Capillaries in tumours are often severely buckled (in a plane perpendicular to the axis) and / or chaotic in their direction. We develop a model of these phenomena using nonlinear solid mechanics. Our model focusses on the immediate surrounding of a capillary. The vessel and surrounding tissue are modelled as concentric annulii. The growth is dependent on the concentration of a nutrient (oxygen) diffusing from the vessel into the tumour interstitium. The stress is modelled using a multiplicative decomposition of the deformation gradient F=F_e F_g. The stress is determined by substituting the elastic deformation gradient F_e (which gives the deformation gradient from the hypothetical configuration to the current configuration) into a hyperelastic constitutive model as per classical solid mechanics. We use a Blatz-Ko model, parameterised using uniaxial compression experiments. The entire system is in quasi-static equilibrium, with the divergence of the stress tensor equal to zero. We determine the onset of buckling using a linear stability analysis. We then investigate the postbuckling behaviour by introducing higher order perturbations in the deformation and growth before using the Fredholm Alternative to obtain the magnitude of the buckle. Our results demonstrate that the growth-induced stresses are sufficient for the capillary to buckle in the absence of external loading and / or constraints. Planar buckling usually occurs after 2-5 times the cellular proliferation timescale. Buckles with axial variation almost always go unstable after planar buckles. Buckles of fine wavelength are initially preferred by the system, but over time buckles of large wavelength become energetically more favourable. The tumoural hoop stress T_{ThetaTheta} is the most invariant (Eulerian) variable at the time of buckling: it is typically of the order of the tumoural Young's Modulus when this occurs.

This thesis is devoted to the mathematical analysis of models describing the energy of defects in crystalline solids via variational methods. The first part of this work studies a discrete model describing the energy of a point defect in a one dimensional chain of atoms. We derive an expansion of the ground state energy using Gamma-convergence, following previous work on similar models [BDMG99,BC07,SSZ11]. The main novelty here is an explicit characterisation of the first order limit as the solution of a variational problem in an infinite lattice. Analysing this variational problem, we prove a regularity result for the perturbation caused by the defect, and demonstrate the order of the next term in the expansion. The second main topic is a discrete model describing screw dislocations in body centred cubic crystals. We formulate an anti plane lattice model which describes the energy difference between deformations and, using the framework defined in [AO05], provide a kinematic description of the Burgers vector, which is a key geometric quantity used to describe dislocations. Apart from the anti plane restriction, this model is invariant under all the natural symmetries of the lattice and in particular allows for the creation and annihilation of dislocations. The energy difference formulation enables us to provide a clear definition of what it means to be a stable deformation. The main results of the analysis of this model are then first, a proof that deformations with unit net Burgers vector exist as globally stable states in an infinite body, and second, that deformations containing multiple screw dislocations exist as locally stable states in both infinite bodies and finite convex bodies. To prove the former result, we establish coercivity with respect to the elastic strain, and exploit a concentration compactness principle. In the latter case, we use a form of the inverse function theorem, proving careful estimates on the residual and stability of an ansatz which combines continuum linear elasticity theory with an atomistic core correction.

Cerebrospinal fluid (CSF) fills a system of cavities at the centre of the brain, known as ventricles, and the subarachnoid space surrounding the brain and the spinal cord. In addition, CSF is in free communication with the interstitial fluid of the brain tissue. Disturbances in CSF dynamics can lead to diseases that cause severe brain damage or even death. So-called infusion tests are frequently performed in the diagnosis of such diseases. In this type of test, changes in average CSF pressure are related to changes in CSF volume through infusion of known volumes of additional fluid. Traditionally, infusion tests are analysed with single compartment models, which treat all CSF as part of one compartment and balance fluid inflow, outflow and storage through a single ordinary differential equation. Poroelastic models of the brain, on the other hand, have been used to simulate spatial changes with disease, particularly of the ventricle size, on larger time scales of days, weeks or months. Wirth and Sobey (2008) developed a two-fluid poroelastic model of the brain in which CSF pressure pulsations are linked to arterial blood pressure pulsations. In this thesis, this model is developed further and simulation results are compared to clinical data. At first, the functional form of the compliance, which governs the storage of CSF in single compartment models, is examined by comparison of two different compliance models with clinical data. The derivations of a single-fluid and a two-fluid poroelastic model of the brain in spherical symmetry are laid out in detail and some of the parameters are related to the compliance functions considered earlier. The finite element implementation of the two-fluid model is described and finally simulation results of the average CSF pressure response and the pressure pulsations are compared to clinical data.

