Automated morphometry in adolescents with OCD and controls, using MR images with incomplete brain coverage

Similar documents
Automated Whole Brain Segmentation Using FreeSurfer

Automated detection of abnormal changes in cortical thickness: A tool to help diagnosis in neocortical focal epilepsy

Regional and Lobe Parcellation Rhesus Monkey Brain Atlas. Manual Tracing for Parcellation Template

Supplementary Information Methods Subjects The study was comprised of 84 chronic pain patients with either chronic back pain (CBP) or osteoarthritis

Gross Organization I The Brain. Reading: BCP Chapter 7

Announcement. Danny to schedule a time if you are interested.

Biological Bases of Behavior. 3: Structure of the Nervous System

Nature Neuroscience doi: /nn Supplementary Figure 1. Characterization of viral injections.

Assessing Brain Volumes Using MorphoBox Prototype

Supplementary Online Content

Magnetic Resonance Angiography

Supplementary information Detailed Materials and Methods

Student Lab #: Date. Lab: Gross Anatomy of Brain Sheep Brain Dissection Organ System: Nervous Subdivision: CNS (Central Nervous System)

DISSECTION OF THE SHEEP'S BRAIN

Online appendices are unedited and posted as supplied by the authors. SUPPLEMENTARY MATERIAL

Sheep Brain Dissection

Group-Wise FMRI Activation Detection on Corresponding Cortical Landmarks

Pediatric MS MRI Study Methodology

Chapter 3. Structure and Function of the Nervous System. Copyright (c) Allyn and Bacon 2004

ANATOMY & PHYSIOLOGY DISSECTION OF THE SHEEP BRAIN LAB GROUP:

APPLICATION OF PHOTOGRAMMETRY TO BRAIN ANATOMY

MR Advance Techniques. Vascular Imaging. Class II

Visualization strategies for major white matter tracts identified by diffusion tensor imaging for intraoperative use

Leah Militello, class of 2018

Discriminative Analysis for Image-Based Population Comparisons

Table 1. Summary of PET and fmri Methods. What is imaged PET fmri BOLD (T2*) Regional brain activation. Blood flow ( 15 O) Arterial spin tagging (AST)

Use of Multimodal Neuroimaging Techniques to Examine Age, Sex, and Alcohol-Related Changes in Brain Structure Through Adolescence and Young Adulthood

The human brain. of cognition need to make sense gives the structure of the brain (duh). ! What is the basic physiology of this organ?

Dissection of the Sheep Brain

Cerebral Cortex 1. Sarah Heilbronner

Cover Page. The handle holds various files of this Leiden University dissertation

BIOL Dissection of the Sheep and Human Brain

Classification and Statistical Analysis of Auditory FMRI Data Using Linear Discriminative Analysis and Quadratic Discriminative Analysis

Repeatability of 2D FISP MR Fingerprinting in the Brain at 1.5T and 3.0T

Review of Longitudinal MRI Analysis for Brain Tumors. Elsa Angelini 17 Nov. 2006

Twelve right-handed subjects between the ages of 22 and 30 were recruited from the

Shape Modeling of the Corpus Callosum for Neuroimaging Studies of the Brain (Part I) Dongqing Chen, Ph.D.

Discriminative Analysis for Image-Based Studies

Organization of The Nervous System PROF. MOUSAED ALFAYEZ & DR. SANAA ALSHAARAWY

CISC 3250 Systems Neuroscience

Organization of The Nervous System PROF. SAEED ABUEL MAKAREM

NIH Public Access Author Manuscript Proc SPIE. Author manuscript; available in PMC 2014 February 07.

Supplementary Information

Department of Human Anatomy GUIDELINES. nuclei. The lateral ventricles. White substance of cerebral hemispheres. course 1

Contributions to Brain MRI Processing and Analysis

Overview of the Nervous System (some basic concepts) Steven McLoon Department of Neuroscience University of Minnesota

MRI-Based Classification Techniques of Autistic vs. Typically Developing Brain

ACR MRI Accreditation: Medical Physicist Role in the Application Process

Functional MRI Mapping Cognition

MR Imaging with the CCSVI or Haacke protocol

High spatial resolution reveals excellent detail in pediatric neuro imaging

Supplementary Online Content

The Central Nervous System I. Chapter 12

CEREBRUM. Dr. Jamila EL Medany

Procedia - Social and Behavioral Sciences 159 ( 2014 ) WCPCG 2014

Funding: NIDCF UL1 DE019583, NIA RL1 AG032119, NINDS RL1 NS062412, NIDA TL1 DA

University of Groningen. The traumatized brain Chalavi, Sima

Magnetic Resonance Imaging. Basics of MRI in practice. Generation of MR signal. Generation of MR signal. Spin echo imaging. Generation of MR signal

Text to brain: predicting the spatial distribution of neuroimaging observations from text reports (submitted to MICCAI 2018)

Differences in brain structure and function between the sexes has been a topic of

Supplementary Online Material Supplementary Table S1 to S5 Supplementary Figure S1 to S4

Introduction to the Central Nervous System: Internal Structure

WHAT DOES THE BRAIN TELL US ABOUT TRUST AND DISTRUST? EVIDENCE FROM A FUNCTIONAL NEUROIMAGING STUDY 1

Study of the CNS. Bent O. Kjos' Richard L. Ehman Michael Brant-Zawadzki William M. Kelly David Norman Thomas H. Newton

ACR MRI Accreditation Program. ACR MRI Accreditation Program Update. Educational Objectives. ACR accreditation. History. New Modular Program

Cerebrum-Cerebral Hemispheres. Cuneyt Mirzanli Istanbul Gelisim University

Medical Neuroscience Tutorial Notes

Department of Cognitive Science UCSD

Fig.1: A, Sagittal 110x110 mm subimage close to the midline, passing through the cingulum. Note that the fibers of the corpus callosum run at a

Hallucinations and conscious access to visual inputs in Parkinson s disease

The human brain weighs roughly 1.5 kg and has an average volume of 1130 cm 3. A sheep s brain weighs in however at kg.

Standardized, Reproducible, High Resolution Global Measurements of T1 Relaxation Metrics in Cases of Multiple Sclerosis

Resistance to forgetting associated with hippocampus-mediated. reactivation during new learning

Activated Fibers: Fiber-centered Activation Detection in Task-based FMRI

P. Hitchcock, Ph.D. Department of Cell and Developmental Biology Kellogg Eye Center. Wednesday, 16 March 2009, 1:00p.m. 2:00p.m.

