Pedestrian detection is a canonical instance of object detection that remains a popular topic of research and a key problem in computer vision due to its diverse applications. These applications have the potential to positively improve the quality of life. In recent years, the number of approaches to detecting pedestrians in monocular and binocular images has grown steadily. However, the use of multispectral imaging is still uncommon. This thesis work presents a novel approach to data and feature fusion of a multispectral imaging system for pedestrian detection. It also includes the design and building of a test rig which allows for quick data collection of real-world driving. An application of the mathematical theory of trifocal tensor is used to post process this data. This allows for pixel level data fusion across a multispectral set of data. Performance results based on commonly used SVM classification architectures are evaluated against the collected data set. Lastly, a novel cascaded SVM architecture used in both classification and detection is discussed. Performance improvements through the use of feature fusion is demonstrated.

Magnetic resonance imaging (MRI) has transformed the world of non-invasive imaging for diagnostic purposes. Modern techniques such as diffusion weighted imaging (DWI), diffusion tensor imaging (DTI), and diffusion spectrum imaging (DSI) have been used to reconstruct fiber pathways of the brain - providing a graphical picture of the so-called "connectome." However, there exists controversy in the literature as to the accuracy of the diffusion tractography reconstruction. Although various attempts at histological validation been attempted, there is still no 3D histological pathway validation of the fiber bundle trajectories seen in diffusion MRI. Such a validation is necessary in order to show the viability of current DSI tractography techniques in the ultimate goal for clinical diagnostic application. This project developed methods to provide this 3D histological validation using the rhesus monkey motor pathway as a model system. By injecting biotinylated dextran amine (BDA) tract tracer into the hand area of primary motor cortex, brain section images were reconstructed to create 3D fiber pathways labeled at the axonal level. Using serial coronal brain sections, the BDA label was digitized with a high resolution digital camera to create image montages of the fiber pathway with individual sections spaced at 1200 micron intervals through the brain. An MRI analysis system, OSIRX, was then used to reconstruct these sections into a 3D volume. This is an important technical step toward merging the BDA fiber tract histology with diffusion MRI tractography of the same brain, enabling identification of the valid and inaccurate aspects of diffusion fiber reconstruction. This will ultimately facilitate the use of diffusion MRI to quantify tractography, non-invasively and in vivo, in the human brain.

OBJECTIFS : de précédentes études ont déjà démontré le potentiel de l’imagerie en tenseur de diffusion (DTI) pour détecter des anomalies médullaires dans la sclérose en plaques (SEP). L’objectif de ce travail était d’appliquer les techniques de DTI à des patients atteints de SEP et présentant une lésion médullaire symptomatique, afin de mettre en évidence une corrélation entre les variations des paramètres DTI et l’évolution clinique, et d’identifier des facteurs pronostiques prédictifs en DTI. METHODE : réalisation d’une étude prospective monocentrique chez des patients ayant une poussée médullaire de SEP traitée par corticoïdes intraveineux. Les patients étaient évalués cliniquement ainsi qu’en IRM (conventionnelle et DTI), à l’inclusion et à 3 mois. RESULTATS : seize patients ont été inclus. Lors du suivi à 3 mois, 12 patients étaient cliniquement améliorés. Tous sauf un avaient des valeurs de Fraction d’Anisotropie (FA) plus basses que les sujets normaux, au sein des lésions inflammatoires et dans la moelle apparemment normale, à l’inclusion ou lors du suivi à 3 mois. Les patients améliorés à 3 mois avaient une réduction significative de la Diffusivité Radiale (DR) (p=0,02) dans les lésions au cours du suivi. Ils avaient également une réduction significative du Coefficient de Diffusion Apparent moyen (p=0,002), de la Diffusivité Axiale (p=0,02) et de la DR (p=0,02) et une augmentation significative des valeurs de FA (p=0,02) dans la moelle apparemment normale. Les huit patients améliorés sur leur score ASIA (American Spinal Injury Association) sensitif à 3 mois avaient une FA initiale au sein des lésions inflammatoires significativement plus élevée (p=0,009) et une DR significativement plus basse (p=0,04) que les patients non améliorés. CONCLUSION : dans la SEP, l’IRM médullaire avec séquences en DTI détecte plus d’anomalies que l’IRM conventionnelle T2. Ces anomalies sont évolutives et corrélées au pronostic clinique, notamment celles observées dans la moelle épinière apparemment normale. Une diminution moins marquée des valeurs de FA initiales et plus marquée de la DR au sein des lésions inflammatoires est associée à un meilleur pronostic fonctionnel. / PURPOSE: the potential of Diffusion Tensor Imaging (DTI) to detect spinal cord abnormalities in patients with multiple sclerosis has already been demonstrated. The objective of this study was to apply DTI techniques to Multiple Sclerosis patients with a recently diagnosed spinal cord lesion, in order to demonstrate a correlation between variations of DTI parameters and clinical outcome, and to try to identify DTI parameters predictive of outcome. METHODS: a prospective single-centre study of patients with spinal cord relapse treated by intravenous steroid therapy. Patients were assessed clinically and by conventional MRI with DTI sequences at baseline and at 3 months. RESULTS: Sixteen patients were recruited. At 3 months, 12 patients were clinically improved. All but one patient had lower Fractional Anisotropy (FA) and Apparent Diffusion Coefficient (ADC) values than normal subjects in either inflammatory lesions or normal-appearing spinal cord. Patients who improved at 3 months presented a significant reduction in the Radial Diffusivity (p=0.05) in lesions during the follow-up period. They also had a significant reduction in the mean ADC (p=0.002), Axial Diffusivity (p=0.02), Radial Diffusivity (p=0.02) and a significant increase in FA values (p=0.02) in normal-appearing spinal cord. Patients in whom the American Spinal Injury Association sensory score improved at 3 months showed a significantly higher FA (p=0.009) and lower Radial Diffusivity (p=0.04) in inflammatory lesion at baseline compared to patients with no improvement. CONCLUSION: DTI MRI detects more extensive abnormalities than conventional T2 MRI. A less marked decrease in FA value and more marked decrease in Radial Diffusivity inside the inflammatory lesion were associated with better outcome.