Le cancer pulmonaire est la principale cause de décès parmi tous les cancers au Canada. Le pronostic est généralement faible, de l'ordre de 15% de taux de survie après 5 ans. Les déplacements internes des structures anatomiques apportent une incertitude sur la précision des traitements en radio-oncologie, ce qui diminue leur efficacité. Dans cette optique, certaines techniques comme la radio-chirurgie et la radiothérapie par modulation de l'intensité (IMRT) visent à améliorer les résultats cliniques en ciblant davantage la tumeur. Ceci permet d'augmenter la dose reçue par les tissus cancéreux et de réduire celle administrée aux tissus sains avoisinants. Ce projet vise à mieux évaluer la dose réelle reçue pendant un traitement considérant une anatomie en mouvement. Pour ce faire, des plans de CyberKnife et d'IMRT sont recalculés en utilisant un algorithme Monte Carlo 4D de transport de particules qui permet d'effectuer de l'accumulation de dose dans une géométrie déformable. Un environnement de simulation a été développé afin de modéliser ces deux modalités pour comparer les distributions de doses standard et 4D. Les déformations dans le patient sont obtenues en utilisant un algorithme de recalage déformable d'image (DIR) entre les différentes phases respiratoire générées par le scan CT 4D. Ceci permet de conserver une correspondance de voxels à voxels entre la géométrie de référence et celles déformées. La DIR est calculée en utilisant la suite ANTs («Advanced Normalization Tools») et est basée sur des difféomorphismes. Une version modifiée de DOSXYZnrc de la suite EGSnrc, defDOSXYZnrc, est utilisée pour le transport de particule en 4D. Les résultats sont comparés à une planification standard afin de valider le modèle actuel qui constitue une approximation par rapport à une vraie accumulation de dose en 4D. / Pulmonary cancer is the main cause of death amongst all cancers in Canada with a prognosis of about 15% survival rate in 5 years. The efficiency of radiotherapy treatments is lower when high displacements of the tumors are observed, mostly caused by intrafraction respiratory motion. Advanced techniques such as radiosurgery and intensity-modulated radiotherapy treatments (IMRT) are expected to provide better clinical results by delivering higher radiation doses to the tumor while sparing the surrounding healthy lung tissues. The goal of this project is to perform 4D Monte Carlo dose recalculations to assess the dosimetric impact of moving tumors in CyberKnife and IMRT treatments using dose accumulation in deforming anatomies. Scripts developed in-house were used to model both situations and to compare the Monte Carlo dose distributions with those obtained with standard clinical plans. Displacement vectors fields are obtained from a 4D CT data set and a deformable image registration (DIR) algorithm which allows a voxel-to-voxel correspondence between each respiratory phase. The DIR is computed by the Advanced Normalization Tools (ANTs) software and is mostly based on diffeormophisms. A modified version of DOSXYZnrc from EGSnrc software, defDOSXYZnrc, is used to transport radiation through non-linear geometries. These results are then compared to a typical 3D plan to determine whether or not the current planification is a good approximation of the true 4D dose calculation.

