PDF Brain Network Analysis 1St Edition Moo K Chung Ebook Full Chapter
PDF Brain Network Analysis 1St Edition Moo K Chung Ebook Full Chapter
PDF Brain Network Analysis 1St Edition Moo K Chung Ebook Full Chapter
K. Chung
Visit to download the full and correct content document:
https://textbookfull.com/product/brain-network-analysis-1st-edition-moo-k-chung/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
https://textbookfull.com/product/fundamentals-of-brain-network-
analysis-1st-edition-bullmore/
https://textbookfull.com/product/social-network-analysis-applied-
to-team-sports-analysis-1st-edition-filipe-manuel-clemente/
https://textbookfull.com/product/heterogeneous-information-
network-analysis-and-applications-1st-edition-chuan-shi/
https://textbookfull.com/product/overtime-moo-u-5-1st-edition-
kat-mizera-heart-eyes-press/
Social Network Analysis in Second Language Research 1st
Edition Kennedy Terry
https://textbookfull.com/product/social-network-analysis-in-
second-language-research-1st-edition-kennedy-terry/
https://textbookfull.com/product/brain-source-localization-using-
eeg-signal-analysis-1st-edition-munsif-ali-jatoi/
https://textbookfull.com/product/microarray-bioprinting-
technology-fundamentals-and-practices-1st-edition-moo-yeal-lee-
eds/
https://textbookfull.com/product/social-network-analysis-and-law-
enforcement-applications-for-intelligence-analysis-morgan-
burcher/
https://textbookfull.com/product/social-network-analysis-
interdisciplinary-approaches-and-case-studies-1st-edition-
xiaoming-fu/
Brain Network Analysis
This tutorial reference serves as a coherent overview of various statistical and math-
ematical approaches used in brain network analysis, where modeling the complex
structures and functions of the human brain often poses many unique computational
and statistical challenges. This book fills a gap as a textbook for graduate students while
simultaneously articulating important and technically challenging topics. Whereas most
available books are graph theory centric, this text introduces techniques arising from
graph theory and expands to include other advanced models in its discussion on
network science, regression, and algebraic topology. Links are included to the sample
data and codes used in generating the book’s results and figures, helping empower
methodological understanding in a manner immediately usable to both researchers and
students.
M o o K . C h u n g is an Associate Professor in the Department of Biostatistics and
Medical Informatics at the University of Wisconsin–Madison and is also affiliated with
the Department of Statistics and Waisman Laboratory for Brain Imaging and Behavior.
He has received the Vilas Associate Award for his research in applied topology to
medical imaging, the Editor’s Award for best paper published in the Journal of Speech,
Language, and Hearing Research for a paper that analyzed computed tomography (CT)
images, and a National Institutes of Health (NIH) Brain Initiative Award for work
on persistent homological brain network analysis. He has written numerous papers in
computational neuroimaging and two previous books on computation on brain image
analysis.
Brain Network Analysis
MOO K. CHUNG
University of Wisconsin–Madison
University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India
79 Anson Road, #06-04/06, Singapore 079906
www.cambridge.org
Information on this title: www.cambridge.org/9781107184862
DOI: 10.1017/9781316882610
© Moo K. Chung 2019
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2019
Printed in the United Kingdom by TJ International Ltd., Padstow, Cornwall
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Chung, Moo K., author.
Title: Brain network analysis / Moo K. Chung.
Description: Cambridge, United Kingdom ; New York, NY : Cambridge University
Press, 2019. | Includes bibliographical references and index.
Identifiers: LCCN 2019000330 | ISBN 9781107184862 (hardback : alk. paper)
Subjects: | MESH: Connectome–methods | Brain–physiology | Nerve
Net–physiology | Neural Pathways–physiology | Models, Neurological |
Mathematical Computing
Classification: LCC QP376 | NLM WL 141.5.B6 | DDC 612.8/2–dc23
LC record available at https://lccn.loc.gov/2019000330
ISBN 978-1-107-18486-2 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
To my parents
for their endless love
Contents
Preface page xi
1 Statistical Preliminary 1
1.1 General Linear Models 1
1.2 Logistic Regression 6
1.3 Random Fields 9
1.4 Statistical Inference on Fields 16
2 Brain Network Nodes and Edges 27
2.1 Brain Templates 27
2.2 Brain Parcellations 28
2.3 Deterministic Connectivity 34
2.4 Probabilistic Connectivity 46
2.5 Parcellation-Free Brain Network 50
2.6 Structural Covariates 55
3 Graph Theory 61
3.1 Trees and Graphs 61
3.2 Minimum Spanning Trees 62
3.3 Node Degree 65
3.4 Shortest Path Length 70
3.5 Clustering Coefficient 71
3.6 Small-Worldness 72
3.7 Fractal Dimension 73
4 Correlation Networks 76
4.1 Pearson Correlations 76
4.2 Partial Correlations 78
4.3 Averaging Correlations 79
vii
viii Contents
Bibliography 302
Index 326
Preface
xi
xii Preface
brain imaging data used in illustrating the methods. Hyekyoung Lee of Seoul
National University and Yuan Wang of University of South Carolina helped me
write chapters related to persistent homology and topological distances. Her-
nando Ombao of the King Abdullah University of Science and Technology and
Dustin Pluta of the University of California–Irvine helped me write chapters
related to the dynamic network models. Andrey Gritsenko of the University of
Wisconsin–Madison performed some of basic image processing on the resting-
state fMRI from Human Connectome Project data and helped compile the
list of Automatic Anatomical Labeling (AAL) parcellation. Although most
figures are produced by myself using MATLAB, some figures are generated
by my current and former students, postdocs, and colleagues. Such figures
are identified in figure captions and the proper credits are given. I am also
indebted to Fred Boehm of the University of Wisconsin–Madison and Feng
Liu of Harvard University for proofreading a few chapters.
