L’essor récent de l’imagerie hybride combinant la Tomographie par Emission de Positons (TEP) à l’Imagerie par Résonance Magnétique (IRM) est une opportunité permettant d’exploiter des images d’un même territoire anatomo-pathologique obtenues simultanément et apportant des informations complémentaires. Cela représente aussi un véritable défi en raison de la différence de nature et de résolution spatiale des données acquises. Cette nouvelle technologie offre notamment des perspectives attrayantes en oncologie, et plus particulièrement en neuro-oncologie grâce au contraste qu’offre l’image IRM entre les tissus mous. Dans ce contexte et dans le cadre du projet PIM (Physique et Ingénierie pour la Médecine) de l’Université Paris-Saclay, l’objectif de cette thèse a été de développer un processus de segmentation multimodale adapté aux images TEP et IRM, comprenant une méthode de détection des volumes tumoraux en TEP et IRM, et une technique de segmentation précise du volume tumoral IRM. Ce processus doit être suffisamment générique pour s’appliquer à diverses pathologies cérébrales, différentes par leur nature même et par l’application clinique considérée. La première partie de la thèse aborde la détection de tumeurs par une approche hiérarchique. Plus précisément, la méthode de détection, réalisée sur les images IRM ou TEP, repose sur la création d’un nouveau critère de contexte spatial permettant de sélectionner les lésions potentielles par filtrage d’une représentation de l’image par max-tree. La deuxième partie de la thèse concerne la segmentation du volume tumoral sur les images IRM par une méthode variationnelle par ensembles de niveaux. La méthode de segmentation développée repose sur la minimisation d’une énergie globalement convexe associée à une partition d’une image RM en régions homogènes guidée par des informations de la TEP. Enfin, une dernière partie étend les méthodes proposées précédemment à l’imagerie multimodale IRM, notamment dans le cadre de suivi longitudinal. Les méthodes développées ont été testées sur plusieurs bases de données, chacune correspondant à une pathologie cérébrale et un radiotraceur TEP distincts. Les données TEP-IRM disponibles comprennent, d’une part, des examens de méningiomes et de gliomes acquis sur des machines séparées, et d’autre part, des examens réalisés sur le scanner hybride du Service Hospitalier Frédéric Joliot d’Orsay dans le cadre de recherches de tumeurs cérébrales. La méthode de détection développée a aussi été adaptée à l’imagerie multimodale IRM pour la recherche de lésions de sclérose en plaques ou le suivi longitudinal. Les résultats obtenus montrent que la méthode développée, reposant sur un socle générique, mais étant aussi modulable à travers le choix de paramètres, peut s’adapter à diverses applications cliniques. Par exemple, la qualité de la segmentation des images issues de la machine combinée a été mesurée par le coefficient de Dice, la distance de Hausdorff (DH) et la distance moyenne (DM), en prenant comme référence une segmentation manuelle de la tumeur validée par un expert médical. Les résultats expérimentaux sur ces données montrent que la méthode détecte les lésions visibles à la fois sur les images TEP et IRM, et que la segmentation contoure correctement la lésion (Dice, DH et DM valant respectivement 0, 85 ± 0, 09, 7, 28 ± 5, 42 mm et 0, 72 ± 0, 36mm). / The recent development of hybrid imaging combining Positron Emission Tomography (PET) and Magnetic Resonance Imaging (MRI) is an opportunity to exploit images of a same structure obtained simultaneously and providing complementary information. This also represents a real challenge due to the difference of nature and voxel size of the images. This new technology offers attractive prospects in oncology, and more precisely in neuro-oncology thanks to the contrast between the soft tissues provided by the MRI images. In this context, and as part of the PIM (Physics in Medicine) project of Paris-Saclay University, the goal of this thesis was to develop a multimodal segmentation pipeline adapted to PET and MRI images, including a tumor detection method in PET and MRI, and a segmentation method of the tumor in MRI. This process must be generic to be applied to multiple brain pathologies, of different nature, and for different clinical application. The first part of the thesis focuses on tumor detection using a hierarchical approach. More precisely, the detection method uses a new spatial context criterion applied on a max-tree representation of the MRI and PET images to select potential lesions. The second part presents a MRI tumor segmentation method using a variational approach. This method minimizes a globally convex energy function guided by PET information. Finally, the third part proposes an extension of the detection and segmentation methods developed previously to MRI multimodal segmentation, and also to longitudinal follow-up. The detection and segmentation methods were tested on images from several data bases, each of them standing for a specific brain pathology and PET radiotracer. The dataset used for PET-MRI detection and segmentation is composed of PET and MRI images of gliomas and meningiomas acquired from different systems, and images of brain lesions acquired on the hybrid PET-MRI system of Frédéric Joliot Hospital at Orsay. The detection method was also adapted to multimodal MRI imaging to detect multiple sclerosis lesions and follow-up studies. The results show that the proposed method, characterized by a generic approach using flexible parameters, can be adapted to multiple clinical applications. For example, the quality of the segmentation of images from the hybrid PET-MR system was assessed using the Dice coefficient, the Hausdorff distance (HD) and the average distance (AD) to a manual segmentation of the tumor validated by a medical expert. Experimental results on these datasets show that lesions visible on both PET and MR images are detected, and that the segmentation delineates precisely the tumor contours (Dice, HD and MD values of 0.85 ± 0.09, 7.28 ± 5.42 mm and 0.72 ± 0.36mm respectively).

