The main objective of this thesis is the automatic modeling, understanding and segmentation of diffusively infiltrative tumors known as Diffuse Low-Grade Gliomas. Two approaches exploiting anatomical and spatial prior knowledge have been proposed. We first present the construction of a tumor specific probabilistic atlas describing the tumors' preferential locations in the brain. The proposed atlas constitutes an excellent tool for the study of the mechanisms behind the genesis of the tumors and provides strong spatial cues on where they are expected to appear. The latter characteristic is exploited in a Markov Random Field based segmentation method where the atlas guides the segmentation process as well as characterizes the tumor's preferential location. Second, we introduce a concurrent tumor segmentation and registration with missing correspondences method. The anatomical knowledge introduced by the registration process increases the segmentation quality, while progressively acknowledging the presence of the tumor ensures that the registration is not violated by the missing correspondences without the introduction of a bias. The method is designed as a hierarchical grid-based Markov Random Field model where the segmentation and registration parameters are estimated simultaneously on the grid's control point. The last contribution of this thesis is an uncertainty-driven adaptive sampling approach for such grid-based models in order to ensure precision and accuracy while maintaining robustness and computational efficiency. The potentials of both methods have been demonstrated on a large data-set of heterogeneous Diffuse Low-Grade Gliomas. The proposed methods go beyond the scope of the presented clinical context due to their strong modularity and could easily be adapted to other clinical or computer vision problems.

The microstructure of the brain at the cellular level provides crucial information for the understanding of the function of the brain. A large volume of high-resolution brain image data from 3D microscopy is an essential resource to study detailed microstructures of the brain. Accordingly, we have worked on obtaining high-resolution image data of entire mouse brains using the Knife-Edge Scanning Microscope (KESM). Furthermore, to disseminate these high-resolution whole mouse brain data sets to the neuroscience research community, we developed a web-based brain atlas, the KESM Brain Atlas (KESMBA). To visualize the data sets in 3D while using only a standard web browser, we employed distance attenuation and Google Maps API. The KESMBA is a powerful tool to analyze and share the KESM mouse brain data sets, but the image loading was slow because of the number of raster image (PNG) tiles and the file size. Moreover, since Google Maps API is governed by a commercial license, it does not provide enough flexibility for customization, extension, and mirroring. To solve these issues, we designed and developed a new KESM mouse brain atlas that uses a vector graphics format called Scalable Vector Graphics (SVG) instead of PNG, and OpenLayers API instead of Google Maps API. The SVG-based KESMBA using OpenLayers allows faster navigation and exploration of the KESM data, and more overlay of layers with the 4 times reduced file size compared to PNG tiles. Due to the reduced file size, the SVG-based KESMBA using OpenLayers is 2.45 times faster than the original atlas. By enhancing the performance, the users can more easily access the KESM data. We expect the SVG-based KESMBA to accelerate new discoveries in neuroscience.

In this thesis, we aim to investigate the correlation between myelination and the gestational age for preterm infants, with the former being an important developmental process during human brain maturation. Quantification of myelin requires dedicated imaging, but the conventional magnetic resonance images routinely acquired during clinical imaging of neonates carry signatures that are thought to be associated with myelination. This thesis thus focuses on structural segmentation and spatio-temporal modelling of the so-called myelin-like signals on T2-weighted scans for early prognostic evaluation of the preterm brain. The segmentation part poses the major challenges of this task: insufficient spatial prior information of myelination and the presence of substantial partial volume voxels in clinical data. Specific spatial priors for the developing brain are obtained from either probabilistic atlases or manually annotated training images, but none of them currently include myelin as an individual tissue type. This causes further difficulties in partial volume estimation which depends on the probabilistic atlases of the composing pure tissues. Our key contribution is the development of an expectation-maximisation framework that incorporates an explicit partial volume class whose locations are configured in relation to the composing pure tissues in a predefined region of interest via second-order Markov random fields. This approach resolves the above challenges without requiring any probabilistic atlas of myelin. We also investigate atlas-based whole brain segmentation that generates the binary mask for the region of interest. We then construct a spatio-temporal growth model for myelin-like signals using logistic regression based on the automatic segmentations of 114 preterm infants aged between 29 and 44 gestational weeks. Lastly, we demonstrate the ability of age estimation using the normal growth model in a leave-one-out procedure.

