Functional Magnetic Resonance Imaging Processing

229 Pages·2014·5.653 MB·English

by  Xingfeng Li (auth.)

Xingfeng Li
Functional
Magnetic
Resonance
Imaging
Processing
Functional Magnetic Resonance Imaging Processing

Xingfeng Li
Functional Magnetic
Resonance Imaging
Processing
123
XingfengLi
IntelligentSystemsResearchCentre
UniversityofUlster
Londonderry
UK
ISBN978-94-007-7301-1 ISBN978-94-007-7302-8(eBook)
DOI10.1007/978-94-007-7302-8
SpringerDordrechtHeidelbergNewYorkLondon
LibraryofCongressControlNumber:2013949044
©SpringerScience+BusinessMediaDordrecht2014
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof
thematerialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,
broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation
storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodology
nowknownorhereafterdeveloped.Exemptedfromthislegalreservationarebriefexcerptsinconnection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’slocation,initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.
PermissionsforusemaybeobtainedthroughRightsLinkattheCopyrightClearanceCenter.Violations
areliabletoprosecutionundertherespectiveCopyrightLaw.
Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication
doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant
protectivelawsandregulationsandthereforefreeforgeneraluse.
While the advice and information in this book are believed to be true and accurate at the date of
publication,neithertheauthorsnortheeditorsnorthepublishercanacceptanylegalresponsibilityfor
anyerrorsoromissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,with
respecttothematerialcontainedherein.
Printedonacid-freepaper
SpringerispartofSpringerScience+BusinessMedia(www.springer.com)
To mymotherYufangZhang andmyfather
YongkangLi