Voxel-based morphometry in clinical neurosciences

RECENT ADVANCES IN CLINICAL MR OF ARTICULAR CARTILAGE

Tumor cut segmentation for Blemish Cells Detection in Human Brain Based on Cellular Automata

Clinically focused workflow with unique ability to integrate fmri, DTI, fiber tracks and perfusion in a single, multi-layered 3D rendering

ASSUMPTION OF COGNITIVE UNIFORMITY

Supporting Information

CHAPTER 13&14: The Central Nervous System. Anatomy of the CNS

Brain anatomy tutorial. Dr. Michal Ben-Shachar 459 Neurolinguistics

SUPPLEMENTARY MATERIAL. Table. Neuroimaging studies on the premonitory urge and sensory function in patients with Tourette syndrome.

Anatomy & Physiology Central Nervous System Worksheet

2D-Sigmoid Enhancement Prior to Segment MRI Glioma Tumour

Ch 13: Central Nervous System Part 1: The Brain p 374

Neuroradiology MR Protocols

Supporting Information

Anatomy and Physiology (Bio 220) The Brain Chapter 14 and select portions of Chapter 16

Quantitative Neuroimaging- Gray and white matter Alteration in Multiple Sclerosis. Lior Or-Bach Instructors: Prof. Anat Achiron Dr.

QIBA/NIBIB Final Progress Report

Human Brain Myelination from Birth to 4.5 Years

Chapter 18: The Brain & Cranial Nerves. Origin of the Brain

Influence of Velocity Encoding and Position of Image Plane in Patients with Aortic Valve Insufficiency Using 2D Phase Contrast MRI

PROPERTY OF ELSEVIER SAMPLE CONTENT - NOT FINAL. Gross Anatomy and General Organization of the Central Nervous System

Title Setting: Subject: Grade Level: Time Frame: Paired Dana Alliance Fact Sheets:

Multimodal Magnetic Resonance Imaging Study of Treatment-Naïve Adults with Attention-Deficit/ Hyperactivity Disorder

Information fusion approach for detection of brain structures in MRI

Brain tissue and white matter lesion volume analysis in diabetes mellitus type 2

Transcription:

Automated morphometry in adolescents with OCD and controls, using MR images with incomplete brain coverage M.Sc. Thesis Oscar Gustafsson gusgustaos@student.gu.se Supervisors: Göran Starck Maria Ljungberg Department of Radiation Physics University of Gothenburg Gothenburg, Sweden January 2013

Abstract From an earlier study of adolescents newly diagnosed with obsessive-compulsive disorder, 3D-images of the brain of a patient group and control group were available. Morphometric analysis, using available software, was proposed to extract possible additional information from the study. As only the central parts of the brain were of interest in the earlier study the images did not completely cover the brain in order to reduce the scan time. A morphometry method that was not affected by the truncated field of view was therefore needed. A morphometry method utilizing the Freesurfer software was developed. Validation of the method was performed where the results, when analyzing images with complete and not complete coverage of the brain, were compared. As the only major effects were found in the truncated area and for structures with diffuse borders, the developed method was considered reliable for analyses of the truncated images from the patients and controls. The analysis incorporated tests for difference in volume of the subcortical structures, right-left asymmetry of the volumes of the subcortical structures, difference in thickness of the cerebral cortex and correlation between cortical thickness and the severity of the disease (measured with CYBOCS). The patients left caudate nucleus was found to be significantly larger than the controls comparing both genders, but also for males only. Furthermore, the putamen and the optic chiasm were found to be larger for female patients as compared to female controls. No volume asymmetry on the subcortical structures, which has not been seen previously among healthy persons, was found. A negative correlation between the thickness of the cerebral cortex in the left fusiform gyrus and the severity of the disease was found among the patients. An area at the left postcentral gyrus was discovered to have thicker cerebral cortex in patients as compared to controls.

Acknowledgements All this would not have been without the people who have supported me during the last months. I would like to express my greatest gratitude to Göran and Maria for being excellent supervisors and for letting me be a part of this project. I would also like to thank all MRphysicists at the Sahlgrenska University Hospital for making this the best period in my education. The tremendous work that the Freesurfer team has put down to make the software available for the public is of course also very appreciated. Lastly, thank you Emma for supporting and encouraging me. You are my rock. ii

Nomenclature AP CSF CYBOCS DBM DTI FOV GE GLM GM FH fsaverage FSL ICV MEG MR MRI MRS OCD Parcellation PBM Pial surface RF RL SBM SE SPM Anterior-Posterior (direction) Cerebrospinal Fluid Children s Yale-Brown Obsessive-Compulsive Scale Deformation Based Morphometry Diffusion Tensor Imaging Field of View Gradient Echo General Linear Model Gray Matter Feet-Head (direction) A template brain provided with Freesurfer FMRIB Software Library Intra Cranial Volume Magnetoecephalography Magnetic Resonance Magnetic Resonance Imaging Magnetic Resonance Spectroscopy Obsessive-Compulsive Disorder A procedure where a surface (e.g. the cerebral cortex) is divided in smaller areas which are given individual labels (e.g. insula) Pattern Based Morphometry The surface produced by Freesurfer that forms the outer surface of the cerebral cortex Radio Frequency Right-Left (direction) Surface Based Morphometry Spin Echo Statistical Parametric Mapping Talairach transformation A 9 parameter affine transformation to the Talairach space TBM TE Tessellation TFE Tensor Based Morphometry Echo Time The process in which the repetition of a geometrical shape is used to produce a surface without any gaps or overlaps. Freesurfer uses triangles to define surfaces Turbo Field Echo iii

TR T1 T1W T2 YBOCS VBM Vertex WM WM-surface Repetition Time Spin-Lattice Relaxation Time T1-weighted Spin-Spin Relaxation Time Yale-Brown Obsessive-Compulsive Scale Voxel Based Morphometry Point were the corners of the triangles which define the surfaces meet White Matter The surface produced by Freesurfer that forms the inner surface of the cerebral cortex iv