1
Statistical Preliminary
This chapter covers the basic statistical methods that are mostly used in univari-
ate voxel-level approaches. However, these basic methods are equally useful
in brain network analysis as well. Most of network modeling techniques are
based on the voxel-level methods. Readers familiar with univariate statistical
methods can skip this chapter.
1
2 Statistical Preliminary
mainly estimated by the least squares estimation and have been implemented
in many statistical packages such as R1 (Pinehiro and Bates, 2002), statistical
parametric mapping (SPM)2 and fMRI-STAT.3
We assume there are n subjects. Let yi be the response variable at a node
or edge, which is mainly coming from images and xi = (xi1, · · · ,xip ) to
be the variables of interest and zi = (zi1, · · · ,zik ) to be nuisance variables
corresponding to the ith subject. Then we have GLM
yi = zi λ + xi β + i ,
where λ = (λ1, · · · ,λk ) and β = (β1, · · · ,βp ) are unknown parameter
vectors to be estimated. We assume to be the usual zero mean Gaussian
noise.
The significance of the variable of interests xi is determined by testing the
null hypothesis
H0 : β = 0 vs. H1 : β = 0.
The fit of the reduced model corresponding to β = 0, i.e.,
yi = zi λ, (1.1)
is measured by the sum of the squared errors (SSE):
n
SSE0 = (yi − zi
λ0 )2,
i=1
where λ0 is the least squares estimation obtained from the reduced model. The
reduced model (1.1) can be written in a matrix form
⎛ ⎞ ⎛ ⎞⎛ ⎞
y1 z11 · · · z1k λ1
⎜ .. ⎟ ⎜ .. .. .. ⎟ ⎜ .. ⎟ .
⎝ . ⎠=⎝ . . . ⎠⎝ . ⎠
yn zn1 ··· znk λn
y Z λ
yi = zi λ + xi β
is measured by
n
SSE1 = λ1 − xi
(yi − zi β 1 )2 ,
i=1
where λ1 and β 1 are the least squares estimation from the full model. The full
model can be written in a matrix form by concatenating the row vectors zi and
xi into a larger row vector (zi ,xi ), and the column vectors λ and β into a larger
column vector (λ,β ) , i.e.,
λ
yi = (zi ,xi ) .
β
Then the parameters of the full model can be estimated in the least squares
fashion. Note that
n
SSE1 = min (yi − zi λ1 − xi β 1 )2
λ1,β 1
i=1
n
≤ min (yi − zi λ0 )2 = SSE0 .
λ0
i=1
So the larger the value of SSE0 − SSE1 , more significant the contribution of
the coefficients β is. Under the assumption of the null hypothesis H0 , the test
statistic is the ratio
(SSE0 − SSE1 )/p
F = ∼ Fp,n−p−k . (1.2)
SSE0 /(n − p − k)
The larger the F value, it is more unlikely to accept H0 .
1.1.1 T-Statistic
When p = 1, the test statistic F is distributed as F1,n−1−k , which is the square
of the student t-distribution with n − 1 − k degrees of freedom, i.e., tn−1−k
2 . In
this case, it is better to use t-statistic. The advantage of using the t-statistic is
that the test statistic can provide the direction of the group difference that the
F -statistic cannot provide.
