3 sep17 21may 6657 t269 revised (edit ndit)
19 3 sep17 21may 6657 t269 revised (edit ndit)19 3 sep17 21may 6657 t269 revised (edit ndit)
19 3 sep17 21may 6657 t269 revised (edit ndit)
 
Performance analysis of neural network models for oxazolines and oxazoles der...
Performance analysis of neural network models for oxazolines and oxazoles der...Performance analysis of neural network models for oxazolines and oxazoles der...
Performance analysis of neural network models for oxazolines and oxazoles der...
 
K-NN Classifier Performs Better Than K-Means Clustering in Missing Value Imp...
K-NN Classifier Performs Better Than K-Means Clustering in  Missing Value Imp...K-NN Classifier Performs Better Than K-Means Clustering in  Missing Value Imp...
K-NN Classifier Performs Better Than K-Means Clustering in Missing Value Imp...
 
Alotaiby2014 article eeg_seizure_detectionandpredicti
Alotaiby2014 article eeg_seizure_detectionandpredictiAlotaiby2014 article eeg_seizure_detectionandpredicti
Alotaiby2014 article eeg_seizure_detectionandpredicti
 
Connectomics_Journal Club
Connectomics_Journal ClubConnectomics_Journal Club
Connectomics_Journal Club
 
AN EFFICIENT PSO BASED ENSEMBLE CLASSIFICATION MODEL ON HIGH DIMENSIONAL DATA...
AN EFFICIENT PSO BASED ENSEMBLE CLASSIFICATION MODEL ON HIGH DIMENSIONAL DATA...AN EFFICIENT PSO BASED ENSEMBLE CLASSIFICATION MODEL ON HIGH DIMENSIONAL DATA...
AN EFFICIENT PSO BASED ENSEMBLE CLASSIFICATION MODEL ON HIGH DIMENSIONAL DATA...
 

Viewers also liked (13)

AngularJS
AngularJSAngularJS
AngularJS
 
IAFF2190FinalPaper(1)
IAFF2190FinalPaper(1)IAFF2190FinalPaper(1)
IAFF2190FinalPaper(1)
 
Supreme Research Project
Supreme Research ProjectSupreme Research Project
Supreme Research Project
 
Papéis Colados 4A
Papéis Colados 4APapéis Colados 4A
Papéis Colados 4A
 
Probleme electrocinetica1
Probleme electrocinetica1Probleme electrocinetica1
Probleme electrocinetica1
 
функцииMs excel
функцииMs excelфункцииMs excel
функцииMs excel
 
Ecommerce site start up
Ecommerce site start upEcommerce site start up
Ecommerce site start up
 
diploma
diplomadiploma
diploma
 
OMB SF181 Nov.13 Redacted
OMB SF181 Nov.13 RedactedOMB SF181 Nov.13 Redacted
OMB SF181 Nov.13 Redacted
 
ali.docx
ali.docxali.docx
ali.docx
 
Top 8 construction administrator resume samples
Top 8 construction administrator resume samplesTop 8 construction administrator resume samples
Top 8 construction administrator resume samples
 
PP 20 tahun 2016 tentang THR PNS, TNI, POLRI & Pejabat Negara
PP 20 tahun 2016  tentang  THR PNS, TNI, POLRI & Pejabat NegaraPP 20 tahun 2016  tentang  THR PNS, TNI, POLRI & Pejabat Negara
PP 20 tahun 2016 tentang THR PNS, TNI, POLRI & Pejabat Negara
 
Keluarga Berencana
Keluarga BerencanaKeluarga Berencana
Keluarga Berencana
 

Similar to Honors Thesis

fazowemetodybadaniamozguver10en.pdf
fazowemetodybadaniamozguver10en.pdffazowemetodybadaniamozguver10en.pdf
fazowemetodybadaniamozguver10en.pdfJerzyAchimowicz2
 
Analyzing Complex Problem Solving by Dynamic Brain Networks.pdf
Analyzing Complex Problem Solving by Dynamic Brain Networks.pdfAnalyzing Complex Problem Solving by Dynamic Brain Networks.pdf
Analyzing Complex Problem Solving by Dynamic Brain Networks.pdfNancy Ideker
 
Positron-emission tomography studies of cross-modality inhibition in selectiv...
Positron-emission tomography studies of cross-modality inhibition in selectiv...Positron-emission tomography studies of cross-modality inhibition in selectiv...
Positron-emission tomography studies of cross-modality inhibition in selectiv...Dr Brendan O'Sullivan
 
The Preface Layer for Auditing Sensual Interacts of Primary Distress Conceali...
The Preface Layer for Auditing Sensual Interacts of Primary Distress Conceali...The Preface Layer for Auditing Sensual Interacts of Primary Distress Conceali...
The Preface Layer for Auditing Sensual Interacts of Primary Distress Conceali...IIRindia
 
asp2011_abstracts_modified
asp2011_abstracts_modifiedasp2011_abstracts_modified
asp2011_abstracts_modifiedChris Skaroupka
 
ARTICLE_RNROY_Neuroergo
ARTICLE_RNROY_NeuroergoARTICLE_RNROY_Neuroergo
ARTICLE_RNROY_NeuroergoAlexandre Moly
 
Final Progress Report- Farzana
Final Progress Report- FarzanaFinal Progress Report- Farzana
Final Progress Report- FarzanaFarzana Ali
 
