Melanoma is the deadliest form of skin cancer if left untreated. Incidence rates of melanoma have been increasing, especially among young adults, but survival rates are high if detected early. Unfortunately, the time and costs required for dermatologists to screen all patients for melanoma are prohibitively expensive. There is a need for an automated system to assess a patient's risk of melanoma using photographs of their skin lesions. Dermatologists could use the system to aid their diagnosis without the need for special or expensive equipment. One challenge in implementing such a system is locating the skin lesion in the digital image. Most existing skin lesion segmentation algorithms are designed for images taken using a special instrument called the dermatoscope. The presence of illumination variation in digital images such as shadows complicates the task of finding the lesion. The goal of this research is to develop a framework to automatically correct and segment the skin lesion from an input photograph. The first part of the research is to model illumination variation using a proposed multi-stage illumination modeling algorithm and then using that model to correct the original photograph. Second, a set of representative texture distributions are learned from the corrected photograph and a texture distinctiveness metric is calculated for each distribution. Finally, a texture-based segmentation algorithm classifies regions in the photograph as normal skin or lesion based on the occurrence of representative texture distributions. The resulting segmentation can be used as an input to separate feature extraction and melanoma classification algorithms. The proposed segmentation framework is tested by comparing lesion segmentation results and melanoma classification results to results using other state-of-the-art algorithms. The proposed framework has better segmentation accuracy compared to all other tested algorithms. The segmentation results produced by the tested algorithms are used to train an existing classification algorithm to identify lesions as melanoma or non-melanoma. Using the proposed framework produces the highest classification accuracy and is tied for the highest sensitivity and specificity.

Melanoma is the deadliest form of skin cancer if left untreated. Incidence rates of melanoma have been increasing, especially among young adults, but survival rates are high if detected early. Unfortunately, the time and costs required for dermatologists to screen all patients for melanoma are prohibitively expensive. There is a need for an automated system to assess a patient's risk of melanoma using photographs of their skin lesions. Dermatologists could use the system to aid their diagnosis without the need for special or expensive equipment. One challenge in implementing such a system is locating the skin lesion in the digital image. Most existing skin lesion segmentation algorithms are designed for images taken using a special instrument called the dermatoscope. The presence of illumination variation in digital images such as shadows complicates the task of finding the lesion. The goal of this research is to develop a framework to automatically correct and segment the skin lesion from an input photograph. The first part of the research is to model illumination variation using a proposed multi-stage illumination modeling algorithm and then using that model to correct the original photograph. Second, a set of representative texture distributions are learned from the corrected photograph and a texture distinctiveness metric is calculated for each distribution. Finally, a texture-based segmentation algorithm classifies regions in the photograph as normal skin or lesion based on the occurrence of representative texture distributions. The resulting segmentation can be used as an input to separate feature extraction and melanoma classification algorithms. The proposed segmentation framework is tested by comparing lesion segmentation results and melanoma classification results to results using other state-of-the-art algorithms. The proposed framework has better segmentation accuracy compared to all other tested algorithms. The segmentation results produced by the tested algorithms are used to train an existing classification algorithm to identify lesions as melanoma or non-melanoma. Using the proposed framework produces the highest classification accuracy and is tied for the highest sensitivity and specificity.

This diploma thesis is concerned with the numerical analysis of welded aluminim structures. In these structures, there are significant decreases in the mechanical properties at the area of the weld and in the heat affected zone as a result of welding. Within this thesis, simulations of quasi-statically loaded welded joints made from EN AW-6082 T6 alloy were performed to investigate the load capacity and ductility of these joints. Computations were performed using a programme system based on an explicit finite element method. To describe material anisotrophy, a nonlinear material model called the Weak texture model was chosen. Material properties of the weld and the heat affected zone were considered to be different from base material. The required material parameters were adopted from available literature, however, material tests and indetification procedure of these parameters were described. In comparison with the experimental data, the results of the numerical simulations showed a relatively good ability of models to capture load capacity of studied welded joints. Nevertheless, due to mesh sensitivity of models caused by localization of deformation, it was not possible to determine ductility of these joints.

La mammographie, une modalité 2D d'imagerie du sein par rayons X, a montré son efficacité pour réduire la mortalité par cancer du sein. Aujourd'hui, la tomosynthèse numérique du sein, une modalité 3D d'imagerie du sein par rayons X, prend une place de plus en plus importante dans la pratique clinique, et est reconnue de plus en plus souvent comme ayant le potentiel de remplacer la mammographie dans un proche avenir. Pour évaluer plusieurs aspects de la tomosynthèse, des études cliniques sont nécessaires. Mais les études cliniques sont coûteuses et présentent des risques supplémentaires pour les patientes dus à l'utilisation de radiations ionisantes. Les études cliniques virtuelles ont pour objectif d'offrir une approche alternative en utilisant des simulations numériques. Dans cette thèse, nous nous intéressons à plusieurs éléments intervenants dans une telle étude clinique virtuelle. Dans un premier temps, nous analysons l'état de l'art sur la caractérisation analytique des champs aléatoires 3D pour la simulation de la texture du sein par rayons X. Nous nous intéressons aussi à l'estimation de caractéristiques statistiques des images du sein par rayons X (densité, indice spectral). Puis nous développons un nouveau modèle de texture 3D du sein basé sur la géométrie stochastique, et qui permet de simuler des images 2D et 3D réalistes du sein. Nous considérons le problème de l'inférence d'un tel modèle à partir d'une base d'images cliniques 3D. Ensuite, nous développons un observateur mathématique basé sur la théorie textit{a contrario} de la perception visuelle, pour modéliser la détection des microcalcifications par des radiologues dans des images 2D et 3D du sein. Tous ces composants sont utilisés pour implémenter une étude clinique entièrement numérique. La pertinence des résultats obtenus montre l'utilité de ces études cliniques virtuelles et nous incite à en développer de plus élaborées dans le futur. / Full field digital mammography, a 2D x-ray breast imaging modality has been proved to reduce the breast cancer mortality. Today, digital breast tomosynthesis, a 3D x-ray breast imaging modality, is being integrated in clinical practice and is believed to replace standard mammography in the near future. To assess the clinical performance of various aspects of tomosynthesis, clinical trials are needed. Clinical trials are burdensome, expensive and may impose increased risk to the patient due to additional radiation exposure. Virtual Clinical Trials aim to offer a more efficient alternative by using computational components. Today, active research is ongoing to develop computational components dedicated to 2D and 3D breast imaging, especially to 3D tomosynthesis. This thesis aims to advance several aspects in the development of Virtual Clinical Trials. First, we focused on analytical characterization of state-of-the-art 3D random field breast texture models. The estimation of statistical characteristics (breast density, spectral index) from clinical x-ray breast images was also studied. Next, we proposed a mathematically traceable 3D breast texture model based on stochastic geometry, that allows to simulate realistic 2D and 3D images. The statistical inference of the texture model parameters from a database of clinical 3D breast images was also tackled. We then developed a mathematical observer based on the textit{a contrario} theory, that allows to model the microcalcification detection process by radiologists in 2D and 3D breast images. Finally, these two proposed components were applied to implement a virtual clinical trial experiment, demonstrating their potential in the conduct of more advanced virtual clinical studies in the future.