Let
c = (0, · · · ,0 ,1, 0, · · · ,0)
k p−1
4 Statistical Preliminary
1.1.2 R-Square
The R-square of a model explains the proportion of variability in measurement
that is accounted by the model. Sometime R-square is called the coefficient
of determination and it is given as the square of a correlation coefficient for a
very simple model. For a linear model involving the response variable yi , the
total sum of squares (SST) measures total total variation in response yi and is
defined as
n
SST = (yi − ȳ)2,
i=1
If the statistics t j are all independent, since t j are close to standard normal,
E(t i t j ) ≈ 0. The dependency increases the variance estimate and reduces
the aggregated t-statistic value. Unfortunately, it is difficult to estimate the
correlations directly since only one t-statistic map is available for each t j .
E(t i t j ) can be empirically estimated by computing correlations over the entries
of t-statistic maps t i and t j (see Figure 1.1).
Ei = 0, Vj = σ 2 .
However, linear model (1.3) is no longer appropriate since
EYj = πi = x
i β
but xi β may not be in the range [0,1]. The inconsistency is caused by trying
to match continuous variables xij to categorical variable Yi directly. To address
this problem, we introduce the logistic regression function g:
1.2 Logistic Regression 7
exp(x
i βi )
πi = g(xi ) = . (1.4)
1 + exp(x
i βi )
Using the logit function, we can write (1.4) as
πi
logit(πi ) = log = x
i βi .
1 − πi
∂ 2 log L(β)
n
I (β) = − − πi (1 − πi )xi x
i .
∂β ∂β
i=1
8 Statistical Preliminary
T is a mean zero Gaussian field if ET (x) = 0 for all x. Because any mean
zero multivariate normal distribution can be completely characterized by its
covariance matrix, a mean zero Gaussian random field T can be similarly
determined by its covariance function R. Two fields T and S are independent
if T (x) and S(y) are independent for every x and y. For mean zero Gaussian
fields T and S, they are independent if and only if the covariance function
R(x,y) = E T (x)T (y)
We will denote the mean-square convergence using the usual limit notation:
lim Th = T .
h→0
The convergence in mean square implies the convergence in mean. This can be
seen from
2 2 2
ETh − T = V Th − T + E|Th − T | .
Now let Th → T in mean square. Each term in the right-hand side should also
converge to zero, proving the statement.
Now we define the derivative of the field in the mean-square sense as
dT (x) T (x + h) − T (x)
= lim .
dx h→0 h
12 Statistical Preliminary
the limit of the usual isotropic Gaussian kernel. White noise is usually
characterized via generalized functions. One example of white noise is the
generalized derivative of Brownian motion (Wiener process) called Gaussian
white noise.
RB (x,y) = min(x,y).
The Gaussian white noise can be used to construct smooth Gaussian random
fields of the form
dB(x)
X(x) = K ∗ W (x) = K ∗ ,
dx
14 Statistical Preliminary
sigma=1
K=K(sigma,dx,dy)/sum(sum(K(sigma,dx,dy)))
w=normrnd(0,0.4,101,101)
smooth_w=w
for i=1:10
figure; imagesc(smooth_w)
smooth_w=conv2(smooth_w,K,’same’)
end;
Figure 1.2 Gaussian random field simulation. Starting with Gaussian white noise
N (0,0.42 ) (top-left), we iteratively apply Gaussian kernel smoothing one, four,
and nine times with bandwidth σ = 1.
Another random document with
no related content on Scribd:
The strength of our political organizations arises from their
development as social bodies; many of the hardest workers in their
ranks are neither office-holders nor yet paid henchmen, but merely
members who have gradually learned to identify their fortunes with
the party whose hall they have come to regard as the head-quarters
in which to spend the most agreeable of their leisure moments.
Under the American system it is impossible for a man to accomplish
anything by himself; he must associate himself with others, and they
must throw their weight together. This is just what the social
functions of the political clubs enable their members to do. The great
and rich society clubs are composed of men who are not apt to take
much interest in politics anyhow, and never act as a body. The great
effect produced by a social organization for political purposes is
shown by the career of the Union League Club; and equally striking
proof can be seen by every man who attends a ward meeting. There
is thus, however much to be regretted it may be, a constant
tendency towards the concentration of political power in the hands of
those men who by taste and education are fitted to enjoy the social
side of the various political organizations.
BOSS METHODS.
A politician who becomes an influential local leader or boss is, of
course, always one with a genuine talent for intrigue and
organization. He owes much of his power to the rewards he is able
to dispense. Not only does he procure for his supporters positions in
the service of the State or city,—as in the custom-house, sheriff’s
office, etc.,—but he is also able to procure positions for many on
horse railroads, the elevated roads, quarry works, etc. Great
corporations are peculiarly subject to the attacks of demagogues,
and they find it much to their interest to be on good terms with the
leader in each district who controls the vote of the Assemblyman and
Alderman; and therefore the former is pretty sure that a letter of
recommendation from him on behalf of any applicant for work will
receive most favorable consideration. The leader is also continually
helping his henchmen out of difficulties, pecuniary and otherwise; he
lends them a dollar or two now and then, helps out, when possible,
such of their kinsmen as get into the clutches of the law, gets a hold
over such of them as have done wrong and are afraid of being
exposed, and learns to mix judicious bullying with the rendering of
service.