Comparison of fnir with other neuroimaging modalities relation between eeg sy...
Comparison of fnir with other neuroimaging modalities relation between eeg sy...Comparison of fnir with other neuroimaging modalities relation between eeg sy...
Comparison of fnir with other neuroimaging modalities relation between eeg sy...M. Raihan
 
Convolutional Networks
Convolutional NetworksConvolutional Networks
Convolutional NetworksNicole Savoie
 
Using Brain Stimulation and Imaging Techniquues to Study Human Movement by Jo...
Using Brain Stimulation and Imaging Techniquues to Study Human Movement by Jo...Using Brain Stimulation and Imaging Techniquues to Study Human Movement by Jo...
Using Brain Stimulation and Imaging Techniquues to Study Human Movement by Jo...Brown Fellows Program
 
STUDY AND IMPLEMENTATION OF ADVANCED NEUROERGONOMIC TECHNIQUES
STUDY AND IMPLEMENTATION OF ADVANCED NEUROERGONOMIC TECHNIQUES STUDY AND IMPLEMENTATION OF ADVANCED NEUROERGONOMIC TECHNIQUES
STUDY AND IMPLEMENTATION OF ADVANCED NEUROERGONOMIC TECHNIQUES acijjournal
 
Magnetoencephalography an emerging biological marker for neurodegenerative an...
Magnetoencephalography an emerging biological marker for neurodegenerative an...Magnetoencephalography an emerging biological marker for neurodegenerative an...
Magnetoencephalography an emerging biological marker for neurodegenerative an...Adonis Sfera, MD
 
Talking psychiatry...Functional neuroimaging of schizophrenia
Talking psychiatry...Functional neuroimaging of schizophreniaTalking psychiatry...Functional neuroimaging of schizophrenia
Talking psychiatry...Functional neuroimaging of schizophreniaProfessor Yasser Metwally
 
Time-Resolved fMRI-fMRS measures simultaneous Neurotransmitters and BOLD-fMRI...
Time-Resolved fMRI-fMRS measures simultaneous Neurotransmitters and BOLD-fMRI...Time-Resolved fMRI-fMRS measures simultaneous Neurotransmitters and BOLD-fMRI...
Time-Resolved fMRI-fMRS measures simultaneous Neurotransmitters and BOLD-fMRI...Uzay Emir
 

Similar to Honors Thesis (20)

fazowemetodybadaniamozguver10en.pdf
fazowemetodybadaniamozguver10en.pdffazowemetodybadaniamozguver10en.pdf
fazowemetodybadaniamozguver10en.pdf
 
Analyzing Complex Problem Solving by Dynamic Brain Networks.pdf
Analyzing Complex Problem Solving by Dynamic Brain Networks.pdfAnalyzing Complex Problem Solving by Dynamic Brain Networks.pdf
Analyzing Complex Problem Solving by Dynamic Brain Networks.pdf
 
Positron-emission tomography studies of cross-modality inhibition in selectiv...
Positron-emission tomography studies of cross-modality inhibition in selectiv...Positron-emission tomography studies of cross-modality inhibition in selectiv...
Positron-emission tomography studies of cross-modality inhibition in selectiv...
 
The Preface Layer for Auditing Sensual Interacts of Primary Distress Conceali...
The Preface Layer for Auditing Sensual Interacts of Primary Distress Conceali...The Preface Layer for Auditing Sensual Interacts of Primary Distress Conceali...
The Preface Layer for Auditing Sensual Interacts of Primary Distress Conceali...
 
asp2011_abstracts_modified
asp2011_abstracts_modifiedasp2011_abstracts_modified
asp2011_abstracts_modified
 
ISBI_poster
ISBI_posterISBI_poster
ISBI_poster
 
ARTICLE_RNROY_Neuroergo
ARTICLE_RNROY_NeuroergoARTICLE_RNROY_Neuroergo
ARTICLE_RNROY_Neuroergo
 
rizk_report
rizk_reportrizk_report
rizk_report
 
Bx36449453
Bx36449453Bx36449453
Bx36449453
 
Final Progress Report- Farzana
Final Progress Report- FarzanaFinal Progress Report- Farzana
Final Progress Report- Farzana
 
fMRI and sMRI.pptx
fMRI and sMRI.pptxfMRI and sMRI.pptx
fMRI and sMRI.pptx
 
Comparison of fnir with other neuroimaging modalities relation between eeg sy...
Comparison of fnir with other neuroimaging modalities relation between eeg sy...Comparison of fnir with other neuroimaging modalities relation between eeg sy...
Comparison of fnir with other neuroimaging modalities relation between eeg sy...
 
Convolutional Networks
Convolutional NetworksConvolutional Networks
Convolutional Networks
 
Using Brain Stimulation and Imaging Techniquues to Study Human Movement by Jo...
Using Brain Stimulation and Imaging Techniquues to Study Human Movement by Jo...Using Brain Stimulation and Imaging Techniquues to Study Human Movement by Jo...
Using Brain Stimulation and Imaging Techniquues to Study Human Movement by Jo...
 