Table of contents Abstract...i Acknowledgements...ii Nomenclature...iii 1. Introduction...1 1.1. Aim...2 1.2. T1-weighted images...2 1.3. Morphometry...2 1.3.1. Voxel based morphometry (VBM)...3 1.3.2. Deformation based morphometry (DBM)...4 1.3.3. Surface based morphometry (SBM)...4 1.4. Obsessive-compulsive Disorder...4 1.4.1. Symptoms...4 1.4.2. Prevalence...4 1.4.3. Classification...4 1.5. Anatomy of the brain...5 2. Materials and methods...7 2.1. Imaging of OCD patients...7 2.2. Evaluation of available software tools...7 2.3. Description of the method...8 2.3.1. Description of Freesurfer...8 2.3.2. Procedures specific to the project...8 2.4. Validation of the method...10 2.4.1. Subjects...10 2.4.2. Image acquisition...10 2.4.3. Image analysis...11 2.4.4. Statistical analysis...11 2.5. Analysis of OCD patients and controls...11 2.5.1. Subjects...11 2.5.2. Image analysis...11 2.5.3. Statistical analysis...12 3. Results...13 3.1. Outcome of the evaluation...13 3.2. Validation of the method...13 3.2.1. Subcortical segmentation...13 3.2.2. Cortical segmentation...13 3.3. Analysis of OCD patients and controls...18 3.3.1. Subcortical segmentation...19 3.3.2. Cortical segmentation...20 4. Discussion...23 4.1. The evaluation process...23 4.2. The validation...23 4.2.1. Corrections...23 4.2.2. Subcortical segmentation...23 4.2.3. Cortical segmentation...24 4.2.4. The truncation...24 4.3. Analysis of OCD patients and controls...24 4.3.1. Corrections...24 4.3.2. Subcortical segmentation...25 4.3.3. Cortical segmentation...25 v

4.4. Statistical testing with small samples...26 4.5. Future aspects...26 5. Conclusions...27 References...28 Appendix I Statistical tests...1 Wilcoxon Signed-Rank test...1 Mann-Whitney U-test...1 vi

1. Introduction During the last years a magnetic resonance (MR) study concerning obsessive-compulsive disorder (OCD) has been going on at Sahlgrenska University Hospital in Gothenburg, Sweden. The patients were treatment naïve, newly diagnosed adolescents. The MR examination of the patients and the age- and gender-matched controls included a T1-weighted (T1W) 3D MR image (MRI), a MR diffusion tensor imaging (DTI) scan and four magnetic resonance spectroscopy (MRS) measurements. The main purpose of the MRI was to give high-resolution images to plan the following DTI and MRS measurements. However, morphometric analysis, the comparison of volumetric measurements, of the MRI data was later proposed as another option to extract useful information. This thesis describes the morphometric analysis of the acquired MR images of the patients and controls. The MRS exams were however concentrated to the central parts of the brain and as the MR images were only acquired for planning, an anatomy scan of the most lateral parts of the brain was less meaningful. To cut the scan time for the relatively young subjects in the study images of the most lateral parts of their heads were not acquired. Therefore, the anatomical images of both patients and controls miss two large pieces of the skull (one on either side) and some gray matter. In some cases a small amount of the white matter is not included either (fig 1.1). a b c d e f g h Figure 1.1. Subjects in the OCD study with the least and the most truncated images. The two images on the left in the upper row (a, b) show an axial and a coronal slice of the subject with the smallest truncation, while the two images on the right in the upper row (c, d) show the subject with the largest truncation. The first and the last sagittal slices of the least truncated subject (e, f) and of the most truncated subject (g, h) are shown in the lower row. 1

1.1. Aim The aim of this study was to develop a method to extract morphometric data from the truncated images of the OCD-patients and healthy controls using available software and to apply the method in a pilot morphometric evaluation of these images. Specifically: - Evaluate freely distributed software in order to chose the most applicable to this study - Develop a method utilizing the chosen software - Investigate the influence of the image truncation on the morphometric results - Apply the method on the OCD- and control-groups 1.2. T1-weighted images The net magnetization, caused by the static magnetic field of the MR-scanner and flipped into the transverse plane by an excitation pulse, relaxes simultaneously in two ways, called spinlattice relaxation and spin-spin relaxation. The spin-spin relaxation, which is caused by the transversal dephasing of the spins, decreases the transverse magnetization while the spinlattice relaxation, which is caused by the interactions between protons and the environment, increases the longitudinal magnetization. The time between the excitation pulse and the time when the transverse magnetization has been reduced to 37 % of its initial value (immediately after the excitation pulse) is called T2 or spin-spin relaxation time. T1 or the spin-lattice relaxation time is the time between the excitation pulse and the time when the longitudinal magnetization has regained 63 % of the difference between the longitudinal magnetization immediately after the excitation pulse and thermal equilibrium. The T2-relaxation time is always shorter than the T1-time and the time constants are specific for the type of tissue and magnetic field strength. The different time constants are used to produce contrast in images. The anatomical images acquired in this project are T1-weighted images. The contrast in a T1- weighted image comes from the difference in T1 for different tissues. The two main imaging techniques are based on the spin echo (SE) and the gradient echo (GE). In order to produce a T1-weighted image with a spin echo based sequence a short repetition time (TR) and a short echo time (TE) is used. The short TR results in an incomplete and different longitudinal relaxation in tissues with different T1 while the short echo time minimizes the influence of the spin-spin relaxation on the signal. For gradient echo based sequences the TR is usually very short (compared to T1) and the use of a large flip angle produces the incomplete longitudinal relaxation instead. The technique used for imaging in this project was a turbo field echo (TFE), aka. MP-RAGE or Snapshot FLASH, which utilizes a preparation (inversion or saturation) pulse follow by a gradient echo sequence with very short TR and TE to produce T1-weighted images. 1.3. Morphometry The word morphometry is composed of the two Greek words morphé, which means form and metron, which means measurement. It refers to the quantitative analysis of form, which in this aspect includes both shape and size. Morphometry is used in many areas, but in this thesis it will from now on refer to brain morphometry. 2

There are a number of different techniques used to preform morphometric analyses (fig 1.2), where voxel based morphometry (VBM) (1), deformation based morphometry (DBM) (2), pattern based morphometry (PBM) (3), surface based morphometry (SBM, in literature also known as surface based analysis) (4, 5) and fibre tracking techniques (6) are just a few examples. The first five of these examples are based on T1-weighted images while the last one is based on diffusion-weighted images such as DTI. The most commonly used techniques are explained below. BrainMorphometry Based on diffusion weighted images Based on T1-weighted images Baseimage Fibre tracking techniques Voxel based morphometry Deformation based morphometry Surface based morphometry Pattern based morphometry Techniques Figure 1.2. Some of the most commonly used brain morphometry techniques 1.3.1. Voxel based morphometry (VBM) During the last decade VBM has become the most commonly used method in brain morphometry, mainly because it is fully automated, which makes it easy and fast to use. It is used for comparing volumes of tissue in groups of subjects. VBM involves four main steps (7). First the relevant tissue, typically gray matter (GM), is segmented, then the tissue maps of all subjects in a group are warped to a common space and smoothed with a isotropical gaussian kernel and at last a voxel-by-voxel statistical analysis is preformed. The spatial normalization is non-linear and aims to correct the global differences and not to match every single structure. If it would, all the tissue maps would look exactly the same and hence the statistical analysis would not give any results. 3