L'objectif principal de cette thèse est la modélisation, compréhension et segmentation automatique de tumeurs diffuses et infiltrantes appelées Gliomes Diffus de Bas Grade. Deux approches exploitant des connaissances a priori de l'ordre spatial et anatomique ont été proposées. Dans un premier temps, la construction d'un atlas probabiliste qui illustre les positions préférentielles des tumeurs dans le cerveau est présentée. Cet atlas représente un excellent outil pour l'étude des mécanismes associés à la genèse des tumeurs et fournit des indications sur la position probable des tumeurs. Cette information est exploitée dans une méthode de segmentation basée sur des champs de Markov aléatoires, dans laquelle l'atlas guide la segmentation et caractérise la position préférentielle de la tumeur. Dans un second temps, nous présentons une méthode pour la segmentation de tumeur et le recalage avec absence de correspondances simultanés. Le recalage introduit des informations anatomiques qui améliorent les résultats de segmentation tandis que la détection progressive de la tumeur permet de surmonter l'absence de correspondances sans l'introduction d'un a priori. La méthode est modélisée comme un champ de Markov aléatoire hiérarchique et à base de grille sur laquelle les paramètres de segmentation et recalage sont estimés simultanément. Notre dernière contribution est une méthode d'échantillonnage adaptatif guidé par les incertitudes pour de tels modèles discrets. Ceci permet d'avoir une grande précision tout en maintenant la robustesse et rapidité de la méthode. Le potentiel des deux méthodes est démontré sur de grandes bases de données de gliomes diffus de bas grade hétérogènes. De par leur modularité, les méthodes proposées ne se limitent pas au contexte clinique présenté et pourraient facilement être adaptées à d'autres problèmes cliniques ou de vision par ordinateur. / The main objective of this thesis is the automatic modeling, understanding and segmentation of diffusively infiltrative tumors known as Diffuse Low-Grade Gliomas. Two approaches exploiting anatomical and spatial prior knowledge have been proposed. We first present the construction of a tumor specific probabilistic atlas describing the tumors' preferential locations in the brain. The proposed atlas constitutes an excellent tool for the study of the mechanisms behind the genesis of the tumors and provides strong spatial cues on where they are expected to appear. The latter characteristic is exploited in a Markov Random Field based segmentation method where the atlas guides the segmentation process as well as characterizes the tumor's preferential location. Second, we introduce a concurrent tumor segmentation and registration with missing correspondences method. The anatomical knowledge introduced by the registration process increases the segmentation quality, while progressively acknowledging the presence of the tumor ensures that the registration is not violated by the missing correspondences without the introduction of a bias. The method is designed as a hierarchical grid-based Markov Random Field model where the segmentation and registration parameters are estimated simultaneously on the grid's control point. The last contribution of this thesis is an uncertainty-driven adaptive sampling approach for such grid-based models in order to ensure precision and accuracy while maintaining robustness and computational efficiency. The potentials of both methods have been demonstrated on a large data-set of heterogeneous Diffuse Low-Grade Gliomas. The proposed methods go beyond the scope of the presented clinical context due to their strong modularity and could easily be adapted to other clinical or computer vision problems.