Abstract
AbstractAbstract
Abstract
 
NCP_Thesis_Francesca_Bocca.pdf
NCP_Thesis_Francesca_Bocca.pdfNCP_Thesis_Francesca_Bocca.pdf
NCP_Thesis_Francesca_Bocca.pdf
 
STUDY AND IMPLEMENTATION OF ADVANCED NEUROERGONOMIC TECHNIQUES
STUDY AND IMPLEMENTATION OF ADVANCED NEUROERGONOMIC TECHNIQUES STUDY AND IMPLEMENTATION OF ADVANCED NEUROERGONOMIC TECHNIQUES
STUDY AND IMPLEMENTATION OF ADVANCED NEUROERGONOMIC TECHNIQUES
 
Magnetoencephalography an emerging biological marker for neurodegenerative an...
Magnetoencephalography an emerging biological marker for neurodegenerative an...Magnetoencephalography an emerging biological marker for neurodegenerative an...
Magnetoencephalography an emerging biological marker for neurodegenerative an...
 
Talking psychiatry...Functional neuroimaging of schizophrenia
Talking psychiatry...Functional neuroimaging of schizophreniaTalking psychiatry...Functional neuroimaging of schizophrenia
Talking psychiatry...Functional neuroimaging of schizophrenia
 
Time-Resolved fMRI-fMRS measures simultaneous Neurotransmitters and BOLD-fMRI...
Time-Resolved fMRI-fMRS measures simultaneous Neurotransmitters and BOLD-fMRI...Time-Resolved fMRI-fMRS measures simultaneous Neurotransmitters and BOLD-fMRI...
Time-Resolved fMRI-fMRS measures simultaneous Neurotransmitters and BOLD-fMRI...
 

