Thesis: Deep learning meets real-time numerical simulation - Applications to surgical training, preoperative planning and surgical guidance

PhD defense: Andrea MENDIZABAL

Title: Deep learning meets real-time numerical simulation - Applications to surgical training, preoperative planning and surgical guidance

Team: IGG

Abstract: Many engineering applications require accurate numerical simulations of non-linear structures in real-time. Some important examples can be found in the field of medicine, in order to develop surgical training systems, or in the field of surgical navigation where augmented reality can bring significant improvements to the clinical gesture. To guarantee the accuracy of the simulations, patient-specific modeling must be pursued by taking into account personalized material parameters and boundary conditions. In the context of augmented surgery for instance, it is essential to perform an elastic registration between the preoperative and the intraoperative images. To this end, a patient-specific biomechanical model must be built to produce real-time finite-element simulations of the deformed organ. This is in practice very difficult to achieve as the problems to be solved are highly complex, in particular when non-linear deformations are considered.
In this work, we propose a method combining finite-element simulations and deep neural networks in order to satisfy the rapidity and accuracy requirements of medical applications. In particular, we present the U-Mesh framework, capable of predicting in real-time the shape of a highly deformable organ like the liver in order to guide surgeons during interventions where following the organ's deformation is crucial for the surgery to be successful.

The jury is composed by:

• M. Elías Cueto (Reviewer) - Professor at university of Zaragoza, Spain
• M. José David Martin-Guerrero (Reviewer) - Professor at University of Valencia, Spain
• M. Souvik Chakraborty (Examiner) - Researcher at Indian Institute of technology of Delhi, India
• M. Yohan Payan (Examiner and president of the jury) - Research director at the University of Grenoble Alpes, France
• M. Stéphane Cotin (Research director) - Research Director at the University of Strasbourg, France

The defense will take place on the 3^rd of December 2020 at 9h30 CET (UTC+1) at Salle de conférence Hygie - IHU Strasbourg (1 place de l’hôpital 67000 Strasbourg). A YouTube live will also be available.

Keywords: Real-time simulation, Deep neural networks, Finite element method, Data-driven simulation, Augmented surgery.