L’IRM de diffusion est une modalité d’imagerie médicale qui suscite un intérêt croissant dans larecherche en neuro-imagerie. Elle permet de caractériser in vivo l’organisation neuronale et apportepar conséquent de nouvelles informations sur les fibres de la matière blanche. En outre, il a étémontré que chaque région corticale a une signature spécifique pouvant être décrite par des mesuresde connectivité. Notre travail de recherche a ainsi porté sur la conception d’une méthode deparcellisation du cortex entier à partir de ces métriques. En se basant sur de précédents travaux dudomaine (thèse de P. Roca 2011), ce travail propose une nouvelle analyse de groupe permettantl’obtention d’une segmentation individuelle ou moyennée sur la population d'étude. Il s’agit d’unproblème difficile en raison de la variabilité interindividuelle présente dans les données. Laméthode a été testée et évaluée sur les 80 sujets de la base ARCHI. Des aspects multimodaux ontété abordés pour comparer nos parcellisations structurelles avec d’autres parcellisations ou descaractéristiques morphologiques calculées à partir des modalités présentes dans la base de données.Une correspondance avec la variabilité de l’anatomie corticale, ainsi qu’avec des parcellisations dedonnées d’IRM fonctionnelle, a pu être montrée, apportant une première validationneuroscientifique. / Résumé anglais :Diffusion MRI is a medical imaging modality of great interest in neuroimaging research. Thismodality enables the characterization in vivo of neuronal organization and thus providinginformation on the white matter fibers. In addition, each cortical region has been shown to have aspecific signature, which can be described by connectivity measures. Our research has focused onthe design of a whole cortex parcellation method driven by these metrics. Based on the previouswork of P. Roca 2011, a new group analysis is proposed to achieve an individual or populationaveraged segmentation. This is a difficult problem due to the interindividual variability present inthe data. The method was tested and evaluated on the 80 subjects of the ARCHI database.Multimodal aspects were investigated to compare the proposed structural parcelliations with otherparcellations or morphological characteristics derived from the modalities present in the database. Aconnection between the variability of cortical anatomy and parcellations of the functional MRI datawas demonstrated, providing a first neuroscientist validation.