Although VBM is the most commonly used technique in the field of brain morphometry, it has some major flaws. For example, VBM is as sensitive to a miss registration as it is to a true structural change. The most commonly used software for preforming VBM are Statistical Parametric Mapping (SPM, http://www.fil.ion.ucl.ac.uk/spm/) and the FMRIB Software Library (FSL, http://fsl.fmrib.ox.ac.uk/). 1.3.2. Deformation based morphometry (DBM) The approach to DBM is similar to that of VBM. However, instead of warping the tissue maps to a common space one uses the deformation fields produced by the warping (2). The deformation fields give macroscopic information; hence DBM can be used to study largescale differences of brain structure while VBM rather aims to study mesoscopic ones. The result can either be displayed by the deformation fields or as the deformations applied to an image. Currently, no global standard for DBM has been adopted and because of that it has not been included in any of the major software packages. 1.3.3. Surface based morphometry (SBM) Unlike VBM, SBM extracts the morphometric measurements from surfaces. The first surface used is the outer boundary of the white matter. The other surface used is the border between gray matter and pial matter. The two surfaces thus encompass the gray matter. From the two surfaces, measurements such as volume, surface area and thickness can be extracted. A strength of the surface based morphometry is the use of surfaces, which are more anatomically relevant than voxels. SBM has therefore been presented as a better alternative for brain morphometry than VBM (8). The software often used in SBM studies is Freesurfer. A more thorough description of Freesurfer based SBM can be read in the method section 2.3.1 below. 1.4. Obsessive-compulsive Disorder 1.4.1. Symptoms Common to all OCD patients is a ritualistic behavior, which has to be completed (compulsion), sometimes a large number of times, to neutralize a perceived threat (obsession). Examples are hand washing or checking if the stove is turned off. If the ritual cannot be completed the patient gets distressed. The anxiety disappears when the ritual can be completed. The difference between normal behavior and disease is when the ritual becomes a problem, e.g. when the hand washing takes several hours every day and therefore makes the person late for work or impairs normal social life. The ritual can be physical actions as well as neutralizing thoughts. In a try to escape the intrusive thoughts, the patients often avoid situations they know might trigger the obsession. 1.4.2. Prevalence Approximately 2 to 3 percent of the world's population is believed to have the disease (9). The onset of the OCD most often occurs in the late childhood or early adulthood. Only 15 percent of the patients are older than 35 years at the time of the onset (10). OCD is evenly distributed on the both genders. Generally, males have a slightly earlier onset though. 1.4.3. Classification To rate the severity of the symptoms the Yale-Brown Obsessive Compulsive Scale (YBOCS) is often used. The obsessions and the compulsions are rated separately with a maximum score 4

of 20 each, resulting in a maximum total score of 40. A higher score is equal to a more severe disease. Children are analyzed with the Children s Yale-Brown Obsessive Compulsive Scale (CYBOCS). The main difference between the two scales is the language in the questions asked, which is CYBOCS is more adopted for children. CYBOCS is used for persons up to the age of 18 years. 1.5. Anatomy of the brain The brain is composed of nervous tissue. A nerve cell has a body, an axon and a number of dendrites. The axon is a long outgrowth (up to a meter) that transfers information to other parts of the brain or body, while the dendrites are used by the nerve cell to communicate with other nerve cells that are nearby. Sections with a large proportion of cell bodies, i.e. where information is transferred between nerve cells, are called gray matter, while sections with a large proportion of axons are called white matter (WM). The brain is mainly composed of white matter covered by gray matter. The outer layer of the brain that is made of gray matter is called the cerebral cortex. On the inside of the brain there are a number of structures mainly made of gray matter called the subcortical structures (fig. 1.3). Some of these (caudate nucleus, putamen, hippocampus) are of greatest interest for this project as there have been reports of deviation in volume for those structures (11-13). The accumbens area is not displayed in figure 1.3. It is however located close to the caudate nucleus and the putamen. In addition to the gray and white matter there are cavities filled with cerebrospinal fluid (CSF) called ventricles. The tissue that produces CSF is called choroid plexus and is situated on the walls of the lateral ventricles. 5

Figure 1.3. The subcortical structures of the brain. Globus pallidus is called pallidum in the rest of the thesis. Figure from The human mind explained, Susan Greenfield ed, 1996, Cassell Publishers, p.34 6

2. Materials and methods The imaging of the OCD patients and the age- and gender-matched controls defined the conditions for everything else in this project and is thus described first. The morphometric analysis of the OCD patients and controls is described last in the Materials and methods section. 2.1. Imaging of OCD patients T1-weighted 3D MR brain images of all subjects were acquired using a TFE sequence on a 1.5 T Philips Achieva MR scanner (Best, the Netherlands), software release 2.6.3 with a transmit receive head coil of bird cage type (same vendor). To secure the position of the subjects heads a vacuum pillow was used. The imaging parameters were: TR 6.796 ms, TE 3.12 ms, flip angle 8 degrees, voxel size 1.25x1.25x1.25 mm, field of view (FOV) (FH, AP, RL) 300x255x125 mm (FH is the Feet-Head direction, AP is the Anterior-Posterior direction, RL is the Right-Left direction), 1 measurement. The images were angulated and positioned using a survey scan so that the mid plane of the 3D volume was aligned with the plane of symmetry of the brain. Also, the bottom of the 3D volume was angulated to align with the bottom of corpus callosum in the sagittal projection. The RL FOV was too small to cover the subjects heads; therefore the most lateral sections of the head were missing in the images. 2.2. Evaluation of available software tools To find a method, which was applicable to this specific case a few important features, that the method and software needed to have, were identified. Here the method is defined as the complete procedure using the chosen software toolbox, while the software toolbox is the chosen software i.e. Freesurfer in this work. The method must: - be documented to work well on children and/or teenagers - give a metric output, e.g. mm, mm 2, mm 3 and in addition the software toolbox must: - have the ability to handle truncated images, especially the not-complete skull - not require any manually placed landmarks - give the option to do manual adjustments, e.g. adjusting incorrect boundaries - give accurate measurements of the subcortical structures Both the method and the software must also be well documented and tested. The reason why a metric output was needed was to make it possible to evaluate if the cut edges produced any substantial errors (see section 2.4 below). The software must not require any manually placed landmarks, as it would increase the time a person has to put in to analyze a single subject and the knowledge needed for analyzing. Some knowledge and time is however needed for examining the images and search for errors in the automatic segmentation. If manual adjustments are possible these unavoidable errors can be removed to give the software the opportunity to produce results, e.g. cortical thickness, that is closer to the true value. The evaluation was done by first finding options fulfilling all binary features, i.e. those that only could be fulfilled or not e.g. give a metric output, and then to maximize the features left, i.e. those that could be more or less fulfilled, e.g. working well on children and/or teenagers by comparing the qualified methods/software tools. The software tools included in the evaluation were BrainSuite, BrainVisa, Freesurfer, FSL, Spherical demon and SPM. 7