Honors Thesis

  • 1. Washington University in Saint Louis Undergraduate Honors Thesis Analysis of Effective Connectivity under Different Anesthetic Conditions Using Resting-State fMRI Data Author: Wenshuai Ye Supervisor: Nan Lin A thesis submitted in fulfilment of the requirements for the degree of Bachelor of Arts in the Department of Mathematics December 2013
  • 2. WASHINGTON UNIVERSITY IN SAINT LOUIS Abstract Department of Mathematics Bachelor of Arts Analysis of Effective Connectivity under Different Anesthetic Conditions Using Resting-State fMRI Data by Wenshuai Ye Functional Magnetic Resonance Imaging (fMRI) is a procedure that measures brain ac- tivity by detecting associated changes in blood flow. Anesthesiologists are interested in learning the change in brain functional activity within networks linking different cortical areas under different sevoflurane concentrations by studying resting-state fMRI data. In this thesis, we perform a mixed effects vector autoregressive model analysis [1] for such data to analyze the connectivity between different regions of interest (ROIs). This model captures the condition-specific connectivity (fixed effect), subject-specific connectivity (random effect), and run-specific connectivity nested within conditions (random effec- t). In our analysis, we use a simplified version of the original model in SAS based on efficiency and feasibility. Our results show that the number of significant effective con- nectivity effects decreases as the anesthetic concentration gets higher. And at the highest concentration level in the experiment, no cross-ROI connectivity remains significant.
  • 3. Acknowledgements I would like to thank Professor Nan Lin for his invaluable assistance and insights leading to the writing of this paper. My sincere thanks also go to the Professor Ronald Freiwald, Professor Edward Spitznagel, Professor Jimin Ding, Professor Mladen Wickerhauser, Doctor Blake Thornton for helpful comments. ii
  • 4. Contents Abstract i Acknowledgements ii 1 Introduction 1 1.1 Resting-State fMRI Study . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Data Collection and Description . . . . . . . . . . . . . . . . . . . . . . . 2 2 The Mixed Effects Vector Autoregressive Model 4 2.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2.2 Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.3 Hypothesis testing and Effect Estimates . . . . . . . . . . . . . . . . . . . 7 3 Connectivity Analysis of fMRI Data 9 3.1 The Granger Causality Test . . . . . . . . . . . . . . . . . . . . . . . . . . 9 3.2 Connectivity Changes Across Conditions . . . . . . . . . . . . . . . . . . . 11 4 Conclusion 14 A Significant Cross-ROI Effects 16 B Estimated Effects of the Granger Causality Testings 17 Bibliography 18 iii
  • 5. Chapter 1 Introduction Functional Magnetic Resonance Imaging (fMRI) is a procedure that measures brain activity by detecting associated changes in blood flow. Usually volunteers are recruited to perform a set of tasks or in rest while their brains are being scanned by a functional Magnetic Resonance Imaging machine. It is a technique developed in the last decade and is mainly used in the area of psychology and neuroscience, including anesthesiology [2]. In recent years there has been explosive growth in the number of neuroimaging studies performed using fMRI [3]. The primary form of fMRI uses the blood-oxygen-level- dependent (BOLD) contrast. As a type of specialized brain scan, it maps neural activity in the brain or spinal cord of humans or animals by imaging the change in blood flow related to energy use by brain cells. fMRI analysis is an interdisciplinary field involving collaboration among people specializing in different skills. From a statistical perspective, this thesis applies the mixed effects vector autoregressive model to an anesthetic study based on fMRI data. 1.1 Resting-State fMRI Study Typical fMRI research focuses on the change in blood-oxygen-level-dependent (BOLD) signal caused by the neural response to external stimulus. Recently the attention has been shifted to investigating the features of the baseline state of the brain, which repre- sents the state of the human brain in the absence of goal-directed neuronal action and external input [4]. This type of fMRI is called resting-state fMRI. As a relatively new and powerful discipline, resting-state fMRI evaluates regional interactions that occur when a subject is not performing explicit tasks. Because brain activity is present even in the absence of an external prompted task, any brain region will have spontaneous 1
  • 6. Chapter 1. Introduction 2 fluctuations in BOLD signal. By studying the BOLD signal, scientists are usually in- terested in two types of connectivity: functional connectivity and effective connectivity. Functional connectivity is defined as statistical dependencies among remote neurophys- iological events, whereas effective connectivity refers explicitly to the influence that one neural system exerts over another [5]. A more distinguishable difference is that func- tional connectivity is defined at the voxel level while effective connectivity is defined for regions of interest, which is discussed later. Driven by the rapid growth of the interest, many programs have been created for processing resting state fMRI data accompanied with these goals. Some common programs include Statistical Parametric Mapping (SP- M) and Analysis of Functional NeuroImages (AFNI). To process and acquire the resting state fMRI data with these programs, scientists have come up with many methods, yet there are two methods that have become the most popular methods for the analysis of functional and effective connectivity in the brain. One is independent component analysis (ICA). The other is the method used in this thesis, which is based on regions of interest (ROIs). Frequently a ROI is drawn on a homogenous area of an MRI image, and the corresponding voxels are extracted from a coregistered fMRI image. ROI analysis of the fMRI data requires scientists to pre-specify ROIs, followed by the extraction of signal from specified ROIs. The selection of ROIs is based on structural or function features, depending on the motivation to perform the ROI analysis [6]. However, even if the ROIs have been determined, it is usually difficult to measure the signal within every ROI. Many imaging experts believe that in order to obtain the information with the best quality, fMRI and electroencephalography (EEG), a test that measures and records the electrical activity of the brain, should be used simultaneously. In this thesis, EEG was adopted in gathering the data. With the preprocessed data in hand, Statisticians have come up with various com- plex models to analyze the connectivity between pre-specified regions based on the ROI analysis and relevant information acquired. One popular model is the mixed effects vector autoregressive model (ME-VAR). 1.2 Data Collection and Description Anesthesiologists are interested in learning the change in brain activity under d- ifferent levels of anesthetic-induced unconsciousness. In this thesis, we analyze fMRI data from a real study using a mixed model to test whether and how different doses of anesthesia affect brain functional connectivity. Washington University medical school acquired and preprocessed the fMRI data by conducting the following experiment.