One of the most important factors on the success of evolutionary algorithms (EAs) about trees is the representation of them. The representation should exhibit efficiency, locality and heritability to enable effective evolutionary computing. Neville proposed three different methods for encoding labeled trees. The first one is similar with Pr&uuml / fer&#039 / s encoding. In 2001, it is reported that, the use of Pr&uuml / fer numbers is a poor representation of spanning trees for evolutionary search, since it has low locality for random trees. In the thesis Neville&#039 / s other two encodings, namely Neville branch numbers and Neville leaf numbers, are studied. For their performance in EA their properties and algorithms for encoding and decoding them are also examined. Optimal algorithms with time and space complexities of O(n) , where n is the number of nodes, for encoding and decoding Neville branch numbers are given. The localities of Neville&#039 / s encodings are investigated. It is shown that, although the localities of Neville branch and leaf numbers are perfect for star type trees, they are low for random trees. Neville branch and Neville leaf numbers are compared with other codings in EAs and SA for four problems: &#039 / onemax tree problem&#039 / , &#039 / degree-constrained minimum spanning tree problem&#039 / , &#039 / all spanning trees problem&#039 / and &#039 / all degree constrained spanning trees problem&#039 / . It is shown that, neither Neville nor Pr&uuml / fer encodings are suitable for EAs. These encodings are suitable for only tree enumeration and degree computation. Algorithms which are timewise and spacewise optimal for &#039 / all spanning trees problem&#039 / (ASTP) for complete graphs, are given by using Neville branch encoding. Computed time and space complexities for solving ASTP of complete graphs are O(nn-2) and O(n) if trees are only enumerated and O(nn-1) and O(n) if all spanning trees are printed , respectively, where n is the number of nodes. Similarly, &#039 / all degree constrained spanning trees problem&#039 / of a complete graph is solvable in O(nn-1) time and O(n) space.

Orientador: Roberto de Alencar Lotufo / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-14T21:44:46Z (GMT). No. of bitstreams: 1 Silva_AlexandreGoncalves_D.pdf: 8468020 bytes, checksum: 8e00e5b8cd107e6379773db874c73089 (MD5) Previous issue date: 2009 / Resumo: Uma imagem em níveis de cinza pode ser interpretada como uma superfície topográfica e representada por uma árvore de componentes, baseada na relação de inclusão de regiões conexas, obtida a partir da decomposição por limiares. Medidas sobre platôs, vales ou montanhas deste relevo são úteis na caracterização de objetos de interesse em sistemas de visão computacional. Este trabalho apresenta métodos de filtragem, segmentação e reconhecimento de padrões derivados da exploração de aspectos semânticos oferecidos por essa estrutura hierárquica, construída de maneira concisa e em tempo quase-linear, mesmo com a introdução de uma série de novos atributos geométricos, topológicos e estatísticos. Havendo menos elementos a processar em relação à quantidade de pixels e, sendo possível a alteração da organização dos mesmos por meio de podas e enxertos, essa representação possibilita a implementação de algoritmos rápidos para operadores conexos antiextensivos. Um importante resultado da árvore estendida de atributos é a formulação genérica e determinação eficiente de novos valores de extinção, utilizados como modelo simplificado de seleção de extremos ou marcadores relevantes para reconstrução morfológica ou segmentação por regiões de influência. Propõe-se também um algoritmo unificado para pesquisa de formas conforme a análise adotada para verificação aproximada da disposição espacial de pixels de cada componente na árvore. / Abstract: A gray-level image can be interpreted as a topographical surface and represented by a component tree, based on the inclusion relation of connected regions, obtained by threshold decomposition. Measures on plateaus, valleys or mountains of this relief are useful in the characterization of objects of interest in computer vision systems. This work presents filtering, segmentation and pattern recognition methods from the exploration of semantic aspects provide by this hierarchical structure, whose can be constructed in a concise way and in quasi-linear time, even with the addition of a set of new geometric, topological and statistical attributes. How there is less elements to process in relation to the amount of pixels, and being