Understanding mysteries of a brain represents one of the great challenges for modern science. Functional magnetic resonance imaging (fMRI) has two features that make it unique amongst other imaging modalities used in behavioral neuroscience. First, it can be entirely non-invasive and second, fMRI has the spatial and temporal resolution to resolve patterns of neuronal activity across the entire brain in less than a minute. fMRI indirectly detects neural activity in different parts of the brain by comparing contrast in MR signal intensity prior to and following stimulation. Areas of the brain with increased synaptic and neuronal activity require increased levels of oxygen to sustain this activity. Enhanced brain activity is accompanied by an increase in metabolism followed by increases in blood flow and blood volume. The enhanced blood flow usually exceeds the metabolic demand exposing the active brain area to high level of oxygenated hemoglobin. Oxygenated hemoglobin increases the MR signal intensity that can be detected in MR scanner. This relatively straight forward scenario is, unfortunately, oversimplified. The fMRI signal change to noise ratio is extremely small. In this work a quantitative analysis strategy to analyze fMRI data was successfully developed, implemented and optimized for the rat brain. Therein, each subject is registered or aligned to a complete volume-segmented rat atlas. The matrices that transformed the subject's anatomy to the atlas space are used to embed each slice within the atlas. All transformed pixel locations of the anatomy images are tagged with the segmented atlas major and minor regions creating a fully segmented representation of each subject. This task required the development of a full 3D surface atlas based upon 2D non-uniformly spaced 2D slices from an existing atlas. A multiple materials marching cube (M3C) algorithm was used to generate these 1277 subvolumes. After this process, they were coalesced into a dozen major zones of the brain (amygdaloid complex, cerebrum, cerebellum, hypothalamus, etc.). Each major brain category was subdivided into approximately 10 sub-major zones. Many scientists are interested in behavior and reactions to pain, pleasure, smell, for example. Consequently, the 3D volume atlas was segmented into functional zones as well as the anatomical regions. A utility (program) called Tree Browser was developed to interactively display and choose different anatomical and/or functional areas. Statistical t-tests are performed to determine activation on each subject within their original coordinate system. Due to the multiple t-test analyses performed, a false-positive detection controlling mechanism was introduced. A statistical composite of five components was created for each group. The individual analyses were summed within groups. The strategy developed in this work is unique as it registers segments and analyzes multiple subjects and presents a composite response of the whole group. This strategy is robust, incredibly fast and statistically powerful. The power of this system was demonstrated by mapping the olfactory system of a rat brain. Synchronized changes in neuronal activity across multiple subjects and brain areas can be viewed as functional neuro-anatomical circuits coordinating the thoughts, memories and emotions for particular behaviors using this fMRI module.

Medical Image Analysis has emerged as an important field in the computer vision community. In this thesis, two important issues in medical imaging are addressed and a solution for each is derived and synergistically combined as one coherent system. Firstly, a novel approach is proposed for High Resolution Volume (HRV) construction by combining different frequency components at multiple levels, which are separated by using a multi-resolution pyramid structure. Current clinical imaging protocols make use of multiple orthogonal low resolution scans to measure the size of the tumor. The highly anisotropic data result in difficulty and even errors in tumor assessment. In previous approaches, simple interpolation has been used to construct HRVs from multiple low resolution volumes (LRVs), which fail when large inter-plane spacing is present. In our approach, Laplacian pyramids containing band-pass contents are first computed from registered LRVs. The Laplacian images are expanded in their low resolution axes separately and then fused at each level. A Gaussian pyramid is recovered from the fused Laplacian pyramid, where a volume at the bottom level of the Gaussian pyramid is the constructed HRV. The effectiveness of the proposed approach is validated by using simulated images. The method has also been applied to real clinical data and promising experimental results are demonstrated. Secondly, a new knowledge-based framework to automatically quantify the volume of enhancing tissue in brain MR images is proposed. Our approach provides an objective and consistent way to evaluate disease progression and assess the treatment plan. In our approach, enhanced regions are first located by comparing the difference between the aligned set of pre- and post-contrast T1 MR images. Since some normal tissues may also become enhanced by the administration of Gd-DTPA, using the intensity difference alone may not be able to distinguish normal tissue from the tumor. Thus, we propose a new knowledge-based method employing knowledge of anatomical structures from a probabilistic brain atlas and the prior distribution of brain tumor to identify the real enhancing tissue. Our approach has two main advantages. i) The results are invariant to the image contrast change due to the usage of the probabilistic knowledge-based framework. ii) Using the segmented regions instead of independent pixels facilitates an approach that is much less sensitive to small registration errors and image noise. The obtained results are compared to the ground truth for validation and it is shown that the proposed method can achieve accurate and consistent measurements.