  • 7. Chapter 1. Introduction 3 Twenty five human volunteers were initially selected for the study. For each volun- teer, a facemask connected to a semi-closed anesthetic gas machine circuit was placed over the nose and mouth. Sevoflurane with various concentrations used for induction and maintenance of general anesthesia was delivered through this machine. Different sevoflurane inspired concentrations were implemented for corresponding targeted con- centrations. During these periods, BOLD-MRI signals were recorded. Meanwhile, EEG signals were recorded continuously to obtain information with higher quality. The re- sponse was designed as followed. Every 30 seconds, volunteers were instructed to respond to block-randomized record- ed auditory commands “Left” or “Right” by squeezing a pressure transducer in the cor- responding hand. The motor response was recorded and synchronized to vital signs, EEG waveforms, and MRI images. Responsiveness was defined as the appropriate re- sponse to command within the allotted 10 seconds of an auditory command, with the clinical endpoints of interest being loss of responsiveness (LOR) and return of respon- siveness (ROR). The focus of the study is on the neural signals recorded at anesthetic concentrations flanking LOR. The fMRI data have been preprocessed and twenty two subjects for analysis were selected from the results of a step that regressed out the global average MR signal during functional connectivity processing. This step was used to ”clean up” the data by removing putative physiologic sources of common noise in the signals. We set two specific goals for this study. The first goal is to estimate and test the significance of effective pairwise connectivity under different levels of anesthetic-induced unconsciousness. The second goal is to examine the changes in brain connectivity as the sevoflurane concentration increases.
  • 8. Chapter 2 The Mixed Effects Vector Autoregressive Model This section presents an introduction and discussion of the mixed effects vector au- toregressive model (ME-VAR). The model captures (i.) the expected condition-specific connectivity matrix (fixed effect) over different runs, (ii.) a subject-specific component (random effect) to model between-subject variation in connectivity, and (iii.) a run- specific component (random effect) to model the random effects of different runs nested within conditions. However, along with the comprehensive and complicated structure of the model come its shortcomings such as the large number of parameters needed to be estimated. When fitting the model in SAS, several issues are likely to appear. Some common problems include “insufficient memory” or “convergence criteria does not meet”. We performed a simplified version of the original model in SAS to overcome these problems based on feasibility and efficiency. 2.1 Setup Consider a brain network with R ROIs. This model is proposed to describe the BOLD response E(s)(t) (dimension R × 1) under different anaesthetic conditions of the resting state, that is, conditions with different sevoflurane concentrations. Our focus lies on the neural signals recorded at anesthetic concentrations flanking loss of responsive- ness (LOR), a concentration denoted as the clinical endpoint of interest. Five conditions were initially selected (labeled as condition 1 - condition 5). However, condition 4 and condition 5 have the same sevoflurane concentrations as condition 2 and condition 1 re- spectively. As such, we only focus on the first three conditions with different sevoflurane 4
  • 9. Chapter 2. The Mixed Effects Vector Autoregressive Model 5 concentrations. For a specific subject s, the model is written as E(s) (t) = P k=1 [β (s) 1j,kW (s) 1 + β (s) 2j,kW (s) 2 + β (s) 3j,kW (s) 3 ]E(s) (t − k) + (s) (t), (2.1) where t = P +1, ..., T are the time points up to time T and s = 1, ..., S denotes the subject with S subjects in total. The vector (s)(t) (dimension R × 1) is a white noise process with E( (s)(t)) = 0 and Cov( (s)) = Γ (dimension R×R). We assume that Cov( (s)) = Γ does not change over time and is constant across all participants. k = 1, ..., P denotes the time lag with maximum order P, and j represents the experimental run, which is nested within conditions. Other basic elements are more complicated and important. To understand the model in a more comprehensive way, we enumerate other detailed elements in the following. The subject-specific indicator parameter for condition i, denoted as W (s) i This parameter is used as a tool to examine brain connectivity across different conditions. Suppose that at time t0 a stimulus, a change of sevoflurane concentrations for condition i was presented; followed by a different sevoflurane concentrations, denoted as condition i at the time t1. We specify the function W (s) i as a time window on which the brain connectivity present is attributed to condition i in subject s. In this case, this time window takes a value of 1 in the interval [t0, t1] and takes the value 0 elsewhere. For example, if the anesthetic concentration in the interval [t0, t1] remains unchanged, then the corresponding W (s) i takes a value of 1. Optimum for the maximum lag order P The optimal order can be objectively selected using information theoretic criteria: Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC), and AIC with a correction for infinite small sample sizes (AICC). BIC penalizes model complexity more severely compared