able to change the organization of these through pruning and grafting, this representation allows the implementation of fast algorithms for connected anti-extensive operators. An important result of the extended attribute tree is the generic formulation and efficient determination of new extinction values, used as simplified model of selecting relevant extremes or markers for morphological reconstruction or segmentation by influence regions. A unified algorithm to search shapes is also proposed according the analysis adopted for approximate verification of the spatial layout of pixels of each component in the tree. / Doutorado / Engenharia de Computação / Doutor em Engenharia Elétrica

Submitted by Luciana Ferreira (lucgeral@gmail.com) on 2014-11-20T11:31:40Z No. of bitstreams: 2 Dissertação - Danilo Alves Martins de Faria - 2013.pdf: 1188393 bytes, checksum: c56b169690f22bbbeaa2ee6fa46ade1c (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2014-11-20T14:17:45Z (GMT) No. of bitstreams: 2 Dissertação - Danilo Alves Martins de Faria - 2013.pdf: 1188393 bytes, checksum: c56b169690f22bbbeaa2ee6fa46ade1c (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2014-11-20T14:17:45Z (GMT). No. of bitstreams: 2 Dissertação - Danilo Alves Martins de Faria - 2013.pdf: 1188393 bytes, checksum: c56b169690f22bbbeaa2ee6fa46ade1c (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2013-10-23 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Network Design Problems (NDPs) are present in many areas, such as electric power distribution, communication networks, vehicle routing, phylogenetic trees among others. Many NDPs are classified as NP-Hard problems. Among the techniques used to solve them, we highlight the Evolutionary Algorithms (EA). These algorithms simulate the natural evolution of the species. However, in its standard form EAs have limitations to solve large scale NDPs, or with very specific characteristics. To solve these problems, many researchers have studied specific forms of representation of NDPs. Among these stands we show Node-Depth-Degre Encoding (NDDE). This representation produces only feasible solutions, regardless of the network characteristics. NDDE has two mutation operators Preserve Ancestor Operator (PAO) and Ancestor Change Operator (CAO) and the recombination operator EHR (Evolutionary History Recombination Operator) that uses historical applications of mutation, and was applied to NDPs more than one tree and had good results. Thus, this work proposes adapt EHR for NDPs classics represented by a single tree. In addition, two evolutionary algorithms are developed: the AE-RNPG, which uses only NDDE, with mutation operators. And the AE-EHR, which makes use of mutation operators and recombination operator EHR to the One Max Tree Problem. The results showed that the AE-EHR obtained better solutions than the EA-RNPG for most instances analyzed. / Problemas de Projeto de Redes (PPRs) estão presentes em diversas áreas, tais como reconfiguração de sistemas de distribuição de energia elétrica, projetos de redes de comunicação, roteamento de veículos, reconstrução de árvores filogenéticas entre outros. Vários PPRs pertencem à classe de problemas NP-Difíceis. Dentre as técnicas utilizadas para resolvê-los, destacam-se os Algoritmos Evolutivos (AE), cujo processo de resolução de um problema simula a evolução natural das espécies. Entretanto, os AEs em sua forma padrão também possuem limitações quanto a PPRs de larga escala, ou com características muito específicas. Para solucionar esses problemas, diversas pesquisas têm estudado formas específicas de estruturas de dados dos PPRs. Dentre essas destaca-se a representação Nó-Profundidade-Grau (RNPG). Essa representação produz apenas soluções factíveis, independente da característica da rede. A RNPG possui dois operadores de mutação Preserve Ancestor Operator (PAO) e Change Ancestor Operator (CAO) e o operador de recombinação EHR (Evolutionary History Recombination Operator), que utiliza o histórico de aplicações dos operadores de mutação, o qual tem sido aplicado a PPRs com mais de uma árvore com bons resultados. Este trabalho propõem a adequação do EHR para PPRs clássicos de uma única árvore. Além disso, são desenvolvidos dois algoritmos evolutivos: o AE-RNPG, que utiliza a RNPG somente com os operadores de mutação; e o AE-EHR, que faz uso tanto dos operadores de mutação quanto do operador de recombinação EHR para o problema da Árvore máxima. Os resultados obtidos mostram que o AE-EHR obtém melhores soluções do que o AE-RNPG para a maioria das instâncias analisadas.