Le développement de méthodes de décomposition de tableaux multi-dimensionnels suscite toujours autant d'attention, notamment d'un point de vue applicatif. La plupart des algorithmes, de décompositions tensorielles, existants requièrent une estimation du rang du tenseur et sont sensibles à une surestimation de ce dernier. Toutefois, une telle estimation peut être difficile par exemple pour des rapports signal à bruit faibles. D'un autre côté, estimer simultanément le rang et les matrices de facteurs du tenseur ou du tenseur cœur n'est pas tâche facile tant les problèmes de minimisation de rang sont généralement NP-difficiles. Plusieurs travaux existants proposent d'utiliser la norme nucléaire afin de servir d'enveloppe convexe de la fonction de rang. Cependant, la minimisation de la norme nucléaire engendre généralement un coût de calcul prohibitif pour l'analyse de données de grande taille. Dans cette thèse, nous nous sommes donc intéressés à l'approximation d'un tenseur bruité par un tenseur de rang faible. Plus précisément, nous avons étudié trois modèles de décomposition tensorielle, le modèle CPD (Canonical Polyadic Decomposition), le modèle BTD (Block Term Decomposition) et le modèle MTD (Multilinear Tensor Decomposition). Pour chacun de ces modèles, nous avons proposé une nouvelle méthode d'estimation de rang utilisant une métrique moins coûteuse exploitant la parcimonie de groupe. Ces méthodes de décomposition comportent toutes deux étapes : une étape d'estimation de rang, et une étape d'estimation des matrices de facteurs exploitant le rang estimé. Des simulations sur données simulées et sur données réelles montrent que nos méthodes présentent toutes une plus grande robustesse à la présence de bruit que les approches classiques. / Last decades, tensor decompositions have gained in popularity in several application domains. Most of the existing tensor decomposition methods require an estimating of the tensor rank in a preprocessing step to guarantee an outstanding decomposition results. Unfortunately, learning the exact rank of the tensor can be difficult in some particular cases, such as for low signal to noise ratio values. The objective of this thesis is to compute the best low-rank tensor approximation by a joint estimation of the rank and the loading matrices from the noisy tensor. Based on the low-rank property and an over estimation of the loading matrices or the core tensor, this joint estimation problem is solved by promoting group sparsity of over-estimated loading matrices and/or the core tensor. More particularly, three new methods are proposed to achieve efficient low rank estimation for three different tensors decomposition models, namely Canonical Polyadic Decomposition (CPD), Block Term Decomposition (BTD) and Multilinear Tensor Decomposition (MTD). All the proposed methods consist of two steps: the first step is designed to estimate the rank, and the second step uses the estimated rank to compute accurately the loading matrices. Numerical simulations with noisy tensor and results on real data the show effectiveness of the proposed methods compared to the state-of-the-art methods.

Objective: Following a sport related concussion (SRC) visible symptoms generally dissipate in 7-10 days post-injury. However, little is known about the cumulative effects of SRCs both in terms of structural damage to the white matter of the brain and neurocognitive performance. To address this issue, the relationship between the number of SRCs (frequency), axonal white matter (WM) damage and neurocognitive performance was examined. There were three predictions. First, increases in SRC frequency will be associated with decreases in performance on neurocognitive tests. Second, the frequency of SRC will be associated with axonal injury measured three WM tracts: the corpus callosum, the fronto-occipital fasciculus and the inferior longitudinal fasciculus. Third, less accurate and slower performance on a response inhibition task (STOP-IT) will be associated with greater axonal injury. Methods: A cross-sectional correlational design was utilised. Participants were rugby players with a history of SRC, rugby players with no history of SRC and control athletes (N=40) who completed a neurocognitive test battery and had a DTI brain scan. The neurocognitive battery consisted of the following standardised tests: Speed and Capacity of Language Processing Test, CogState Electronic Battery, Stroop Colour and Word Test, Controlled Oral Word Association Test, the Trail Making Test and the experimental test STOP-IT Electronic Test. White matter axonal injury was measured by DTI using fractional anisotropy (FA) and mean diffusivity (MD) metrics. The DTI data was processed using FSL to extract FA and MD DTI metrics in three a-priori regions of interest. Results: Spearman’s correlation analyses did not find significant associations between SRC frequency and neurocognitive performance on the FAS (rs=0.053, 95% CI [-0.27, 0.36]), TMT-A (rs=0.058, 95% CI [-0.26, 0.37]), TMT-B (rs= -0.046, 95% CI [-0.27, 0.36]) and the Stroop Interference (rs= -0.25, 95% CI [-0.07, 0.52]). Similarly, no significant Spearman’s correlations were found between SRC frequency and the computerised neurocognitive tests STOP-IT-SSRT (rs= -0.04, 95% CI [-0.28, 0.35])), STOP-IT–Accuracy (rs= -0.05, 95% CI [-0.27, 0.36]), CogState Detection subtest (rs= -0.15, 95% CI [-0.17, 0.44]), CogState Identification subtest (rs= -0.065, 95% CI [-0.26, 0.37]), CogState One card learning subtest (rs= 0.24, 95% CI [-0.08, 0.52]) or the CogState One back task subtest (rs= 0.06, 95% CI [-0.26, 0.37]). In terms of the DTI data there were no significant associations between SRC frequency and axonal injury measured by FA values in the CC (rs= 0.005, 95% CI [-0.31, 0.32]), ILF (rs= 0.028, 95% CI [-0.29, 0.34]) or FOF (rs= -0.022, 95% CI [-0.30, 0.33]). The same was pattern was found for MD values in the CC (rs= 0.081, 95% CI [-0.24, 0.39]), ILF (rs= -0.16, 95% CI [-0.16, 0.45]) or FOF (rs= -0.15, 95% CI [-0.17, 0.44]) Finally, there were no significant Spearman’s correlations between axonal injury FA values and the STOP-IT SSRT in any of the ROIs: CC (rs= 0.005, 95% CI [-0.31, 0.32]), ILF (rs= 0.028, 95% CI [-0.29, 0.34]) or FOF (rs= -0.022, 95% CI [-0.30, 0.33]). Equally, there were no significant correlations between MD values STOP-IT SSRT in the CC (rs= -0.028, 95% CI [-0.29, 0.34]), ILF (rs= -0.16, 95% CI [-0.16, 0.45]) or FOF (rs= -0.15, 95% CI [-0.17, 0.44]). Likewise, there were no significant Spearman’s correlations between accuracy on the STOP-IT and FA values and in any of the ROIs: CC (rs= 0.19, 95% CI [-0.13, 0.48]), ILF (rs= -0.045, 95% CI [-0.27, 0.35]) and FOF (rs= -0.032, 95% CI [-0.29, 0.34]), or MD values in the CC (rs= -0.11, 95% CI [-0.21, 0.41]), ILF (rs= 0.017, 95% CI [-0.30, 0.33]) or FOF (rs= 0.082, 95% CI [-0.24, 0.39]). This study did not find support for the hypothesis that cumulative SRCs are associated with poorer performance on neurocognitive tests or with axonal injury as measured by FA and MD DTI metrics. Conclusion: The null findings suggest that there are no cumulative effects of SRCs. The current findings are inconsistent with previous cross-sectional research that indicates that there are long-term changes to diffusivity measures present after single SRCs as well as cumulative effects in contact sport athletes. Likewise they are at odds with evidence suggesting that after three SRCs neurocognitive performance can be affected. The study needs to be extended to include a larger sample to ensure the results are not due to low statistical power.