with the other two approaches. We take BIC as the desirable criterion to simplify the model in SAS for the complexity of the model in our analysis. We first use P = 3. If every parameter from every region turns out estimable with this maximum order, we choose both higher and lower maximum orders and compare their BIC. On the other hand, if some of the parameters cannot be estimated with P = 3, only lower maximum orders should be used. To sum up, the first priority for choosing the model is to guarantee that every single parameter we are interested can be estimated in SAS. Then, from these models, we select the one with the lowest BIC. The subject-specific direct connectivity matrix at lag k, denoted as β (s) ij,k The subject-specific connectivity matrix at lag k, denoted β (s) ij,k, includes a subject- specific random effect to describe the variability in the connectivity pattern across dif- ferent subjects. This R × R matrix quantifies exactly how E(s)(t − k) directly predicts
  • 10. Chapter 2. The Mixed Effects Vector Autoregressive Model 6 E(s)(t) under experimental condition i. It also includes an indicator j denoting repli- cates (runs) of the experiment nested in conditions. As a result, β (s) ij,k contains both random components and fixed components. To correctly specify the model in SAS, it is important to separate them, that is, to decompose β (s) ij,k into random and fixed components. Before decomposition, we demonstrate the parameters of the model in vector and matrix forms. Below is an example to illustrate the matrices and the vectors in the model under condition 1.        E(s)(t)(1) E(s)(t)(2) ... E(s)(t)(R)        = P k=1        β (s) 1j,k(1, 1) β (s) 1j,k(1, 2) · · · β (s) 1j,k(1, R) β (s) 1j,k(2, 1) β (s) 1j,k(2, 2) · · · β (s) 1j,k(2, R) ... ... ... ... β (s) 1j,k(R, 1) β (s) 1j,k(R, 2) · · · β (s) 1j,k(R, R)        ×        E(s)(t − 1)(1) E(s)(t − 1)(2) ... E(s)(t − 1)(R)        + (s) (t) (2.2) 2.2 Decomposition Modeling variation across replicates One important element in the connectivity matrix β (s) ij,k is the subscript j indicating that brain connectivity is allowed to vary over runs. Here, we decompose the connectivity matrix into two components. β (s) ij,k = β (s) i,k + φ (s) i(j),k, (2.3) where β (s) i,k is the condition-subject-specific connectivity matrix and φ (s) i(j),k is the run- specific effect nested within conditions or subjects. It models the deviation of each run from the common condition-subject-specific connectivity matrix associated to lag k. The elements of the matrix φ (s) i(j),k are assumed to be mutually independent normal random variables. φ (s) i(j),k(r, r ) ∼ N(0, σ2 k(r, r ) + σ 2 k (r, r )) (2.4) where σ2 k(r, r ) comes from the subject effect and σ 2 k (r, r ) comes from the run effect. Modeling variation across subjects Another important element in the connectivity matrix β (s) ij,k is the superscript (s) indicating that brain connectivity is allowed to vary over subjects. This random com- ponent can be separated from the connectivity. It allows us to further decompose β (s) i,k into a fixed component and a random component. β (s) ij,k = βi,k + b (s) k + φ (s) i(j),k, (2.5)
  • 11. Chapter 2. The Mixed Effects Vector Autoregressive Model 7 where βi,k is the expected condition-specific connectivity matrix and b (s) k is the subject- specific effect which models an individual subject’s deviation from the common population- specific connectivity matrix associated to lag k. We model the elements of the matrix b (s) k as identically distributed independent variables following normal distribution. b (s) k (r, r ) ∼ N(0, σ2 k(r, r )). (2.6) Decomposition of the ME-VAR model To illustrate the features of the model, we assume lag order to be P = 1 for simplicity so that conditional mean of the observed BOLD signal at time t depends only on the observations at the previous time point t − 1. The model for subject s is specified as E(s) (t) = [β (s) 1j,1W (s) 1 + β (s) 2j,1W (s) 2 + β (s) 3j,1W (s) 3 ]E(s) (t − 1) + e(s) (t). (2.7) From the decomposition derived, we can decompose the connectivity matrix into a ran- dom and two fixed components. The model for subject s becomes E(s) (t) =[(β1,1 + b (s) 1 + φ (s) 1(j),1)W (s) 1 + (β2,1 + b (s) 1 + φ (s) 2(j),1)W (s) 2 + (β3,1 + b (s) 1 + φ (s) 3(j),1)W (s) 3 ]E(s) (t − 1) + (s) (t). (2.8) However, in real study, even though common variance have been assumed across subjects and runs, the combined effects of all the factors may still paralyze the software. In this case, we may have to simplify the model by ignoring the run-specific connectivity matrix (random effect) and analyze the data for each replicate separately. For each run, the model then becomes E(s) (t) = P k=1 [β (s) 1,kW (s) 1 + β (s) 2,kW (s) 2 + β (s) 3,kW (s) 3 ]E(s) (t − k) + (s) (t). (2.9) Without the run-specific connectivity component, β (s) 1,k can only be decomposed into a random component and a fixed component; β (s) i,k = βi,k + b (s) k . (2.10) 2.3 Hypothesis testing and Effect Estimates As mentioned, there are two goals for our analysis. The first one is to estimate and test the significance of effective pairwise connectivity. This process contains 50 × 50 × 3
  • 12. Chapter 2. The Mixed Effects Vector Autoregressive Model 8 hypothesis testings. The other goal is to examine the changes in brain connectivity across conditions. The process also contains 50 × 50 × 3 hypothesis testings. Testing for significance of connectivity Our interest is to determine whether the past activity at ROI r during condition i can significantly explain the current activity at ROI r. Given For the model with maximum order P, the null and the alternative hypothesis can be written as H0 : βi,1(r, r ) = βi,2(r, r ) = ... = βi,P (r, r ) = 0, and H1 : At least one of the coefficients parameters are not 0. (2.11) βi,k(r, r ) denotes the expected value of β (s) ij,k(r, r ), which is the expectation of the random effect run indicated by the subscript j and the random effect subject indicated by the superscript (s). In this case, we can say that activity at ROI r “Granger causes” the activity at ROI r if the null hypothesis is rejected. That is, the activity at ROI r provides statistically significant information about future activity at ROI r. Testing for changes of connectivity across conditions Our second interest is to investigate the differences of brain connectivity across different conditions. As