Le besoin tant en haute qualité d'imagerie qu'en structures légères est l'un des principaux moteurs pour la conception des télescopes spatiaux. Un contrôle efficace du front d'onde va donc devenir indispensable dans les futurs grands observatoires spatiaux, assurant une bonne performance optique tout en relâchant les contraintes sur la stabilité globale du système. L'optique active consiste à contrôler la déformation des miroirs, cette technique peut être utilisée afin de compenser la déformation des grands miroirs primaires, afin de permettre l'utilisation d'instrument reconfigurable ou afin de fabriquer des miroirs asphériques avec le polissage sous contraintes. Dans ce manuscrit, la conception de miroirs actifs dédiés à l'instrumentation spatiale est présentée. Premièrement, un système compensant la déformation d'un grand miroir allégé dans l'espace est conçu et ses performances sont démontrées expérimentalement. Avec 24 actionneurs, le miroir MADRAS (Miroir Actif Déformable et Régulé pour Applications Spatiales) effectuera une correction efficace du front d'onde dans un relais de pupille du télescope. Deuxièmement, un harnais de déformation pour le polissage sous contraintes des segments du télescope géant européen de 39 m (E-ELT) est présenté. La performance du procédé est prédite et optimisée avec des analyses éléments finis et la production en masse des segments est considérée. Troisièmement, deux concepts originaux de miroirs déformables avec un nombre minimal d'actionneurs ont été développés. VOALA (Variable Off-Axis parabola) est un système à trois actionneurs et COMSA (Correcting Optimized Mirror with a Single Actuator) est un système à un actionneur. / The need for both high quality images and light structures is one of the main driver in the conception of space telescopes. An efficient wave-front control will then become mandatory in the future large observatories, ensuring the optical performance while relaxing the specifications on the global system stability. Consisting in controlling the mirror deformation, active optics techniques can be used to compensate for primary mirror deformation, to allow the use of reconfigurable instruments or to manufacture aspherical mirror with stress polishing. In this manuscript, the conception of active mirrors dedicated to space instrumentation is presented. Firstly, a system compensating for large lightweight mirror deformation in space, is designed and its performance are experimentally demonstrated. With 24 actuators, the MADRAS mirror (Mirror Actively Deformed and Regulated for Applications in Space) will perform an efficient wave-front correction in the telescope's pupil relay. Secondly, a warping harness for the stress polishing of the 39 m European Extremely Large Telescope segments is presented. The performance of the process is predicted and optimized with Finite Element Analysis and the segments mass production is considered. Thirdly, two original concepts of deformable mirrors with a minimum number of actuators have been developed. The Variable Off-Axis parabola (VOALA) is a 3-actuators system and the Correcting Optimized Mirror with a Single Actuator (COMSA) is a 1-actuator system.

Cette thèse traite la problématique liée à la modélisation du visage dans le but de l’analyse faciale.Dans la première partie de cette thèse, nous avons proposé le Modèle Actif d’Apparence Multi-Objet. La spécificité du modèle proposé est que les différentes parties du visage sont traités comme des objets distincts et les mouvements oculaires (du regard et clignotement) sont extrinsèquement paramétrées.La deuxième partie de la thèse porte sur l'utilisation de la modélisation de visage dans le contexte de la reconnaissance des émotions.Premièrement, nous avons proposé un système de reconnaissance des expressions faciales sous la forme d’Action Units. Notre contribution porte principalement sur l'extraction des descripteurs de visage. Pour cela nous avons utilisé les modèles AAM locaux.Le second système concerne la reconnaissance multimodale des quatre dimensions affectives :. Nous avons proposé un système qui fusionne des caractéristiques audio, contextuelles et visuelles pour donner en sortie les quatre dimensions émotionnelles. Nous contribuons à ce système en trouvant une localisation précise des traits du visage. En conséquence, nous proposons l’AAM Multi-Modèle. Ce modèle combine un modèle global extrinsèque du visage et un modèle local de la bouche. / The work in this thesis deals with the problematic of face modeling for the purpose of facial analysis.In the first part of this thesis, we proposed the Multi-Object Facial Actions Active Appearance Model (AAM). The specificity of the proposed model is that different parts of the face are treated as separate objects and eye movements (gaze and blink) are extrinsically parameterized. This increases the generalization capabilities of classical AAM.The second part of the thesis concerns the use of face modeling in the context of expression and emotion recognition. First we have proposed a system for the recognition of facial expressions in the form of Action Units (AU). Our contribution concerned mainly the extraction of AAM features of which we have opted for the use of local models.The second system concerns multi-modal recognition of four continuously valued affective dimensions. We have proposed a system that fuses audio, context and visual features and gives as output the four emotional dimensions. We contribute to the system by finding the precise localization of the facial features. Accordingly, we propose the Multi-Local AAM. This model combines extrinsically a global model of the face and a local one of the mouth through the computation of projection errors on the same global AAM.

