Towards an accurate brain tractography

190 Pages·2012·46.83 MB·English

by  Eleftherios Garyfallidis

TOWARDS AN ACCURATE BRAIN
TRACTOGRAPHY
ELEFTHERIOS GARYFALLIDIS
Wolfson College
University of Cambridge
United Kingdom
May 2012
AdissertationsubmittedtotheUniversityofCambridgeforthedegree
ofDoctorofPhilosophy
Supervisors
DR. IAN NIMMO-SMITH
MedicalResearchCouncil
CognitionandBrainSciencesUnit
Cambridge,UK
DR. GUY B. WILLIAMS
WolfsonBrainImagingCentre
DepartmentofClinicalNeurosciences
UniversityofCambridge
Cambridge,UK
i
Dedication
To my parents
STAVROS AND AMALIA
forhiscretivityandherpatience.
Ineededbothtoexcel.
To the giants of
OPEN SOURCE SOFTWARE
whopushedtheworldforward.
ii
Acknowledgements
This PhD was funded by the Engineering and Physical Sciences Research
Council, the Vergottis Foundation, the Board of Graduate Studies of the
UniversityofCambridge,theMedicalResearchCouncilandWolfsonCol-
lege,Cambridge,UnitedKingdom.
I would like to thank my supervisor, Dr. Ian Nimmo-Smith, for the
excellent guidance and advice throughout the entire course of my PhD.
He has been an endless source of inspiration and support for my work.
I am also thankful to my second supervisor Dr. Guy Williams for his re-
searchsuggestionsandencouragement. Thecombinationofthestatistical
knowledge from my first supervisor and the knowledge in MR physics
from my second supervisor was a key component for the completion of
thisthesis.
I am grateful to Dr. Matthew Brett who had been such a great friend
andpromoterofmywork. Hissuggestionshavebeenenormouslyhelpful
especially on software development. My great appreciation to my friend
and collaborator Dr. Marta Correia who helped with designing and col-
lecting the MR data sets used in this work. The same goes for dearest Mr.
John Griffithsand Dr. VirginiaNewcombe whowere the firstusers ofmy
workinCambridge.
AbigthankyoutoStephanGerhardwithwhomIworkedtogivebirth
to a new visualization library, to Frank Yeh for his suggestions on trac-
tography algorithms and to Dr. Emanuele Olivetti on Machine Learning
approaches.
MygreatrespecttoDr. DavidJarvisAlumniDeanofWolfsonCollege,
Professor Mike Proctor and Professor Peter Haynes from the Department
ofAppliedMathematicsandTheoreticalPhysicswhohelpedmecontinue
my career in research and supported me at the most difficult moments of
my life. My gratitude also to Dr. Rik Henson who introduced me to my
supervisorsatmycurrentdepartment.
Lifewouldbeunbearablewithoutfriends. Muchloveandgratitudeto
my great friends from my homeland: Vassilis Dimitriadis, Dimitris Prit-
sos, Alexandros Triantafyllidis, Dr. Periklis Akritidis, Iason Oikonomidis
and Dr. Vassilis Tsiaras. And the great friends I made here in Cambridge:
Dr. Euan Spence, Dr. Ian Charest, Dr. Andreas Georgiou, Dr. Daniel Hol-
land, Dr. Mina Mprimpari, Dr. Nikos Dikaios, Foivos Karachalios, Yian-
iii
nis Mattheoudakis, Dr. Ioanna Boulouta, Dr. Johan Carlin, Dr. Annika
Linke, Alex Walther, Michael Worthington, Dimitra Datsiou, Dr. Pierre
DeFouquieres,Dr. KaorukoYamazakiandmywonderfulhousemateDr.
MagdaRapti.
I should not forget my first teachers who helped me develop my sci-
entific abilities and strengthened my courage: Professor Nikolaos Vassi-
las,ProfessorAntonisArgyros,ProfessorAlexandrosTomaras,SteliosKa-
tradis and Vassilis Valavanis. My great acknowledgement to my college
tutor Dr. Giles Yeo who was always available to help with any bureau-
cratic issues. Finally, I would like to thank my fellow colleagues and de-
velopers at NeuroImaging in Python for their hard work and devotion to
theprinciplesoffreesoftwareandopenscience.
iv
Disclaimer
Thisdissertationistheresultofmyownworkandcontainsnothingwhich
is the outcome of work done in collaboration with others, except where
statedexplicitly.
No part of this dissertation has previously been submitted for any de-
greeordiplomaatanyinstitution.
Thisdissertationdoesnotexceed60,000wordsinlength(includingta-
bles,footnotes,bibliographyandappendices).
Throughoutthisdissertationthepluralpronoun‘we’isusedforstylis-
tic reasons and should be taken to refer to either the singular author, the
reader and the author or, when stated explicity, the author and collabora-
tors. Theformassumedshouldbeapparentfromthecontext.
EleftheriosGaryfallidis
Cambridge,10May2012
v
Publications
Conferences
Garyfallidis E, Brett M, Nimmo-Smith I (2010), “Fast Dimensionality
Reduction for Brain Tractography Clustering”, 16th Annual Meeting of the
OrganizationforHumanBrainMapping.
Garyfallidis E, Brett M, Tsiaras V, Vogiatzis G, Nimmo-Smith I (2010),
