DeepAneurysm // DeepAneurysm

The idea behind our project is to modernize and leverage numerical Abdominal Aortic Aneurysm (AAA) simulations with Artificial Intelligence (AI) to make their use easier in the medical community and the procurement of the biomechanical patient-specific data possible within less than one minute, which is a real challenge. On the fundamental science side, and as mentioned above, our objective is to investigate state-of-the-art data-driven identification algorithms for high-dimensional spatio-temporal problems from previous high-fidelity simulations, possibly readjusted with patient data/measurements. We hope they will enable live visualizations of quantities and fields of interest (stress fields, risk assessment according to the evolution) for decision-making (surgical decision, surgical procedure).

Current computational Physics-based solvers can provide accurate results but are too time-consuming for medical use. A breakthrough is thus needed to lower the computational time by at least 3 or 4 orders of magnitude. We believe that a smart synergy between Full-Order models (FOM) and AI-based strategies can take up this challenge.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

The idea behind our project is to modernize and leverage numerical Abdominal Aortic Aneurysm (AAA) simulations with Artificial Intelligence (AI) to make their use easier in the medical community and the procurement of the biomechanical patient-specific data possible within less than one minute, which is a real challenge. On the fundamental science side, and as mentioned above, our objective is to investigate state-of-the-art data-driven identification algorithms for high-dimensional spatio-temporal problems from previous high-fidelity simulations, possibly readjusted with patient data/measurements. We hope they will enable live visualizations of quantities and fields of interest (stress fields, risk assessment according to the evolution) for decision-making (surgical decision, surgical procedure).

Current computational Physics-based solvers can provide accurate results but are too time-consuming for medical use. A breakthrough is thus needed to lower the computational time by at least 3 or 4 orders of magnitude. We believe that a smart synergy between Full-Order models (FOM) and AI-based strategies can take up this challenge.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Début de la thèse : 01/10/2025

Financement d'un établissement public Français

Université de Technologie de Compiègne

71 Sciences pour l'ingénieur

Etudiant MSc ou équivalent Analyse numérique Calcul scientifique Mécanique des fluides, mécanique des structures Machine learning / Scientifique Machine Learning Reduced-order models, deep learning Software, python programming
MSc student Scientific Computing Fluid mechanics, solid mechanics Machine learning / Scientifique Machine Learning Reduced-order models, deep learning Software, python programming