the sevoflurane concentration changes, we need to determine if the connectivity within networks linking the thalamus, frontal and parietal cortical areas change too. The null and the alternative hypothesis can be written as H0 : βi,k(r, r ) = βi ,k(r, r ), and H1 : βi,k(r, r ) = βi ,k(r, r ). (2.12) This hypothesis testing measures the difference between the expected causality effect of the brain activity at ROI r and lag order k on ROI r under condition i and that under condition i . If the null hypothesis is rejected, we can say that the brain connectivity between ROI r and ROI r under condition i is different from that under condition i . Both hypothesis testings contain multiple correlated testings. Therefore, it is nec- essary to control the expected proportion of incorrectly rejected null hypotheses. False Discovery Rate (FDR) is used to correct for multiple comparisons. Compared with Fam- ilywise Error Rate (FWER), FDR increases the power at the cost of increasing the rate of type 1 errors. It is particularly useful in the analysis using datasets with relative- ly small sample sizes and large numbers of variables being measured per sample. The threshold q-value for the FDR is 0.1.
  • 13. Chapter 3 Connectivity Analysis of fMRI Data This chapter presents the result of our study. We use statistical packages R and SAS 9.2 to perform the analysis. R is mainly used for structuring the dataset and graph- ing relevant plots, whereas SAS is the primary tool to analyze the dataset under proc MIXED. Under models that incorporate the run-specific connectivity matrix, SAS shows insufficient memory of the computer. Therefore, to guarantee that every parameter is estimable in SAS, we simplify the model by using the data from the first run. That is, we only fitted the model to the first replicate of the data. Moreover, models with 2 lags and 3 lags do not guarantee that every parameter is estimable. In this situation, we pick the model with 1 lag eventually when specifying it in SAS, regardless of the BIC. 3.1 The Granger Causality Test We test the significance of each element in the connectivity matrices βi,k to investi- gate the lag cross-dependence structure in each condition. k equals 1 here since we fitted the model with only 1 lag. To identify the effective connectivity in pairs of brain regions, we perform a Granger causality test, as discussed in Chapter 2. For each condition, a heatmap (Figure 3.1) containing information on the effect of all regions of interest(ROIs) is presented. Column ROIs are response ROIs. Row ROIs represent regions that help us predict the influence on the response ROIs. In other words, Column regions are the response in our model, whereas row regions are regions with a time lag (lag 1). Each heatmap in Figure 3.1 represents the connectivity between different ROIs under each condition. Black here denotes that the Granger causality is significant after FDR 9
  • 14. Chapter 3. Connectivity Analysis of fMRI Data 10 Figure 3.1: Granger Causality Heatmaps correction. As for those insignificant effects, we use grey and white colors. The color gets deeper as the connectivity gets stronger. We notice that the lag dependence of a ROI on its own region is always significant under any condition. Besides, there are 19 significant cross-dependence effects under condition 1, as shown in the heatmap. Under condition 2 with the increase of sevoflurane concentration, the number of significant cross-dependence effects reduces to 9, all of which are not the same ROIs as those shown in condition 1. As the sevoflurane concentration increases further, none of the cross- dependence effects becomes significant. The only significant lag dependence of a ROI is its own region. The details of the significant cross-dependence effects are listed in Appendix A. The overall pattern implies that as the concentration of sevoflurane goes
  • 15. Chapter 3. Connectivity Analysis of fMRI Data 11 up, the overall connectivity between two ROIs decreases. To improve the accuracy of our findings, we implement a test of changes in connectivity across different conditions. 3.2 Connectivity Changes Across Conditions To investigate whether there is a change in connectivity across conditions, we test if the two connectivity matrix βi,k and βi ,kare equal as the sevoflurane concentration changes. This procedure is implemented in SAS 9.2. We then used the “qvalue” package in R to calculate the corrected threshold for the p-values by converting them to q-values. It turns out that none of the q-values are significant in this case. As a result, none of the connectivity changes is actually significant. However, it is still valuable to dig out more information based on SAS outputs and heatmaps. Our focus lies primarily on the significant connectivity based on the Granger causality test discussed in the previous section. If the connectivity does not exist, it does not make much sense to discuss the changes in connectivity further. These heatmaps in Figure 3.2 represent the connectivity changes under different conditions. Darker colors indicate that the changes are larger. However, since none of the changes are significant, black does not mean the changes are significant. It only implies the changes are strong. The heatmaps help us to identify patterns. As shown, there is no explicit pattern in the first two heatmaps. However, with a more significant increase in the sevoflurane concentration, the connectivity changes become noticeable as shown in the third heatmap. The diagonal color gets deeper in general compared with the first two heatmaps. From the pattern, we guess that the changes in connectivity are in one direction - either the connectivity gets stronger when the sevoflurane concentration goes up, or it gets weaker. Figure 3.1 shows that the connectivity weakens with the increase in sevoflurane concentration. However, since our focus is on the significant connectivity, it is necessary to investigate those values to see the direction of connectivity changes. To find out the direction, we referred to the significant estimated values of the Granger causality effect under all conditions. From Figure 3.1, we can see no cross-dependence under condition 3. As a result, for each ROI, we only focused on the causality effects from itself with a time lag (lag 1). Appendix B shows the estimated effect of each region under each condition. From condition 1 to condition 2 as the sevoflurane concentration increases, the estimated ef- fects of the connectivity go up except for ROI36, ROI42, and ROI46. However, from condition 2 to condition 3, as the sevoflurane concentration increases further, the con- nectivity trend in general becomes vaguer, though the connectivity still weakens in total as shown in Figure 3.3. Each line represents a sevoflurane concentration across ROIs.