The aim of this doctoral thesis was to study the influence of memory, pain, age and education on the regional cerebral blood flow (rCBF), i.e. brain function, in early Alzheimer's disease (AD) and in chronic neck pain patients in comparison to healthy controls and in healthy elderly per se. This was done by optimizing single photon emission computed tomography (SPECT) as a method to study rCBF with the tracer Technetium-99m (99mTc) hexamethylpropyleneamine oxime (HMPAO) and by matching all image data to a brain atlas before evaluation. The rCBF-SPECT was evaluated and developed to obtain higher diagnostic accuracy in AD and in chronic neck pain patients it was used to study basic pain related cerebral processes in chronic pain of different origin. A new semimanual registration method, based on fiducial marker, suitable for investigations with low spatial resolution was developed. The method was used to reconstruct images with an improved attenuation and scatter correction by using an attenuation-map calculated from the patients' previously acquired CT images. The influence of age and education on rCBF was evaluated with statistical parametric mapping, SPM in healthy elderly. The main findings were age related changes in rCBF in regions close to interlobar and interhemispheric space but not in regions typically affected in early AD, except for the medial temporal lobe. The theory of a 'cognitive reserve' in individuals with a longer education was supported with findings in the lateral temporal lobe, a region related to semantic memory, and in the frontal lobe. A cross-sectional study of chronic neck pain patients showed extensive rCBF changes in coping related regions in a non-traumatic pain patients compared to both healthy and a pain group with a traumatic origin, i.e. whiplash syndrome. The whiplash group displayed no significant differences in rCBF in comparison with the healthy controls. This suggests different pain mechanisms in these groups. The AD-patients showed a significantly lower rCBF in temporoparietal regions including left hippocampus. These changes were associated to episodic memory performance, and especially to face recognition. The diagnostic sensitivity for AD was high. The face recognition test (episodic memory) was used in AD patients to improve the sensitivity of method, i.e. memory-provoked rCBF-SPECT (MP-SPECT). The results were compared to healthy controls and the reductions of rCBF in temporoparietal regions were more pronounced in mild AD during provocation. Memory provocation increased the sensitivity of AD-related rCBF changes at group level. If a higher sensitivity for AD at the individual level is verified in future studies, a single MP-SPECT study might then be of help to set diagnosis earlier. In conclusion rCBF in temporoparietal regions are associated to an impaired episodic memory in early AD. Changes in these regions do not have a strong connection to chronological age. The diagnostic sensitivity of rCBF-SPECT in AD is high and there is a potentially higher sensitivity if memory provoked investigations are used. The findings in this thesis have given an increased knowledge of underlying cerebral pain processing in non-traumatic and traumatic (whiplash) neck pain. Preliminary results supporting the theory of 'cognitive reserve' by showing a correlation between long education and preserved rCBF was found in healthy elderly.

Les images angiographiques sont utilisés pour différentes tâches comme le diagnostique, le suivie de pathologies et la planification d'interventions chirurgicales. Toutefois, en raison du faible ratio signal sur bruit et le contenu complexe des données (informations clairsemées), l'analyse des images angiographiques est une tâche fastidieuse et source d'erreurs. Ces différentes considérations ont motivé le développement de nombreuses techniques d'analyse.Les travaux de cette thèse s'organisent autour de deux axes de recherches : d'une part la segmentation des images angiographiques, et d'autre part la modélisation des réseaux vasculaires cérébraux. En segmentation, l'automatisation induit généralement un coût de calcul élevé, alors que les méthodes interactives sont difficiles à utiliser en raison de la dimension et de la complexité des images. Ces travaux présentent un compromis entre les deux approches, en utilisant le concept de segmentation à base d'exemple. Cette stratégie qui utilise les arbres de coupes de façon non standard,conduit à des résultats satisfaisant, lorsqu'elle est appliqué sur des données d'ARM cérébrales 3D. Les approches existantes, en modélisation, reposent exclusivement sur des informations relatives aux vaisseaux. Ces travaux ont exploré une nouvelle voie, consistant à utiliser à la fois les informations vasculaires et morphologiques ( c-à-d structures cérébrales) pour améliorer la précision et la pertinence des atlas obtenus. Les expériences soulignent des améliorations dans les principales étapes du processus de création de l'atlas impacté par l'utilisation de l'information morphologique. Un exemple d'atlas cérébraux a été réalisé. / Angiographie images are useful data for several tasks, e.g., diagnosis, pathology follow-up or surgery planning. However, due to low SNR (noise,artifacts), and complex semantic content (sparseness), angiographie image analysis is a time consurning and error prone task. These consideration have motivated the development of numerous vesse! filtering, segmentation, or modeling techinques.This thesis is organized around two research areas : the segmentation anù the moùeling. Segmentation of cerebral vascular networks from 3D angiographie data remains a challenge. Automation generally induces a high computational cost and possible errors, white interactive methods are hard to use due to the dimension and the complexity of images. This thesis presents a compromise between both approaches by using the concept of example-based segmentation. This strategy, which uses component-trees in a non-standard fashion, leads to promising results, when applied on cerebral MR angiographie data. The generation of cerebrovascular atlases remains a complex and infrequently considered issue. The existing approaches rely on information exclusively related to the vessels. This thesis investigate a new way, consisting of using both vascular and morphological information (i.e. Cerebral structures) to improve the accuracy and relevance of the obtaines vascular atlases. Experiments emphasize improvments in the main steps of the atlas generation process impacted by the use of the morphological information. An example of cerebrovascular atlas obtained from a dataset of MRAs acquired form different acquisition devices has been provided.