“Identification of corresponding tracks in diffusion MRI tractographies”, 18th
Proceedings of the International Society of Magnetic Resonance in Medi-
cine.
Correia MM, Williams GB, Yeh F-C, Nimmo-Smith I, Garyfallidis E
(2011),“RobustnessofdiffusionscalarmetricswhenestimatedwithGeneralized
Q-Sampling Imaging acquisition schemes”, 19th Proceedings of the Interna-
tionalSocietyofMagneticResonanceinMedicine.
Garyfallidis E, Brett M, Amirbekian B, Nguyen C, Yeh F-C, Olivetti
E, Halchenko Y, Nimmo-Smith I (2011), “Dipy - a novel software library for
diffusion MR and tractography”, 17th Annual Meeting of the Organization
forHumanBrainMapping.
Garyfallidis E, Gerhard S, Avesani P, Nguyen T, Tsiaras V, Nimmo-
Smith I, Olivetti E (2012), “A software application for real-time, clustering-
based exploration of tractographies”, 18th Annual Meeting of the Organiza-
tionforHumanBrainMapping.
Olivetti E, Nguyen TB, Garyfallidis E (2012), “The Approximation of
the Dissimilarity Projection”, 2nd IEEE International Workshop on Pattern
RecognitioninNeuroImaging.
Garyfallidis E, Nimmo-Smith I (2012), “Cartesian grid q-space recon-
struction”,HARDIReconstructionWorkshopofthe9thIEEEInternational
SymposiumonBiomedicalImaging.
Journals
GaryfallidisE,BrettM,CorreiaM,WilliamsGB,Nimmo-SmithI(2012),
“QuickBundlesforReal-TimeTractographySegmentation”,IEEETransactions
ofMedicalImaging,submitted.
Chamberlain SR, Hampshire A, Menzies LA, Garyfallidis E, Grant
JE, Odlaug BL, Craig K, Fineberg N, Sahakian BJ (2010), “Reduced brain
vi
white matter integrity in trichotillomania: a diffusion tensor imaging study.”,
ArchivesofGeneralPsychiatry67(9):965-71.
Outreach
GaryfallidisE,Nimmo-SmithI(2010),“SurfingyourBrainSuper-High-
ways”. Presented at the 350th celebration of the Royal Society, London,
UK. In the presence of Her Majesty Queen Elizabeth II and His Royal
HighnesstheDukeofEdinburgh.
Software
DiffusionImaginginPython(DIPY):Availableatdipy.org.
FreeOnShades(FOS):Availableatfos.me.
vii
Abstract
Theobjectiveofthisthesisistoimproveonthemethodsforinferringneu-
ral tracts from diffusion weighted magnetic resonance imaging (dMRI).
Accordingly, I present improvements to the reconstruction, integration,
segmentationandregistrationmodalitiesofdMRIanalysis.
I compare and evaluate different Cartesian-grid q-space dMRI acqui-
sition schemes, using methods based on the Fourier transform of the dif-
fusionsignal,withreconstructionsbydiffusionspectrumimagingorgen-
eralised q-ball imaging methods. I propose a new reconstruction method
called diffusion nabla imaging (DNI) which works with all these acqui-
sition schemes, using an algorithm that directly approximates the orien-
tation distribution function using the Laplacian of the signal in q-space.
DNIhasimpressiveaccuracyonlowanglecrossings.
Most previously published reconstruction methods are closely linked
to their own specific track integration method. I have formulated a gen-
eral, non-inferential, deterministic tractography algorithm (EuDX) which
isbasedonEulerintegrationandtrilinearinterpolation,whichworkswith
voxel level information about fibre orientations including multiple cross-
ings, and employs a range of stopping criteria. The purpose of this algo-
rithmistobefaithfultothereconstructionresultsratherthantrytocorrect
orenhancethembyintroducingregionalorglobalconsiderations.
Ihavedevelopedanentirelynew,fullyautomatic,lineartime,cluster-
ing method (QuickBundles) which reduces massive tractographies to just
a few bundles. These bundles are characterised by representative tracks
which are multi-purpose and can be used for interaction with the data
or as the basis for applying higher-complexity clustering methods which
wouldhavebeenimpossibleortooslowwiththefulldataset. QuickBun-
dlesiscurrentlythefastestknowntractographyclusteringalgorithm.
AfterapplyingQuickBundlestotractographiesfromdifferentsubjects,
I show how to use the representative tracks to identify robust landmarks
withineachsubjectwhichIusetodirectlyregisterthedifferenttractogra-
phies together in a highly efficient way. The resulting correspondences
provide important evidence for the anatomical plausibility of the derived
bundles. I demonstrate how these methods can be used for group analy-
sis,andforatlascreation.
This thesis contributes to the understanding of the diffusion signal in
viii
the context of dMRI acquisitions and builds on this foundation towards
a more robust brain tractography which approximates more closely the
underlyingfibrearchitecture.
ix