2.3. Description of the method 2.3.1. Description of Freesurfer The Freesurfer software tool for image analysis (version 5.1) was used for segmentation and reconstruction of the cortex of the brain. The software tool and documentation are freely available for download (http://surfer.nmr.mgh.harvard.edu/). The images of all subjects were set to a common coordinate system with an automated Talairach transformation, i.e. a 9 parameter affine transformation to the Talairach space that used in Freesurfer. To correct for inhomogeneous signal intensity, which may be caused by factors such as inhomogeneous radio-frequency (RF) excitation or non-uniform reception sensitivity, intensity normalization was applied (14). The skull and other non-brain tissue were removed from the 3D-image of each subject using a procedure that utilizes watershed and surface deformation (15).The subcortical structures, including the accumbens area, amygdala, the caudate nucleus, hippocampus, pallidum, putamen and thalamus, and the optic chiasm were segmented into volumetric structures (16, 17). A tessellation of the gray matter to white matter boundary, where the border is covered by triangles, was followed by an automated topology correction to create a surface covering the white matter (18, 19). Surface deformation following intensity gradients was applied to place the border between gray and white matter (called WM surface) and the border between gray matter and CSF (called pial surface) optimally, i.e. at the point where the shift in intensity reaches a maximum which defines the transition to the other tissue (4, 20, 21). This step completes the cortical models and the following steps are instead connected to mapping and parcellation, i.e. labeling of different areas of cortex, which are needed for a smooth way of categorizing the data extracted from the cortical model for further analysis. To start with, the smoothed WM surface was inflated in a process that tries to minimize the metric distortions, i.e. attempting to preserve distance and area (5). This grants the possibility of a registration to a spherical atlas, which is able to match geometrical parts of the brain across subjects (22). For the parcellation of the cerebral cortex a method based on patterns of the sulci and gyri was used (23, 24). A measurement of the cortical thickness was calculated as the smallest distance between the WM surface and the pial surface for each vertex i.e. the point where the triangles, which build the surfaces, meet (21). As the spatial gradients of image intensity were used to produce the maps, these are not dependent of absolute signal intensity. The measurements were made on the surfaces created and were thus not restricted by the voxel size (1.25 mm); hence submillimeter differences between groups were detectable. The measurement of intra cranial volume (ICV) was based on the transformation of the image stack to the template used (25). The technique uses the fact that the determinant of the transformation matrix is inversely proportional to the ICV. Validations of the measurement of cortical thickness has been performed both compared to histological samples (26) but also compared to manual measurements in MR-images (27, 28). The reproducibility of the morphometric results produced by Freesurfer has been verified for different MR scanner manufacturers as well as across different magnetic field strengths (29, 30). 2.3.2. Procedures specific to the project The Freesurfer software cannot handle a FOV that is larger than 256 mm in any direction. As the FH FOV is 300 mm in the images of the study, a flag had to be set (-cw256) that automatically truncated the images down to 256 mm. The segmentation process was manually inspected by the author to assure proper Talairach transformation, complete but not excessive skull strip, WM and pial surfaces that 8

corresponded to the true ones and correct segmentation. Where errors were observed manual adjustments were made and the algorithm was re-run. This was done iteratively until no errors that would influence any result, which were important to the study, were present. The manual adjustments of the Talairach transformation was done in the following steps; first the corpus callosum of the image and the template were aligned in a central sagittal slice, then translation, rotation and scaling were used to match the outer boarders of the brain in all directions (sagittal, axial and coronal) as good as possible (fig 2.1). The skull strip was corrected by manually removing any remaining non-brain pixels. The surfaces were inspected using the coronal slices. Since the surfaces are constrained to be smooth all errors could be seen using only one projection. If there were errors in both the WM and the pial surface the WM surface was manually adjusted first as it is used as a starting point when the pial surface is created. The WM surface was corrected by adding or removing pixels from the white matter image. A few of the subjects had errors in the pial surface at insula. This is a known error of the software (31) and was manually adjusted by changing the segmentation classification of the voxels labeled as vessel to putamen at the area of the error (fig. 2.2). The inspection of the segmentation was focused on obvious errors such as misclassification where part of a ventricle was classified as white matter or where sections were not classified at all (fig 2.3). Figure 2.1. Example of correction of the Talairach transformation. In the left image (uncorrected) the white matter of the subject expands beyond the border of the template brain. In the right image (corrected) it does not. 9

Figure 2.2. Example of correction of the pial surface (showing the right insula). The pial surface crosses the WM surface in the uncorrected image (left), but not in the corrected one (right). The correction was made by changing the segmentation class of a few voxels from vessel (purple) to putamen (pink) Figure 2.3. Example of correction of the segmentation. Voxels without segmentation class in the uncorrected image (left) were classified as WM and GM in the corrected one (right). 2.4. Validation of the method 2.4.1. Subjects A group of six healthy adults (3 males and 3 females, mean age 22.3 years, range 21-26) were scanned. They were all recruited from the education for medical physicists at the University of Gothenburg. All subjects signed an informed consent before the start of their examination. 2.4.2. Image acquisition The same parameters and the same MR scanner was used as in the scanning of the OCD study, with the only exception that the RL FOV was increased to 187.5 mm, i.e. 150 sagittal slices instead of 100. Furthermore, a new software release (R 3.2.1) was installed on the MR- 10

