Nariman Khaledian, Tangram. Capturing Contact in Mitral Valve Dynamic Closure with Fluid-Structure Interaction Simulation.
Realistic fluid-structure interaction (FSI) simulation of the mitral valve opens the way toward planning for surgical repair. In the literature, blood leakage is identified by measuring the flow rate, but detailed information about closure efficiency is missing. We present in this paper an FSI model that improves the detection of blood leakage by building a map of contact.
Methods – Our model is based on the immersed boundary method that captures a map of contact and perfect closure of the mitral valve, without the presence of orifice holes, which often appear with existing methods. We also identified important factors influencing convergence issues.
Results – The method is demonstrated in three typical clinical situations: mitral valve with leakage, bulging, and healthy. In addition to the classical ways of evaluating MV closure, such as stress distribution and flow rate, the contact map provides easy detection of leakage with identification of the sources of leakage and a quality assessment of the closure.
Conclusions – Our method significantly improves the quality of the simulation and allows the identification of regurgitation as well as a spatial evaluation of the quality of valve closure. Comparably fast simulation, ability to simulate large deformation, and capturing detailed contact are the main aspects of the study.
Ongoing work – Currently, we are focused on implementing more complex material characteristics in the model to simulate a more realistic behavior of the mitral valve motion. Implementation of real mitral valve geometry based on medical imaging is another ongoing work that we are currently focused on.
David Desobry, Pixel. Quadrilateral meshing for large deformations.
We present a fully automatic method to produce quadrilateral meshes that stay valid despite deformations during a numerical simulation. Our approach is based on the Quadcover algorithm, used in computer graphics to generate high quality quad meshes. Frame field based algorithm like Quadcover are currently not used for simulations of high deformation because the quadrilateral meshes generated have no reason to remain of sufficient quality when we deform their geometry. We explain in this talk how we can take into account deformations that we already computed to produce quad meshes adapted to these coming geometrical changes.
Ana Rodriguez Cordero, Caramba. Differential cryptanalysis in block ciphers
In symmetric cryptography, block ciphers are widely used to secure the exchange of messages between parties. Cryptanalysis is responsible for testing the security of these ciphers to ensure that communications are secure. During this talk we will see what block ciphers are and how to study their security against differential attacks.
Radhouane Jilani, Tangram. Simulation prédictive pour la neuroradiologie interventionnelle.
Les AVC ischémiques tuent chaque année plusieurs millions de personnes dans le monde. La thrombectomie manuelle, consistant à amarrer le caillot sanguin pour déboucher l’artère est l’un des traitements les plus sûrs et les plus efficaces actuels. Elle nécessite la navigation d’un cathéter dans l’arbre vasculaire du patient jusqu’à atteindre le caillot, ce qui est un geste difficile. L’objectif de ma thèse est le développement d’une simulation précise et interactive qui aide à la planification d’une telle opération. Ainsi, il est crucial de modéliser avec précision le cathéter, son environnement et leurs interactions. Le modèle de cathéter de cosserat sera utilisé car il offre le meilleur compromis entre temps de calcul et précision. Les vaisseaux seront modélisés implicitement et un nouvel algorithme de gestion des contacts sera développé. Après une introduction à ma thèse et à l’état de l’art, je présenterai trois simulations qui reproduisent une expérience réelle de déploiement de cathéter.
Tom Masini, ABC. Dimension de Natarajan à marge des PMC.
Nous nous intéressons à l’utilisation des dimensions combinatoires pour majorer la probabilité d’erreur des classifieurs multi-classes à marge. Dans ce contexte, la dimension combinatoire de référence est la gamma-dimension. Cependant, son utilisation pâ tit de l’absence d’un bon résultat structurel. Nous proposons de contourner cette difficulté en substituant à cette dimension la dimension de Natarajan à marge. Pour illustrer le potentiel de cette démarche, nous étudions le cas particulier des perceptrons multi-couches.
Bastien Laboureix, Adagio. Connexité des hyperplans arithmétiques.
Quand vous étiez petits, on vous a dit que les droites étaient continues et n’avaient pas d’épaisseur. Mais, en prenant une image numérique de droite et en zoomant au maximum, on observe plutô t un amas de pixel qu’une forme rectiligne continue. Et si nous étudiions ces droites numériques, discrètes, plutô t que leurs homologues euclidiennes ? Et si nous nous intéressions à une nouvelle géométrie ? Car, au-delà des droites, on peut regarder des segments, des polygones, voire monter en dimension et étudier des plans, des sphères, des hyperplans. Evidemment, la géométrie euclidienne ne s’applique pas sur les pixels : plus de division, adieu les limites, tout n’est plus qu’arithmétique ! Recommençons donc la géométrie depuis le début, via un problème simple sur les plans !
Yoann Coudert-Osmont, Pixel. Closed space-filling curves with controlled orientation for 3D printing.
We explore the optimization of closed space-filling curves under orientation objectives. By solidifying material along the closed curve, solid layers of 3D prints can be manufactured in a single continuous extrusion motion. The control over orientation enables the deposition to align with specific directions in different areas, or to produce a locally uniform distribution of orientations, patterning the solidified volume in a precisely controlled manner. Our optimization framework proceeds in two steps. First, we cast a combinatorial problem, optimizing Hamiltonian cycles within a specially constructed graph. We rely on a stochastic optimization process based on local operators that modify a cycle while preserving its Hamiltonian property. Second, we use the result to initialize a geometric optimizer that improves the smoothness and uniform coverage of the cycle while further optimizing for alignment and orientation objectives.
Youssef Assis, Tangram. Aneurysm detection using deep learning.
The detection of intracranial aneurysms from Magnetic Resonance Angiography (MRA) 3D images is a rapidly growing problem of clinical importance, which is not only time-consuming for radiologists but also extremely difficult to automate. To overcome data scarcity and class imbalance problems related to aneurysm data, state-of-the-art methods are model-centric approaches, which are focused on network architecture and loss functions, but only achieve fair sensitivity with a high false positives rate. Our work mainly follows a data-centric approach, where we exploit the maximum of our weak annotations, a small-patch approach was combined with an adapted data sampling strategy, taking into account the specificities of aneurysms. Moreover, we investigated the impact of various network architectures for aneurysm detection, and compared them using performance metrics that are more relevant to object detection.
Leo Valque, Gamble. Rounding polygonal meshes
To build a 3D mesh, Boolean operations are often used. But the resulting points have rational coordinates and each term of the rational requires greater precision, so you will probably need to round the coordinates of those points. However, rounding can create intersections and corrupt your mesh. I will present our new algorithm to solve this problem in 3D and the first results of its implementation.