  • 16. Chapter 3. Connectivity Analysis of fMRI Data 12 Figure 3.2: Connectivity Changes Heatmaps The horizontal axis denotes ROIs, whereas the vertical axis denotes estimated effects for each corresponding ROIs.
  • 17. Chapter 3. Connectivity Analysis of fMRI Data 13 Figure 3.3: Connectivity Plot across ROIs
  • 18. Chapter 4 Conclusion In this thesis, we explored the effect of anesthesia on brain effective connectivity. We used a mixed effects vector autoregressive model (ME-VAR) to analyze a resting-state fMRI dataset. This model captures the expected condition-specific component (fixed effect), the subject specific component (random effect), and the run-specific component (random effect). Theoretically, this is a comprehensive model for the analysis of the resting-state FMRI data. However, when applying to large datasets with many ROIs, it exceeds the limit of SAS and hence simplification of the model is necessary. Based on feasibility, we removed the run-specific component from the model and only demon- strated the result from the first replicate of the dataset. Since evaluation or testing of a single item does not allow for item-to-item variation and may not represent the batch or process, the model without a run-specific matrix is likely to contain bias. Moreover, even with the simplified model and data information, the maximum lag order is restricted to 1 in SAS, which is another limit of the model in this analysis. Based on the heatmap, the result shows that 69 Granger causality effects are signif- icant under condition 1. 50 of them are effects within ROIs over time. Under condition 2, 59 Granger causality effects are significant. Among them, 50 effects within ROIs over time are still significant. As the sevoflurane concentration goes up further (condition 3), there are only 50 significant Granger causality effects, all of which are effects within ROIs over time. We then tested for the significance of the connectivity changes. It turns out that none of the changes is significant. Despite the insignificance presented, we can still find a trend that from condition 1 to condition 2 with the increase in sevoflurane concentration, the connectivity generally becomes weaker, with few exceptions. From condition 2 to condition 3, the direction of the changes become vaguer. However, the connectivity still gets weaker on aggregate. In the future study, the difference of sevoflurane concentration 14
  • 19. Chapter 4. Conclusion 15 among conditions can be more substantial in order for us to visualize the connectivity changes better.
  • 20. Appendix A Significant Cross-ROI Effects condition1 condition2 condition3 ROI01→ROI15 ROI06→ROI49 ROI19→ROI49 ROI07→ROI49 ROI10→ROI49 ROI43→ROI36 ROI28→ROI12 ROI44→ROI36 ROI12→ROI49 ROI29→ROI14 ROI30→ROI36 ROI46→ROI36 ROI31→ROI36 ROI38→ROI31 ROI41→ROI31 ROI36→ROI38 ROI37→ROI36 ROI36→ROI48 ROI42→ROI31 ROI48→ROI32 ROI38→ROI36 ROI40→ROI44 ROI39→ROI36 ROI49→ROI17 ROI37→ROI48 ROI40→ROI36 ROI38→ROI48 ROI40→ROI48 16
  • 21. Appendix B Estimated Effects of the Granger Causality Testings ROIs condition1 condition2 condition3 ROIs condition1 condition2 condition3 ROI1 0.7952652 0.7406001 0.7209555 ROI26 0.7709983 0.6702235 0.6885903 ROI2 0.7169583 0.6979163 0.5990974 ROI27 0.7894696 0.7809740 0.6930602 ROI3 0.8088205 0.7619170 0.6537976 ROI28 0.7304779 0.7205294 0.7228392 ROI4 0.8199955 0.7108279 0.6914172 ROI29 0.7273849 0.7055247 0.6709614 ROI5 0.6966409 0.6447304 0.6453244 ROI30 0.7810669 0.7444860 0.7055461 ROI6 0.7178230 0.7269425 0.6152923 ROI31 0.6942039 0.6709644 0.6105873 ROI7 0.7428886 0.6928639 0.7060899 ROI32 0.7216514 0.7042265 0.6398730 ROI8 0.7533285 0.7180890 0.6002316 ROI33 0.7076750 0.6858878 0.6270619 ROI9 0.7448033 0.6964896 0.6999977 ROI34 0.6417452 0.5972246 0.6158163 ROI10 0.7405824 0.7127173 0.7392774 ROI35 0.6354104 0.6075662 0.5722229 ROI11 0.7403181 0.7103553 0.6728991 ROI36 0.4412667 0.5161668 0.5357359 ROI12 0.7315507 0.7362889 0.6877589 ROI37 0.7790736 0.7009024 0.6537599 ROI13 0.7367875 0.6523549 0.6761815 ROI38 0.7876667 0.7411852 0.7356888 ROI14 0.7412977 0.6965885 0.6984047 ROI39 0.6479298 0.6074324 0.6146055 ROI15 0.7366455 0.7094993 0.6603111 ROI40 0.6345684 0.5881167 0.6609347 ROI16 0.7049970 0.6632437 0.6330410 ROI41 0.7285654 0.6526174 0.6368883 ROI17 0.7710784 0.6583128 0.5939490 ROI42 0.6411857 0.6785500 0.6836995 ROI18 0.7503926 0.7099066 0.6509170 ROI43 0.6711933 0.6094010 0.6600454 ROI19 0.7755234 0.7190092 0.7280341 ROI44 0.6931007 0.6377876 0.5954626 ROI20 0.7672113 0.7196823 0.6337438 ROI45 0.7214215 0.7022580 0.7021112 ROI21 0.6921319 0.6425770 0.5799561 ROI46 0.6991743 0.7237863 0.7231898 ROI22 0.7276675 0.6848949 0.6267191 ROI47 0.7591686 0.6389944 0.6334283 ROI23 0.7499073 0.6820988 0.6768443 ROI48 0.7615498 0.6183721 0.6661121 ROI24 0.7289385 0.6023442 0.6239701 ROI49 0.6787235 0.6522500 0.6039819 ROI25 0.7416147 0.6635005 0.6665899 ROI50 0.7133189 0.7122303 0.5711190 17
  • 22. Bibliography [1] Cristina Gorrostieta, Hernando Ombao, Patrick Bdard b, and Jerome N. Sanes. Investigating brain connectivity using mixed effects vector autoregressive models. NeuroImage, 59:3347 – 3355, 2012. doi: 10.1016/j.neuroimage.2011.08.115. [2] Nicole A. Lazar, William F. Eddy, Christopher R. Genovese, and Joel Welling. Sta- tistical issues in fmri for brain imaging. International Statistical Review, 69:105127, 2001. doi: 10.1111/j.1751-5823.2001.tb00482.x. [3] Martin A. Lindquist. The statistical analysis of fmri data. Statistical Science, 23: 439 – 464, 2008. doi: 10.1214/09-STS282. [4] J. S. Damoiseaux, S. A. R. B. Rombouts, F. Barkhof, P. Scheltens, C. J. Stam, S. M. Smith, and C. F. Beckmann. Consistent resting-state networks across healthy subjects. PNAS, 103:1384813853, 2006. doi: 10.1073/pnas.0601417103. [5] Karl J. Friston. Functional and effective connectivity: A review. Brain Connectivity, 1:13 – 36, 2011. doi: 10.1089/brain.2011.0008. [6] Russell A. Poldrack. Region of interest analysis for fmri. Soc Cogn Affect Neurosci, 2:6770, 2007. doi: 10.1093/scan/nsm006. 18