system. 2.4.3. Image analysis The images from the validation group were used to create two sets of image data, one with images with complete coverage of the brain and one with truncated images, mimicking the OCD-images. The image stacks were truncated by deleting the first 25 and the last 25 sagittal slices, thus using the remaining 100 images as input. The slices to delete were chosen by examining where the original FOV would have truncated the images. Since all images were centered to the midplane of the brain it was expected that the number of slices to delete on either side would be the same, i.e. 25. As a measure of the truncation, the number of deleted slices containing gray matter was counted. The datasets were then treated in the same way as the OCD datasets including manual inspection and correction of any visible errors as described above. 2.4.4. Statistical analysis To compensate for different brain size, the output of the subcortical segmentation, i.e. volume measurements of the subcortical structures, was normalized to the value for the images with complete coverage of the brain of each subject, making all measurements for the images with complete coverage equal to unity. For each structure a Wilcoxon signed-rank test with a significance level of p<0.05 was used to search for significant difference between the images with complete coverage and the truncated ones. The mean of the normalized volumes generated from the truncated images was observed as well as its standard deviation. To analyze the output from the cortical segmentation, i.e. maps of cortical thickness, the percentage difference between images with complete coverage and the truncated ones was calculated in every vertex for all subjects. The result was then mapped to a template brain (fsaverage that comes with the software) and smoothed with a gaussian kernel with a FWHM = 15 mm. These images were visually inspected for areas of large differences in cortical thickness. 2.5. Analysis of OCD patients and controls 2.5.1. Subjects T1-weighted 3D datasets from 13 adolescents that were newly diagnosed with OCD and 11 age- and gender-matched controls were acquired in a previous study (table 2.1). Due to major errors in the segmentation, which would have required some parts of the brain to be manually segmented, one of the male control subjects was not analyzed. The patients were recruited from Queen Silvia s Children Hospital. The study was approved by the Ethical Review Board at University of Gothenburg. All subjects and their parents signed an informed consent before the start of the examination. 2.5.2. Image analysis The 3D images of the subjects were processed with the Freesurfer software and inspected and corrected as described in section 2.3.2. The head was positioned close to the image top; therefore the -cw256 flag could not be used as the automatic procedure to truncate the images in the FH direction could not identify which part of the image that should be truncated without losing important information. Consequently, the top of the head was truncated in most images. Therefore, the image processing software mipav (Medical Image Processing, Analysis, and Visualization, version 6.0.1, http://mipav.cit.nih.gov) was used to manually truncate the images in the FH direction. 11

Table 2.1. Subject information Experimental group OCD (n = 13) Controls (n = 10) Age (years) 14.4 (1.84) 15.4 (1.94) Males 6 4 Females 7 6 CYBOCS Total 23.8 (5.30) - Obsessions 12.2 (2.95) - Compulsions 11.6 (2.52) - Values are mean (SD) 2.5.3. Statistical analysis The volumes of the subcortical structures except the optic chiasm were normalized to the ICV for all subjects. The results were tested using a Mann-Whitney U-test with a significance level of p<0.05. Patients and controls were compared both as complete groups and for each gender separately. Right-left volume asymmetry was tested with a Wilcoxon signed rank test (significance level: p<0.05) in those structures that it was applicable to (thalamus, the caudate nuclei, putamen, pallidum, hippocampus, amygdala, accumbens area, ventral diencephalon and choroid plexus). Patients and controls were tested separately, both as complete groups and for each gender subgroup. The cortical measurements were analyzed by applying a general linear model (GLM) on the vertex level. The cortical thickness of each subject was mapped to a template surface (fsaverage) and smoothed with a gaussian kernel (FWHM 10 mm). Difference in cortical thickness between patients and controls as well as correlation between CYBOCS and cortical thickness of patients were tested with a t-test (p<0.01). Only clusters, i.e. multiple adjacent significant vertices, with an area larger than 50 mm 2 were registered. Both analyses were performed with both genders and with each gender separately. 12

3. Results 3.1. Outcome of the evaluation The freely available Freesurfer-software with its surface based morphometry was found as the best choice according to the features listed. This was based on the facts that: - SBM has been seen to produce better results than VBM when analyzing children (32) - The output of the Freesurfer based SBM is metric - Freesurfer uses a skull stripping procedure that works from the brain outwards rather than from the skull inwards (15). The missing skull due to the truncation of the images is thus not a problem as the technique finds the brain tissue and removes everything else instead of finding the non-brain tissue to remove - Freesurfer is one of few SBM-software tools that is fully automatic (32) - It is possible to preform manual adjustments in Freesurfer - Freesurfer has proved superior to other commonly used software for subcortical segmentation (33, 34) Furthermore, SBM splits the measurement into surface and thickness and therefore reduces the risk to miss differences such as increased surface combined with decreased thickness, which might add up to no difference in volume that is the main parameter in VBM (35). 3.2. Validation of the method It was found that the left side of the brain was slightly more truncated than the right side, on average 1 more slice was truncated on the left side. From the correction table (table 3.1) it can be seen that the Talairach transformation and skull strip were the steps that were most affected by the truncation, showing clear differences between the images with complete coverage of the brain and the truncated ones, while the others (WM surface, pial surface and segmentation) were more affected by the individual variation of anatomy. Pairs, of complete and truncated images, of corrections can be seen in the table for the later corrections. Not seen in the table is the systematic error of the Talairach transformation in the RL direction for all truncated images, i.e. the transformed images were shorter in the RL direction than they should have been. Before the manual adjustments of the Talairach transformation were made, the ICV was consistently smaller for the truncated images than for the ones with complete coverage of the brain when the subjects were compared individually, with a mean difference of 5 percent (range: 1.4 % - 8.6 %). 3.2.1. Subcortical segmentation As seen in table 3.2, there were no significant differences (p<0.05) among the segmented structures analyzed when the signed rank test was used. The mean fraction, comparing images with complete coverage and truncated ones, was in general close to unity. Two structures (right hippocampus and right choroid plexus) showed differences slightly over five percent and only one (right amygdala) had a difference between 2 % and 5 %. 3.2.2. Cortical segmentation The major differences in cortical thickness were found in the lateral parts of the brain (fig. 3.1-3.4). Differences can however also be seen at the bottom of the frontal lobe in both hemispheres. The magnitude of these was smaller than the ones on the lateral structures. 13

Table 3.1. Corrections made during the validation process Subject no. Talairach transformation (complete/truncated) Skull strip (comp/trunc) Correction WM surface (comp/trunc) Pial surface (comp/trunc) Segmentation (comp/trunc) 1 C/C -/C C/C C/- C/- 2 C/C -/C -/- -/- -/C 3 C/C -/C C/C C/C C (p) / C (p) 4 -/C -/C -/- C/C -/- 5 C/C C/C -/- -/- -/- 6 -/C -/C C/C C/C C (p) / C (p) C means that a correction was preformed, - means that no corrections were performed C (p) means that the segmentation was changed to correct the pial surface 14

