L'objective de la thèse est de comprendre s'il est envisageable d'utiliser les modèles qui décrivent la croissance tumorale (systèmes d'EDP) pour des applications médicales. En particulier, les modèles paramétriques sont calibrés en utilisant les données d'imagerie médicale d'un patient. Une fois calibré, le modèle donne une représentation de la croissance tumorale. Des techniques différentes sont proposées. Un approche classique basé sur la sensibilité est comparé à un approche réduit basé sur la Proper Orthogonal Decomposition. Des cas réalistes concernants l'étude des métastases dans les poumons ont été mis à point en collaboration avec l'Institut Bergonié. Des exigence pratique de traitement de l'image ont motivé l'étude des méthodes de recalage non-rigide des images et parmi ceux là, le transport optimale. Un étude de la numérique du problème de Monge-Kantorovich est décrit, avec des cas test numérique. Des applications concernants l'application de la distance de Wasserstein à la réduction de modèle sont envisagées. / The main purpose of this work was to understand if and wether PDE based modeling of tumor growth may be used in realistic applications. Models proposed in the literature are parametric. The goal is to identify parameters in such a way that the pathology evolution of a given patient is recovered. The identification is performed by means of inverse problems, taking medical images as data.Different techniques were tested: a classical Sensitivity approach is compared to a reduced one, based on Proper Orthogonal Decomposition. Realistic cases were set up in collaboration with Institut Bergonié, concerning lung metastasis evolution.Practical needs when dealing with medical images pushed us to interest to Optimal transport theory and Monge-Kantorovich problem. A numerical study was carried out and a family of lagrangian methods proposed. A perspective on the using of Wasserstein distance in model reduction concludes this work.

Le cancer du sein est le cancer le plus fréquent chez les femmes. Il y a une multitude de solutions proposées concernant une éventuelle intervention médicale pour le cancer du sein ‐ une en particulier est la chirurgie mammaire conservatrice (tumoréctomie). Le but de la tumoréctomie est de parvenir à un contrôle local du cancer, ainsi que de préserver une forme du sein qui satisfait les besoins esthétiques de la femme. Bien que ces objectifs sont généralement atteint, il reste encore parfois des résultats inattendus,tels qu'une tumeur récurrence locale, ou des résultats cosmétiques insuffisants.L'objectif de cette thèse est de proposer une plateforme de calcul, qui contribue à la tumoréctomie. Cela comprend:1) Une étude de la dynamique de croissance des tumeurs du sein.2) Une étude sur la prédiction du contour du sein grâce a la chirurgie virtuelle.3) Un modèle de calcul de la forme finale du sein après cicatrisation. / Breast cancer is the most common cancer among women in the developed as well as the developing countries. There are a plethora of proposed solutions regarding possible medical interventions for breast cancer–one in particular is Breast Conserving Therapy (BCT). BCT comprises of complete surgical excision of the tumor (partialmastectomy), and post-operative radiotherapy for the remaining breast tissue. This is a feasible treatment for most women with breast cancer. The goal of BCT is toachieve local control of the cancer, as well as to preserve breast shape that appeases awoman’s cosmetic concerns. Although these goals are usually achieved, there are still occasional unexpected results, such as reexcision of the tumor due to a positive margin assessment, tumor local recurrence, unsatisfactory cosmetic results, and breastpain. Other than surgical experience and judgment, there are currently no toolswhich can predict the outcome of partial mastectomy on the contour and deformity of the treated breast. The objective of this dissertation is to propose computational framework, which contributes to BCT operations, this was achieve by exploring two areas.On the one hand we developed a multiscale model adapted for breast cancer tumor growth, ductal carcinoma in situ (DCIS). The model features included: nutrients growth limitation, wall degradation enzyme and HER2 chemical expression tumor phenotype. Our model successfully simulate some pattern of DCIS carcinoma.Among the interesting result we showed that the enzyme contributed to a greater tumor size and that when HER2 was over expressed, the growth limiting factor wasthe EGFR. On the other hand, we developed a virtual surgery box to simulate BCT surgery. The box will input MRI patient data and will output cosmetic and functional indicator to rate the impact of the surgery. It appears that stiffness of the tissue, resection radius as well as the lump quadrant location are the most sensitive parameters to the indicators. A healing model was also embedded to simulate the wound closure after resection, this model was stress dependent and illustrate anasymmetric wound closure progression.The tools developed in this research allows a new type of field convergence between the surgery and computation field. At the local level it will allow surgeons and patient to be able to communicate on the pertinence and necessity of performing alumpectomy surgery, enabling to anticipate the possible outcome of the operation.On the global aspect this type of tool gives birth to a new type of field: computational surgery, where computer scientist and surgeons work hand in hand to provide the best and the most reliable service to the patients.