Meshless methods provide a highly continuous approximation field, convenient for thin structures like shells. Nevertheless, the lack of Kronecker Delta property makes the formulation of essential boundary conditions not straightforward, as the trial and test fields cannot be tailored to boundary values. Similar problem arise when different approximation regions must be joined, in a multi-region problem, such as kinks, folds or joints. This work presents three approaches to impose both kinematic conditions: the well known Lagrange Multiplier method, used since the beginning of the Element Free Galerkin method; a pure penalty approach; and the recently rediscovered alternative of Nitsche\'s Method. We use the EFG discretization technique for thick Reissner-Mindlin shells and adapt the weak form as to separate displacement and rotational degrees of freedom and obtain suitable and separate stabilization parameters. This approach enables the modeling of discontinuous shells and local refinement on multi-region problems. / Métodos sem malha geram campos de aproximação com alta continuidade, convenientes para estruturas finas como cascas. No entanto, a ausência da propriedade de Delta de Kronecker dificulta a formulação de condições essenciais de contorno, já que os campos de aproximação e teste não podem ser moldados aos valores de contorno. Um problema similar aparece quando diferentes regiões de aproximação precisam ser juntadas em um problema multi-regiões como dobras, vincos ou junções. Este trabalho apresenta três métodos de imposição ambas condições cinemáticas: o já conhecido método dos multiplicadores de Lagrange, usado desde o começo do método de Galekin sem elementos (EFG); uma abordagem de penalidade pura; e o recentemente redescoberto método de Nitsche. Nós usamos a técnica de discretização com EFG para cascas espessas de Reissner-Mindlin e adaptamos a forma fraca de forma a separar graus de liberdade de deslocamento e rotação e obter coeficientes de estabilização diferentes e apropriados. Essa abordagem permite a modelagem de cascas discontínuas e o refinamento local em problemas multi-regiões.

Les changements anatomiques des patients au cours du traitement de radiothérapie peuvent engendrer des conséquences dosimétriques significatives sur les volumes cibles (VC) ou les organes à risques (OARs). Le processus de radiothérapie adaptative peut compenser ces variations, cependant son déploiement en clinique est ralentit par une charge de travail supplémentaire considérable pour les équipes médicales et aucun logiciel n'est disponible pour une utilisation en clinique. Nous avons développé un outil d'alerte dosimétrique in vivo simple permettant d'identifier rapidement les situations où une adaptation de traitement est requise pour un patient. L'évaluation dosimétrique des traitements délivrés a été réalisée sur l'imagerie embarquée 3D (CBCT) dont la précision des calculs de dose a dû être évaluée. L'outil d'alerte permet de s'affranchir d'une nouvelle délinéation de volumes d'intérêt et est basé sur des critères objectifs et quantifiables constitués par le dépassement des limites dosimétriques définies pour chacun des volumes considérés. La précision et la détectabilité de l'outil ont été validées puis il a été appliqué dans une étude rétrospective de 10 patients ORL afin de surveiller l'administration du traitement et d'identifier les patients pour lesquels une adaptation du traitement aurait pu être envisagée. Dans son implémentation clinique, le processus de radiothérapie adaptative requiert des algorithmes de recalage déformable capable de suivre les déformations locales d'un patient se produisant au cours du traitement, seulement leur utilisation n'est pas encore validée. Nous avons procédé à l'évaluation de la précision d'un algorithme de recalage déformable, de type Block Matching présentant l'avantage d'être adapté à l'imagerie multimodale CT/CBCT, en comparaison par rapport à un algorithme de recalage rigide. Une étude a été menée pour 10 patients ORL en se basant sur la comparaison de contours de volumes d'intérêt pour 76 CBCT. Les paramètres de similarité utilisés consistaient en l'Indice de Similarité Dice, la distance de Hausdorff robuste (en mm) et la différence de volume absolu (en cm3) / Changing anatomy during radiotherapy can lead to significant dosimetric consequences for organs at risk (OARs) and/or target volumes. Adaptive radiotherapy can compensate for these variations however its deployment for clinical work is hampered by the