Preface
This book is about how to analyze perfusion-weightedimaging (PWI), functional
MRI (fMRI), diffusion-weightedimaging(DWI),and structuralMRI (sMRI) data
for investigating brain functions. Broadly speaking, there are two approaches to
studybrainfunctionin vivo:oneisbolusinjectionmethodandtheotherisnonin-
vasive(noinjections)method,e.g.,blood-oxygen-level-dependent(BOLD)contrast
method.For the tracer injection method,we will introducedynamicsusceptibility
contrastimaging(DSC-MRI),whichwefocusontheDiracdeltafunction(impulse
function)asaninputforstudyingthebrainbloodflowsystem.Thebasicindicator
theory is explained in details. Both linear and nonlinear regression methods are
employed to smooth the DSC-MRI concentration time course. To solve the ill-
posed problem for the residual function estimation, weighted damping method,
i.e., Levenberg–Marquardt(LM) algorithm is introduced to solve Toeplitz matrix
regularizationproblem.Cerebralbloodflowparametersarethenestimatedbasedon
theindicatortheory.
BOLD-fMRI processing is the main part of this book; this includes both acti-
vation detection(segmentationview of the brain)and effectiveconnectivitystudy
(integration view of the brain). We begin with the first-level activation detection
analysis, and we introduced the generalized linear model with autoregression
model for error correction in activation detection. The threshold correction for
the activation map using false discoveryrate (FDR) and family-wise error (FWE)
is introduced subsequently. Then mixed model is presented for the second-level
analysis. To calculate the regression parameters, i.e., variance in the mixed effect
model,Newton–Raphson(NR),LM,andtrustregionmethodsaregiveninChap.3.
In recentyears,there hasbeenincreasinginterestin studyingeffectiveconnec-
tivity using fMRI method. Generally speaking, there are three methods to study
effective connectivityfrom the viewpoint of system identification, i.e., black-box,
gray-box,and white-box methods. Since the black-boxmethod is model free and
easy to apply, we concentrate on introducing this method. This includes model
selectionforfirst-levelandrobustregressionforsecond-leveleffectiveconnectivity
analysis.Wealsoapplythismethodforresting-statefMRIstudy.
vii
viii Preface
The third part of this book is about processing diffusion-weighted imaging
(DWI). The basic principle of MRI diffusion imaging is to study the motion of
watermolecules.Thefirstconceptistheapparentdiffusioncoefficient(ADC)which
quantifies the magnitude of the water diffusion on one dimension. Because water
diffusion is really 3D processing, diffusion tensor imaging (DTI) is introduced to
describethismotion.Basedonthisinformation,wecaninferthefiberdirectionsin
thehumanbrain.ButDTImethodcannotresolvetheproblemofcrossingfiberissue;
to circumvent this limitation, high angle resolution diffusion imaging (HARDI)
was proposed, and Q-ball imaging (QBI) and diffusion spectrum imaging (DSI)
havebeendevelopedforstudyingdiffusionorientationmap.Toestimateorientation
distribution function (ODF) from QBI/DSI, regularization methods need to be
adopted.Weintroducethecommonlyusedmethod,i.e.,generalizedcrossvalidation
(GCV)methodforODFregularization.
Finally, sMRI data analysis method is presented. Instead of concentrating on
sMRI image segmentation and registration, we present voxel-based morphometry
(VBM) method and its application to Alzheimer’s disease (AD) study. To begin
with,wegivetheprocessingstepsforVBManalysisbasedoncross-sectionalstudy,
andthenweprovidedlongitudinalVBMforsMRIdataanalysis.Furthermore,asan
example,weapplythismethodtoADstudytodemonstratehowtousethismethod.
Based on longitudinal VBM, we investigate the causality relationship between
differentbrainregionsatdifferentstagesofdiseaseprogression.
I assume the reader has a certain background in computer programming,
numericalanalysis, statistics, and medical image analysis. This book can be used
forgraduatestudentswhoareinterestedinstudyingmedicalimageanalysis,partic-
ularlyfMRIimageanalysis.Itcanalsobeusedasareferencebookforradiologists,
psychologists, neurologists, medical image physicists, computer scientists, and
biomedicalengineersforstudyingMRIimageprocessing.
I wish to thankProf. RobertHess and Prof.Kathy MullenfromMcGill Vision
Research, McGill University, Canada, for providing a huge amount of fMRI
data for this book. I appreciate Dr. Arun L. W. Bokde and Dr. Elizabeth Kehoe
from Cognitive Systems group, Trinity College Dublin, Ireland, for collecting
resting-statefMRI andemotionalfacialexperimentalfMRIdata.IrecognizeProf.
Cyril Pouponfrom NeuroSpin, France, and Dr. Jennifer Campbell from Montreal
Neurological Institute, McGill University, Canada, for providing the human and
biologicalratQBIdatasets.IamobligatedtoProf.StefanTeipelandDr.Maximilian
LerchefromtheUniversityofRostockinGermanytoallowmetousetheirDSIdata.
Furthermore,IamespeciallygratefultoProf.HabibBenaliandDr.ArnaudMesse
fromfunctionalimaginglaboratory,INSERM/UPMC,Paris6thUniversity,France,
fortheirhelpindevelopingthemethodforprocessingQBI.Finally,Iacknowledge
Prof. Thomas Martin McGinnity from University of Ulster, UK, for helping my
work.Mostof all, I am indebtedto mywife, Feijun Wang, forherunderstanding,
patience,andsupportforwritingthisbook.
Londonderry,UK XingfengLi
Contents
1 MRIPerfusion-WeightedImagingAnalysis .............................. 1
1.1 PerfusionImaging....................................................... 2
1.1.1 Indicator–DilutionTheoryforDSC-MRI ..................... 3
1.1.2 MTTandCBVCalculation .................................... 6
1.1.3 DSC-MRITimeSeriesAnalysis............................... 8
1.2 Gamma-VariateFitting.................................................. 11
1.2.1 LinearRegressionMethodforGamma-VariateFitting....... 11
1.2.2 NonlinearRegressionMethodforGamma-VariateFitting ... 13
1.2.3 BaselineEliminationforGamma-VariateFitting ............. 16
1.2.4 Linear Method and Nonlinear Method
forGamma-VariateFitting ..................................... 18
1.3 AIFSelection............................................................ 19
1.3.1 RobustMethodforAIFDetermination........................ 20
1.3.2 DeconvolutionCalculationandResidualFunction
Estimation....................................................... 22
1.3.3 SVDMethodforDeconvolution............................... 24
1.3.4 L2NormRegularizationforPWIStudy....................... 25
1.3.5 PiecewiseLinearMethodforRidgeRegression
ParameterEstimation........................................... 26
1.3.6 CBF,MTT,CBV,ArrivalTime,andT-maxMaps ............ 29
1.4 DispersionEffectsinDSC-MRI........................................ 32
1.4.1 LocalDensityRandomWalkforConcentrationTimeCourse 32
1.4.2 ConvolutionMethodtoStudyDisperseEffect................ 33
1.5 SummaryofthePWIAlgorithm ....................................... 33
References..................................................................... 35
2 First-LevelfMRIDataAnalysisforActivationDetection............... 39
2.1 fMRIExperimentalDesign............................................. 40
2.1.1 BlockDesign.................................................... 41
2.1.2 RandomERDesign............................................. 42
2.1.3 Phase-EncodedDesign ......................................... 44
ix