Table 3.2. The results from the subcortical segmentation of the validation subjects. The volume of all structures from the truncated images was divided by the volume of the corresponding structure in the images with complete coverage of the brain. The mean and standard deviation of this quotient is found in the right column. The quotients were also used as input to the Wilcoxon test together with the normalized volume from the images with complete coverage (always equal to unity). Structure Statistical test (Wilcoxon signed rank test) p-value Comparison Fraction (mean (SD)) Left Cerebellum White Matter 0.6875 1.011 (0.045) Left Cerebellum Cortex 1.0000 1.013 (0.033) Left Thalamus Proper 0.6875 0.997 (0.013) Left Caudate Nucleus 0.3125 1.015 (0.020) Left Putamen 0.0938 1.008 (0.009) Left Pallidum 0.3125 0.984 (0.028) Brain Stem 0.6875 1.004 (0.017) Left Hippocampus 0.4375 1.005 (0.012) Left Amygdala 1.0000 1.007 (0.062) Left Accumbens-area 0.8438 0.993 (0.074) Left Ventral DC 0.5625 1.013 (0.032) Left choroid plexus 0.8438 0.994 (0.040) Right Cerebellum White Matter 0.5625 1.000 (0.035) Right Cerebellum Cortex 1.0000 1.009 (0.040) Right Thalamus Proper 0.5625 0.990 (0.021) Right Caudate Nucleus 0.2188 1.008 (0.012) Right Putamen 0.6875 0.996 (0.012) Right Pallidum 1.0000 0.995 (0.030) Right Hippocampus 0.0938 1.053 (0.051) Right Amygdala 0.5625 1.026 (0.051) Right Accumbens-area 0.6875 1.010 (0.051) Right Ventral DC 0.2188 0.981 (0.028) Right choroid plexus 0.3125 0.943 (0.089) Optic Chiasm 0.8438 0.983 (0.099) Corpus Callosum Posterior 1.0000 1.002 (0.015) Corpus Callosum Mid Posterior 0.8438 0.993 (0.027) Corpus Callosum Central 0.3125 1.006 (0.009) Corpus Callosum Mid Anterior 0.8438 1.005 (0.017) Corpus Callosum Anterior 0.5625 1.004 (0.012) Left hemisphere Cortex Volume 1.0000 0.993 (0.020) Right hemisphere Cortex Volume 0.2188 0.989 (0.014) Cortex Volume 0.3125 0.991 (0.014) Left hemisphere Cortical White Matter Volume 0.3125 1.002 (0.004) Right hemisphere Cortical White Matter Volume 1.0000 1.002 (0.004) Cortical White Matter Volume 0.2188 1.002 (0.003) Subcortical Gray Matter Volume 1.0000 1.007 (0.017) Total Gray Matter Volume 0.4375 0.995 (0.011) Supra Tentorial Volume 0.3125 0.996 (0.008) Intra Cranial Volume 0.6875 1.009 (0.028) 15

Figure 3.1. Mean thickness difference between images with complete coverage of the brain and truncated images of the left hemisphere overlaid on the inflated surface of the fsaverage. Scale is percent normalized to the value of the image with complete coverage. Top left image is lateral view, top right image is medial view, bottom left image is inferior view and bottom right image is superior view Figure 3.2. Mean thickness difference between images with complete coverage of the brain and truncated images of the right hemisphere overlaid on the inflated surface of the fsaverage. Scale is percent normalized to the value of the image with complete coverage. Top left image is lateral view, top right image is medial view, bottom left image is inferior view and bottom right image is superior view 16

Figure 3.3. The standard deviation of the cortical thickness difference between images with complete coverage of the brain and truncated images of the left hemisphere overlaid on the inflated surface of the fsaverage. Scale is percent normalized to the value of the image with complete coverage. Top left image is lateral view, top right image is medial view, bottom left image is inferior view and bottom right image is superior view Figure 3.4. The standard deviation of the cortical thickness difference between images with complete coverage of the brain and truncated images of the right hemisphere overlaid on the inflated surface of the fsaverage. Scale is percent normalized to the value of the image with complete coverage. Top left image is lateral view, top right image is medial view, bottom left image is inferior view and bottom right image is superior view 17

3.3. Analysis of OCD patients and controls Corresponding to the systematic error seen in the validation process adjustments of the Talairach transformation were needed for all subjects (table 3.3). The majority of the adjustments on the pial surface were in the right insula (fig. 2.2). Table 3.3. Corrections made on the patient and control groups Subject no. Talairach transformation Correction Skull strip WM surface Pial surface Segmentation Patient 1 F C - - C C (p) Patient 2 F C C - - - Patient 3 F C - - C C (p) Patient 4 F C - - - - Patient 5 F C - - - - Patient 6 F C - - C C (p) Patient 7 F C - - - C Patient 1 M C - C - - Patient 2 M C - - - - Patient 3 M C - - C C (p) Patient 4 M C - - C C (p) Patient 5 M C - - C C (p) Patient 6 M C - - - - Control 1 F C - - - - Control 2 F C - C C C (p) Control 3 F C - - C C (p) Control 4 F C C - - - Control 5 F C - - - - Control 6 F C - C C C (p) Control 1 M C C - C C (p) Control 2 M C - - C C (p) Control 3 M C - - C C (p) Control 4 M C C - - - Patient 1 F means that patient number 1 was female. M means male. C means that a correction was preformed, - means that no corrections were performed C (p) means that the segmentation was changed to correct the pial surface 18

3.3.1. Subcortical segmentation Significant differences in volume (p<0.05) between the patient group and the control group were found in the left caudate nucleus as well as in the left and right hippocampus when analyzing both genders (table 3.4). When comparing the female groups significant differences in volume were found in the left putamen, the left and right hippocampus as well as in the optic chiasm. When comparing the male groups a significant difference in volume (p<0.05) was found in the left caudate nucleus. Right-left asymmetry was found in several structures for both patients and controls (table 3.5). In three structures, patients putamen, controls caudate nucleus and controls amygdala, the p-value was less than 0.01). Table 3.4. Structures with a significant volume difference (p<0.05) between the patient group and the control group Both genders Structure p-value Median difference Left Caudate nucleus 0.0101 +10 % Left Hippocampus 0.0277 +13 % Right Hippocampus 0.02 +7 % Females Structure p-value Median difference Left Putamen 0.035 +7 % Left Hippocampus 0.035 +12 % Right Hippocampus 0.0221 +5 % Optic Chiasm 0.00117 +16 % Males Structure p-value Median difference Left caudate nucleus 0.00952 +18 % Median difference is normalized to controls. 19

Table 3.5. Structures with a significant (p<0.05) right-left asymmetry Both gender Patients Controls Structure p-value Median difference p-value Median difference Accumbens area n.s. 0.0195 +10 % Amygdala 0.0105-13 % 0.00195-22 % Caudate Nucleus n.s. 0.00586-5 % Choroid plexus 0.0215-7 % n.s. Hippocampus n.s. 0.0488-7 % Putamen 0.00171 +5 % 0.0273 +3 % Females Patients Controls Structure p-value Median difference p-value Median difference Accumbens area 0.0469 +6 % n.s. Amygdala 0.0156-13 % 0.0313-19 % Choroid plexus 0.0313-4 % n.s. Hippocampus n.s. 0.0313-7 % Putamen 0.0469 +5 % n.s. Males Patients Controls Structure p-value Median difference p-value Pallidum 0.0313 +8 % n.s. Putamen 0.0313 +3 % n.s. Median difference Median difference is compared to the right structure. A positive difference is thus equal to a larger left structure. n.s. is not significant 3.3.2. Cortical segmentation Significantly increased cortical thickness (p<0.01, cluster area > 50 mm 2 ) for patients compared to controls was found in the postcentral gyrus in the left hemisphere (fig. 3.5). Significant negative correlation (p<0.01, cluster area > 50 mm 2 ) was found in the fusiform gyrus (fig. 3.6) and at the middle temporal gyrus in the left hemisphere. As the temporal lobe was affected by the truncation the correlation found there was rejected. 20