La thèse explore trois questions méthodologiques en imagerie de diffusion (DTI) clinique du cerveau, dans le contexte d’une étude sur le VIH. La première question est comment améliorer la résolution du DTI. Le deuxième problème est comment créer un atlas multimodal spécifique à la population. La troisième question porte sur le calcul des statistiques pour comparer les zones de matière blanche entre les contrôles et patients. Les DTI cliniques ont une résolution spatiale et un rapport signal sur bruit faibles, ce qui rend difficile le calcul de statistiques significatives. Nous proposons un algorithme de super-résolution pour améliorer la résolution qui utilise un a priori spatial anisotrope. Cette méthode démontre une amélioration de l’anisotropie fractionnelle et de la tractographie. Pour normaliser spatialement les images du cerveau dans un système de coordonnées commun, nous proposons ensuite de construire un atlas multimodal spécifique á la population. Ceci permet de créer un atlas probabiliste de la matière blanche qui est consistant avec l’atlas anatomique. Cet atlas peut être utilisé pour des statistiques basées sur des régions d’intérêt ou pour le raffinement d’une segmentation. Enfin, nous améliorons les résultats de la méthode TBSS (Tract-Based Spatial Statistics) en utilisant le recalage des images DTI. Contrairement á la méthode TBSS traditionnelle, nous utilisons ici des statistiques multivariées. Nous montrons que ceci permet de détecter des différences dans les régions de matière blanche qui étaient non significatives auparavant, et de les corréler avec les scores des tests neuropsychologiques. / The thesis explores three major methodological questions in clinical brain DTI, in the context of a clinical study on HIV. The first question is how to improve the DTI resolution. The second problem addressed in the thesis is how to create a multimodal population specific atlas. The third question is on the computation of statistics to compare white matter (WM) regions among controls and HIV patients. Clinical DTIs have low spatial resolution and signal-to-noise ratio making it difficult to compute meaningful statistics. We propose a super-resolution (SRR) algorithm for improving DTI resolution. The SRR is achieved using anisotropic regularization prior. This method demonstrates improved fractional anisotropy and tractography. In order to spatially normalize all images in a consistent coordinate system, we create a multimodal population specific brain atlas using the T1 and DTI images from a HIV dataset. We also transfer WM labels from an existing white matter parcellation map to create probabilistic WM atlas. This atlas can be used for region of interest based statistics and refining manual segmentation. On the statistical analysis side, we improve the existing tract based spatial statistics (TBSS) by using DTI based registration for spatial normalization. Contrary to traditional TBSS routines, we use multivariate statistics for detecting changes in WM tracts. With the improved method it is possible to detect differences in WM regions and correlate it with the neuropschylogical test scores of the subjects.

Magister Scientiae - MSc / Neurological diseases are socially disabling and often mortal. To efficiently combat these diseases, a deep understanding of involved cellular processes, gene functions and anatomy is required. However, differential regulation of genes across anatomy is not sufficiently well understood. This study utilized large-scale gene expression data to define the regulatory networks of genes expressing in the hippocampus to which multiple disease pathologies may be associated. Specific aims were: ident i fy key regulatory transcription factors (TFs) responsible for observed gene expression patterns, reconstruct transcription regulatory networks, and prioritize likely TFs responsible for anatomically restricted gene expression. Most of the analysis was restricted to the CA3 sub-region of Ammon’s horn within the hippocampus. We identified 155 core genes expressing throughout the CA3 sub-region and predicted corresponding TF binding site (TFBS) distributions. Our analysis shows plausible transcription regulatory networks for twelve clusters of co-expressed genes. We demonstrate the validity of the predictions by re-clustering genes based on TFBS distributions and found that genes tend to be correctly assigned to groups of previously identified co-expressing genes with sensitivity of 67.74% and positive predictive value of 100%. Taken together, this study represents one of the first to merge anatomical architecture, expression profiles and transcription regulatory potential on such a large scale in hippocampal sub-anatomy. / South Africa