increased workload for the medical staff and there is still no commercialized software available for clinical use. We developed a simple in vivo dosimetric alert tool allowing rapid identification of patients who might benefit from an adaptive radiotherapy. Dosimetric evaluation of delivered treatment has been conducted onto 3D on board imaging (CBCT) whose dose calculation accuracy has been evaluated. The tool does not require a new volume of interest delineation. Tool alert is based on objectives and quantifiable criteria defined by the exceeding volumes of interest dose thresholds. Tool precision and detectability have been validated and applied in a retrospective study on 10 head and neck patients. The tool allows detecting patients where an adaptive treatment could have been considered. In its clinical implementation, adaptive radiotherapy process requires deformable matching algorithms to follow patient local's deformations occurring during treatment. Nevertheless, their use has not been validated. We conducted an evaluation of the Block Matching deformable algorithm, suitable for multimodality imaging (CT/CBCT), in comparison to rigid algorithm. A study has been conducted for 10 head and neck patients based on volume of interest contours comparison for 76 CBCT. Similarity parameters used consisted on Dice Similarity Index, Robust Hausdorff Distance (in mm) and the absolute volume difference (in cc)

O remifentanil é um opióide potente comparável ao fentanil, mas, possui rápida eliminação plasmática e por isso necessariamente deve ser adiministrado por infusão contínua. A administração transdérmica pode liberar fármacos de forma prolongada, contínua e controlada. Características que facilitariam a administração do remifentanil, contornariam sua limitação farmacocinética e ampliariam os casos em que este fármacos seria útil. Lipossomas deformáveis são formados por fosfolipídios e por tensoativo que normalmente é um surfactante de cadeia simples que aumenta a flexibilidade da bicamada lipídica da membrana e torna os lipossomas deformáveis. Os lipossomas deformáveis são capazes de disponibilizar, de forma expressiva, várias substâncias transdermicamente incluindo macromoléculas e podem ser um eficiente sistema de liberação controlada de remifentanil. O presente estudo desenvolve diferentes dispersões de lipossomas deformáveis de remifentanil e caracteriza essas dispersões quanto a tamanho de diâmetro dos lipossomas, eficiência de encapsulação dos lipossomas, elasticidade da membrana, permeação e retenção cutânea in vitro. As dispersões de lipossomas deformáveis preparadas foram compostas por remifentanil, fosfatidilcolina de soja, tensoativo tween 80 ou tween 20 e etanol a 10% ou a 20% e foram preparadas por dois métodos: sonicação e hidratação de filme e sonicação. O tamanho e a distribuição de tamanho foram obtidos por espalhamento dinâmico de luz, a eficiência de encapsulação por diálise, a elasticidade por filtração, seguido de determinação do tamanho. Células de difusão de Franz e pele de orelha de porco foram usadas para estudar a permeação e a retenção cutânea in vitro. Os lipossomas deformáveis preparados apresentaram pequeno diâmetro (40 a 71 nm) quando comparado a outros na literatura e seus tamanhos foram dependentes da formulação. Todos os lipossomas deformáveis preparados apresentaram boa elasticidade e permeação dependente da eficiência de encapsulação e do tamanho de diâmetro. A solução hidroetanólica com remifentanil e a mistura física dos excipientes com remifentanil não apresentaram permeação e retenção. A eficiência de encapsulação dos lipossomas deformáveis é coerente com outros resultados na literatura e foi dependente da formulação e do método de preparação. A avaliação da permeação de dispersões de lipossomas deformáveis com remifentanil não encapsulado foi feita para se obter informações sobre os mecanismos de ação pelos quais estes lipossomas agem. Verificou-se que o transporte de substâncias encapsuladas nos lipossomas foi o mecanismo de maior importância / Remifentanil is a potent opioid comparable to fentanyl. Remifentanil has rapid systemic elimination and, therefore, it necessarily has to be administered by continuous parenteral infusion. Transdermal delivery can provide extended, continuous and controlled delivery of drug. Characteristics that could make the remifentanil administration easier, could resolve its phamacocinetcs limitation and could enlarge the cases that this drug would be useful. Deformable liposomes consist