Figure 3.5. Significant difference between patients and controls at the postcentral gyrus in the left hemisphere (orange cluster). The blue cluster was rejected as it was too small (area < 50 mm 2 ). The diagram shows the cortical thickness at the vertex with the highest t-value. Blue dots are controls and red squares are patients. The subjects on the left side of the diagram are female and the subjects on the right side of the diagram are male. 21

R 2 =0,77 Figure 3.6. Significant correlation between CYBOCS and cortical thickness was found in the fusiform gyrus in the left hemisphere (inferior view). The scale in the color bar is the t-value. The diagram shows the cortical thickness for all patients at the vertex with the highest t-value. 22

4. Discussion 4.1. The evaluation process Mandatory features of the software were listed in order to reduce the subjectivity and to ensure that the most appropriate software actually was chosen. The selected features were based on the purpose of the project, i.e. to be able to extract morphometric results that are as little as possible affected by the truncation. As a morphometric analysis usually involves complete coverage of the brain, no similar study was available to compare the choice of software to. Freesurfer fulfilled the listed criteria well. Common to morphometry in general is the problem that brains of different size and shape has to be warped to a common template in order to permit a pointwise comparison. The misregistrations it can cause may result in false positive as well as false negative results. The alternative is to use the mean result, e.g. mean thickness, of a certain area, e.g. insula, and compare that measurement instead. That method will however introduce other problems. If the areas are chosen to be large, the number of false negatives will increase, as the areas that differ most likely are smaller than the chosen area. Hence, the differences are likely to be reduced because of the averaging. On the other hand, if the areas are chosen to be small, there might be a problem to map the same areas on all subjects, i.e. the same problem as when a pointwise comparison is performed. In this study the two concepts were combined as a smoothing kernel was used after the warping. The smoothing is comparable to the use of a mean result, but with the option to perform a pointwise comparison still available. The false positives should thus be reduced without increasing the number of false negatives in the same extent. 4.2. The validation 4.2.1. Corrections A connection between the truncation of the images and the ICV was found. The effect comes from the fact the ICV was calculated from the determinant of the transformation matrix used in the Talairach transformation. After the transformation the truncated images were found to be consistently smaller in the RL direction (fig. 2.1 left image), i.e. where the images were truncated, compared to the images with complete coverage of the head. This resulted in smaller calculated ICV when the transformation of the truncated images was not corrected. A thorough inspection of the Talairach transformation was thus vital as ICV was used to normalize all volume measurements in the analysis of the OCD patients and the case matched controls. This was the only effect in Freesurfer s processing that could be seen originating from the truncation. The need of other corrections, such as adjusting surfaces, was most likely due to properties of the images and inherent uncertainties of the processing streams employed by Freesurfer. 4.2.2. Subcortical segmentation As the subcortical segmentation operates in the central parts of the brain, it is not affected by the truncation as long as the truncation is small. The structures with a large difference (> 5 %) were those that are harder to segment due to diffuse borders to other structures. It is confirmed by observing the standard deviation that is generally larger for these structures compared to other structures. The result of the validation is promising with regard to using the method in the OCD study in this project as no subcortical structure showed a significant difference between images with complete coverage of the brain and truncated images and the mean difference was generally 23

small. 4.2.3. Cortical segmentation The cortical thickness was more affected by the truncation on the left side than on the right side. This was probably caused by the slightly larger truncation of the left side. A larger truncation can thus be connected to a larger effect on the cortical thickness. That is an expected effect as the most lateral cerebral cortex in the images was thinner or completely absent due to the truncation. However, due to the constraints on the surfaces (smoothness), the pial surface was in most cases slightly outside the WM surface. The measured cortical thickness was thus not zero in these areas although it should have been in some cases. No exact level on the mean and standard deviation of the cortical thickness was set where an area of cortex had to be discriminated. The limits set on the scale (2 % to 10 % for the mean) are however well below the differences displayed in figure 3.5 (approximately 0.9 mm i.e. 30-40 %) and are thus set low enough. If the mere reason of the limits were to display areas that should not be used in the subsequent analysis of the patients they could have been set higher (5 % or maybe 10 % for the mean), but they were instead set to visualize the effect of the truncation. To summarize, the validation of the cortical segmentation showed that the effects from the truncation were concentrated to the areas, which were truncated. It should thus be possible to use the results from rest of the cortex when analyzing the OCD patients and controls. 4.2.4. The truncation The manual truncation performed in the validation part of this project was made to mimic the actual one in the OCD study, i.e. the truncation was performed as if the old scanning parameters would have been used also on the subjects in the validation process. The distribution of head size and shape was more limited in the validation group, i.e. both larger and smaller heads were found among the patients and controls as compared to the validation group. Therefore, some effects caused by the truncation in the patient and control groups may have been overlooked. To overcome this problem a larger group with a wider distribution of head sizes and shapes could be studied. Another alternative is to successively truncate the images used in the validation in order to find a critical level where the truncation starts to influence the results of the morphometry in a substantial way in the rest of the brain. 4.3. Analysis of OCD patients and controls 4.3.1. Corrections The template was usually fitted to the lateral borders of the truncated brain. In the subsequent manual adjustment of the Talairach transformation, the operator had to estimate the lateral extent/shape of the brain based on the rest of the image. Even though this manual procedure was not exact, the ICV estimates depending on the Talairach transformation after adjustments were closer to the true values as seen in the validation. The most common adjustment on the segmentation was to change the classification of some voxels from right vessel to right putamen in order to correct an erroneous pial surface (fig. 2.2). This could have influenced the result when testing for right-left asymmetry for the putamen, as the right one was made somewhat larger. It does certainly affect the mean difference, however, as the result of the test for some of the groups was a slightly larger left structure, the adjustments made should not have resulted in false positive results, but rather weaker p-values and smaller differences. For those groups where a significant difference could not be seen, it may be a result of the adjustments on the segmentation. As a result no significant difference with a larger right putamen should have been missed, as the adjustments 24