But, in addition to all this, the boss owes very much of his
commanding influence to his social relations with various bodies of
his constituents; and it is his work as well as his pleasure to keep up
these relations. No débutante during her first winter in society has a
more exacting round of social duties to perform than has a prominent
ward politician. In every ward there are numerous organizations,
primarily social in character, but capable of being turned to good
account politically. The Amalgamated Hack-drivers’ Union, the
Hibernian Republican Club, the West Side Young Democrats, the
Jefferson C. Mullin Picnic Association,—there are twenty such
bodies as these in every district, and with, at any rate, the master
spirits in each and all it is necessary for the boss to keep on terms of
intimate and, indeed, rather boisterous friendship. When the
Jefferson C. Mullin society goes on a picnic, the average citizen
scrupulously avoids its neighborhood; but the boss goes, perhaps
with his wife, and, moreover, enjoys himself heartily, and is hail-
fellow-well-met with the rest of the picnickers, who, by the way, may
be by no means bad fellows; and when election day comes round,
the latter, in return, no matter to what party they may nominally
belong, enthusiastically support their friend and guest, on social, not
political, grounds. The boss knows every man in his district who can
control any number of votes: an influential saloon-keeper, the owner
of a large livery stable, the leader among a set of horse-car drivers, a
foreman in a machine-shop who has a taste for politics,—with all
alike he keeps up constant and friendly relations. Of course this fact
does not of itself make the boss a bad man; there are several such I
could point out who are ten times over better fellows than are the
mild-mannered scholars of timorous virtue who criticise them. But on
the whole the qualities tending to make a man a successful local
political leader under our present conditions are not apt to be
qualities that make him serve the public honestly or disinterestedly;
and in the lower wards, where there is a large vicious population, the
condition of politics is often fairly appalling, and the boss of the
dominant party is generally a man of grossly immoral public and
private character, as anyone can satisfy himself by examining the
testimony taken by the last two or three legislative committees that
have investigated the affairs of New York city. In some of these
wards many of the social organizations with which the leaders are
obliged to keep on good terms are composed of criminals, or of the
relatives and associates of criminals. The testimony mentioned
above showed some strange things. I will take at random a few
instances that occur to me at the moment. There was one case of an
assemblyman who served several terms in the Legislature, while his
private business was to carry on corrupt negotiations between the
Excise Commissioners and owners of low haunts who wished
licenses. The president of a powerful semi-political association was
by profession a burglar; the man who received the goods he stole
was an alderman. Another alderman was elected while his hair was
still short from a term in State Prison. A school trustee had been
convicted of embezzlement, and was the associate of criminals. A
prominent official in the Police Department was interested in
disreputable houses and gambling saloons, and was backed
politically by their proprietors.
CONCLUSIONS.
In conclusion, it may be accepted as a fact, however unpleasant,
that if steady work and much attention to detail are required, ordinary
citizens, to whom participation in politics is merely a disagreeable
duty, will always be beaten by the organized army of politicians to
whom it is both duty, business, and pleasure, and who are knit
together and to outsiders by their social relations. On the other hand,
average citizens do take a spasmodic interest in public affairs; and
we should therefore so shape our governmental system that the
action required by the voters should be as simple and direct as
possible, and should not need to be taken any more often than is
necessary. Governmental power should be concentrated in the
hands of a very few men, who would be so conspicuous that no
citizen could help knowing all about them; and the elections should
not come too frequently. Not one decent voter in ten will take the
trouble annually to inform himself as to the character of the host of
petty candidates to be balloted for, but he will be sure to know all
about the mayor, comptroller, etc. It is not to his credit that we can
only rely, and that without much certainty, upon his taking a
spasmodic interest in the government that affects his own well-being;
but such is the case, and accordingly we ought, as far as possible, to
have a system requiring on his part intermittent and not sustained
action.
FOOTNOTES:
[9] The Century, November, 1886.
[10] Since succeeded every year or two by some other anti-
Tammany Democratic organization or organizations.
[11] All this has been changed, vastly for the better, by the
ballot reform laws, under which the State distributes the printed
ballots; and elections are now much more honest than formerly.
[12] As a consequence of our investigation the committee, of
which I was chairman, succeeded in securing the enactment of
laws which abolished these enormous salaries.
VII
SIX YEARS OF CIVIL SERVICE REFORM[13]