of phospholipid and surfactant that is often a single chain surfactant that increases the membrane lipid bilayer flexibility and makes the liposome deformable. Deformable liposomes are able to expressively improve skin delivery of many drugs, including macromolecules and could be an efficient controlled delivery system of remifentanil. The aim of this study has been to develop different deformable liposomes dispersions of remifentanil and to characterize those dispersions by liposome diameter size, liposome entrapment efficiency, membrane elasticity and in vitro skin permeation and deposition. The deformable liposomes consisted of remifentanil, soybean phosphatidylcholine, surfactant tween 80 or tween 20, and ethanol at 10% or 20%. They have been prepared by two methods: sonication or conventional mechanical dispersion and sonication. The size was determined by light scattering, the entrapment efficiency by dialysis, the membrane elasticity by filtration and size determination. Franz diffusion cells and ear porcine skin wore used to evaluate the in vitro skin permeation and the skin deposition. The deformable liposomes showed small size (40 a 71 nm) compared with others from literature and their sizes wore dependent of the formulation. All deformable liposomes showed good membrane elasticity and showed permeation that was dependent of the formulation and diameter size. The remifentanil hidroetanolic solution and the excipients physical mixture with remifentanil did not have any permeation and retention. The deformable liposomes entrapment efficiency has been coherent with others literature results and also was dependent of the formulation and preparation method. The permeation of deformable liposomes dispersions wore evaluated to get information about the deformable liposome action mechanisms. The transport of encapsulated substances on liposome was the action mechanism of major importance.

Cette thèse explore l’utilisation des modèles de contours déformables pour la segmentation basée sur la forme des images médicales. Nous apportons des contributions sur deux fronts: dans le problème de l’apprentissage statistique, où le modèle est formé à partir d’un ensemble d’images annotées, et le problème de l’inférence, dont le but est de segmenter une image étant donnée un modèle. Nous démontrons le mérite de nos techniques sur une grande base d’images à rayons X, où nous obtenons des améliorations systématiques et des accélérations par rapport à la méthode de l’état de l’art. Concernant l’apprentissage, nous formulons la formation de la fonction de score des modèles de contours déformables en un problème de prédiction structurée à grande marge et construisons une fonction d’apprentissage qui vise à donner le plus haut score à la configuration vérité-terrain. Nous intégrons une fonction de perte adaptée à la prédiction structurée pour les modèles de contours déformables. En particulier, nous considérons l’apprentissage avec la mesure de performance consistant en la distance moyenne entre contours, comme une fonction de perte. L’utilisation de cette fonction de perte au cours de l’apprentissage revient à classer chaque contour candidat selon sa distance moyenne du contour vérité-terrain. Notre apprentissage des modèles de contours déformables en utilisant la prédiction structurée avec la fonction zéro-un de perte surpasse la méthode [Seghers et al. 2007] de référence sur la base d’images médicales considérée [Shiraishi et al. 2000, van Ginneken et al. 2006]. Nous démontrons que l’apprentissage avec la fonction de perte de distance moyenne entre contours améliore encore plus les résultats produits avec l’apprentissage utilisant la fonction zéro-un de perte et ce d’une quantité statistiquement significative.Concernant l’inférence, nous proposons des solveurs efficaces et adaptés aux problèmes combinatoires à variables spatiales discrétisées. Nos contributions sont triples: d’abord, nous considérons le problème d’inférence pour des modèles graphiques qui contiennent des boucles, ne faisant aucune hypothèse sur la topologie du graphe sous-jacent. Nous utilisons un algorithme de décomposition-coordination efficace pour résoudre le problème d’optimisation résultant: nous décomposons le graphe du modèle en un ensemble de sous-graphes en forme de chaines ouvertes. Nous employons la Méthode de direction alternée des multiplicateurs (ADMM) pour réparer les incohérences des solutions individuelles. Même si ADMM est une méthode d’inférence approximative, nous montrons empiriquement