This thesis develops novel analytic models of scalar-tensor theories with quadratic coupling. In this framework, the coupling strength between scalar and matter is regulated in a way that allows the vacuum expectation value to vanish for low matter densities while becoming non-vanishingly large in the high-density regime. This results in significant deviations from the predictions of General Relativity in the strong-gravity regime. In astrophysics, we addressed the core-collapse supernova problem to account for the apparently missing energy required to explain the observed powerful explosions. We assumed a small, massless scalar gravitational field, thus allowing General Relativity to be recovered in the weak-gravity asymptotic limit. The non-trivial effects coming from the coupling function in the presence of a high-density field were analyzed at the instant of neutron star formation. Our results show that the scalar gravitational field evolves from a cosmological value to a new equilibrium via a Higgs-like mechanism. Additionally, the calculations associated with the gravitational binding energy shift and relevant relaxation timescale are explicitly shown. The full theory space of the model was also investigated for positive values of the coupling parameter. We studied a mechanism to address the stalled shock issue in core-collapse scenarios, which involved the application of sufficiently large positive values to the coupling parameter. Our results show that pulsating neutron stars act like optical cavities in which resonant scalar waves are parametrically amplified. It implies that the surface of a neutron star acts like an anti-phase reflector, releasing traveling scalar gravitational waves similar to an optical laser. In cosmology, the same framework was applied to a generic Friedman-Robertson-Walker universe involving general metric coupling and scalar potential functions. In cosmology, the same framework was applied to a generic Friedman-Robertson-Walker universe involving general metric coupling and scalar potential functions. We developed a mechanism which allowed the scalar field to be dynamically trapped, thus generating a potential capable of driving primordial inflation. Our results show that a trapped scalar field produces non-trivial dynamical consequences when applied to standard cosmology. Additionally, our analytic solutions for the generic inflationary behaviour, produce acceptable duration and e-foldings, thus recovering the Hubble parameter which is consistent with the present-day value. A feature of our cosmological model is that the universe can undergo several accelerating or decelerating phases, even though the scalar potential and metric coupling are monotonic functions overall. As this is important for the current dark energy problem, the quasi-static motion of the gravitational field induced by the scalar potential in the early universe, is investigated for a small value of the scalar field with normalized metric at the present time. Our results show that a variable Lambda Cold Dark Matter universe emerges naturally from the quadratic model.