que notre implémentation fournit une solution exacte pour les exemples considérés. Deuxièmement, nous accélérons l’optimisation des modèles graphiques en forme de chaîne en utilisant l’algorithme de recherche hiérarchique A* [Felzenszwalb & Mcallester 2007] couplé avec les techniques d’élagage développés dans [Kokkinos 2011a]. Nous réalisons une accélération de 10 fois en moyenne par rapport à l’état de l’art qui est basé sur la programmation dynamique (DP) couplé avec les transformées de distances généralisées [Felzenszwalb & Huttenlocher 2004]. Troisièmement, nous intégrons A* dans le schéma d’ADMM pour garantir une optimisation efficace des sous-problèmes en forme de chaine. En outre, l’algorithme résultant est adapté pour résoudre les problèmes d’inférence augmentée par une fonction de perte qui se pose lors de l’apprentissage de prédiction des structure, et est donc utilisé lors de l’apprentissage et de l’inférence. [...] / This thesis explores the use of discriminatively trained deformable contour models (DCMs) for shape-based segmentation in medical images. We make contributions in two fronts: in the learning problem, where the model is trained from a set of annotated images, and in the inference problem, whose aim is to segment an image given a model. We demonstrate the merit of our techniques in a large X-Ray image segmentation benchmark, where we obtain systematic improvements in accuracy and speedups over the current state-of-the-art. For learning, we formulate training the DCM scoring function as large-margin structured prediction and construct a training objective that aims at giving the highest score to the ground-truth contour configuration. We incorporate a loss function adapted to DCM-based structured prediction. In particular, we consider training with the Mean Contour Distance (MCD) performance measure. Using this loss function during training amounts to scoring each candidate contour according to its Mean Contour Distance to the ground truth configuration. Training DCMs using structured prediction with the standard zero-one loss already outperforms the current state-of-the-art method [Seghers et al. 2007] on the considered medical benchmark [Shiraishi et al. 2000, van Ginneken et al. 2006]. We demonstrate that training with the MCD structured loss further improves over the generic zero-one loss results by a statistically significant amount. For inference, we propose efficient solvers adapted to combinatorial problems with discretized spatial variables. Our contributions are three-fold:first, we consider inference for loopy graphical models, making no assumption about the underlying graph topology. We use an efficient decomposition-coordination algorithm to solve the resulting optimization problem: we decompose the model’s graph into a set of open, chain-structured graphs. We employ the Alternating Direction Method of Multipliers (ADMM) to fix the potential inconsistencies of the individual solutions. Even-though ADMMis an approximate inference scheme, we show empirically that our implementation delivers the exact solution for the considered examples. Second,we accelerate optimization of chain-structured graphical models by using the Hierarchical A∗ search algorithm of [Felzenszwalb & Mcallester 2007] couple dwith the pruning techniques developed in [Kokkinos 2011a]. We achieve a one order of magnitude speedup in average over the state-of-the-art technique based on Dynamic Programming (DP) coupled with Generalized DistanceTransforms (GDTs) [Felzenszwalb & Huttenlocher 2004]. Third, we incorporate the Hierarchical A∗ algorithm in the ADMM scheme to guarantee an efficient optimization of the underlying chain structured subproblems. The resulting algorithm is naturally adapted to solve the loss-augmented inference problem in structured prediction learning, and hence is used during training and inference. In Appendix A, we consider the case of 3D data and we develop an efficientmethod to find the mode of a 3D kernel density distribution. Our algorithm has guaranteed convergence to the global optimum, and scales logarithmically in the volume size by virtue of recursively subdividing the search space. We use this method to rapidly initialize 3D brain tumor segmentation where we demonstrate substantial acceleration with respect to a standard mean-shift implementation. In Appendix B, we describe in more details our extension of the Hierarchical A∗ search algorithm of [Felzenszwalb & Mcallester 2007] to inference on chain-structured graphs.