The logistical complexities of performing multi-site longitudinal diffusion-weighted imaging (DWI) studies requires careful construction of analysis tools and procedures. Proposed clinical trials for therapies in neurodegenerative disease are known to re- quire several hundred subjects, thus mandating multiple site participation to obtain sufficient sample sizes. DWI is an important tool for monitoring diffusivity properties of white matter (WM) in disease progression. The multi-site nature of clinical trials requires new strategies in DWI processing and analysis to reliably measure longitudi- nal WM changes. This work describes the process of developing and validating robust analysis methodologies to process multi-site DWI data in a rare, neurodegenerative disease. Key processing components to accomplish a robust DWI processing system include: DICOM conversion, automated quality control, unbiased atlas construction, fiber tracking, and statistical analysis. Extensive validation studies were performed to characterize methodological results within and across the common confounds inherent in multi-site clinical trials. The conversion and automated quality control tools optimized for this work both enhanced the ability to reliably obtain repeat diffusion tensor image (DTI) scalar measurements in a reliability analysis of healthy controls scanned at multiple sites using multiple scanner vendors. A DTI scalar analysis performed on focused WM regions showed it was possible to detect significant mean differences of DTI scalars among separate groups of a neurodegenerative disease population. The DTI scalar analysis paved the way for an atlas-based cross-sectional fiber tracking analysis. In the cross-sectional fiber tracking analysis, multi-site data was brought into the same space, making major fiber tracts terminating in the focused WM regions of the scalar analysis from all participants comparable. Significant differences in diffusivity were found throughout each tract among separate groups of the neurodegenerative disease population. In addition, multiple neuropsychological cognitive variables that have a documented ability to track disease progression of the neurodegenerative disease, strongly correlated with many of the DTI scalars in each tract. The findings of the cross-sectional fiber tracking analysis were reinforced by similar findings produced by a longitudinal fiber tracking analysis. Collectively, these findings suggest that cogni- tive deficits seen in the neurodegenerative disease population could be explained by changes in diffusivity of the tracts explored in this work. In addition to the longi- tudinal fiber tracking analysis examining diffusivity, methods for a WM morphology analysis using parallel transport to apply longitudinal volume changes to a template was explored.

Digital medical atlases can contain anatomical information which is valuable for medical doctors in diagnosing and treating illnesses. The increased availability of such atlases has created an interest for computer algorithms which are capable of integrating such atlas information into patient specific dataprocessing. The field of medical image registration aim at calculating how to match one medical image to another. Here the atlas information could give important hints of which kinds of motion are plausible in different locations of the anatomy. Being able to incorporate such atlas specific information could potentially improve the matching of images and plausibility of image registration - ultimately providing a more correct information on which to base health care diagnosis and treatment decisions. In this licentiate thesis a generic signal processing framework is derived : Global Linear Optimization (GLO). The power of the GLO framework is first demonstrated quantitatively in a very high performing image denoiser. Important proofs of concepts are then made deriving and implementing three important capabilities regarding adaptive filtering of vector fields in medica limage registration: Global regularization with local anisotropic certainty metric. Allowing sliding motion along organ and tissue boundaries. Enforcing an incompressible motion in specific areas or volumes. In the three publications included in this thesis, the GLO framework is shown to be able to incorporate one each of these capabilities. In the third and final paper a demonstration is made how to integrate more and more of the capabilities above into the same GLO to perform adaptive processing on relevant clinical data. It is shown how each added capability improves the result of the image registration. In the end of the thesis there is a discussion which highlights the advantage of the contributions made as compared to previous methods in the scientific literature. / Dynamic Context Atlases for Image Denoising and Patient Safety

The current study investigates the effects of vascular disease on white matter health by comparing participants with atherosclerotic vascular disease (AVD) to healthy control participants (HC). The comparison between groups will help elucidate the differences between early-stage mild vascular disease and normal aging processes in terms of their effects on white matter health as measured by diffusion tensor imaging (DTI). Relationships between white matter health and depression, attention, and processing speed are studied by the application of a variety of DTI neuroimaging techniques, which will allow investigation of these relationships at the levels of global, lobe-wise, and subregional analysis. The specific subregion of interest in the depression study is Brodmann Area 25, which has shown significant relationships with depressive symptomatology in patients with treatment refractory depression, but has not been studied in the context of aging, vascular disease, or subthreshold depressive symptoms. Results indicate that there are significant differences between AVD and HC participants in global and regional FA measures. Within the AVD group, significant relationships of FA with depressive symptoms and attentional function have been observed in the current study. Several unexpected findings emerged, most important of which was the observation that there is a significant relationship between FA in Brodmann Area 25 and depressive symptoms in AVD participants which is specific to the right hemisphere. These findings have implications for the treatment of depressive symptoms in older adults and participants with vascular disease.