Emulating parametrized blood clot simulations through model order reduction combined with deep learning techniques

Alessandro Longhi (1) , Rodrigo Méndez Rojano (2) , Amélie Fau (3) , Didier Lucor (1, 4)

1 LISN - Laboratoire Interdisciplinaire des Sciences du Numérique
2 Cornell University [New York]
3 LMPS - Laboratoire de Mécanique Paris-Saclay
4 DATAFLOT - DAtascience, trAnsition, Fluid instability, contrOl, Turbulence LISN

Blood-contacting-medical devices are often affected by clotting complications. Computational models involving blood flow description interacting with thrombus formation through biochemical mechanisms have been proposed to help predicting thrombosis risk. However, they remain very expensive and plagued with various modelling and parametric uncertainties. Therefore, more efficient modeling strategies are needed in order to promote more systematic use of thrombosis simulations in medical device development and regulation. We investigate the combined use of reduced modelling and deep learning techniques to tackle this issue and test our approach on the clotting assay PFA-100\textregistered.

Domaines: Biomécanique [physics.med-ph]; Machine Learning [stat.ML]

Alessandro Longhi, Rodrigo Méndez Rojano, Amélie Fau, Didier Lucor. Emulating parametrized blood clot simulations through model order reduction combined with deep learning techniques. CFC2023 Cannes, Apr 2023, Cannes, France. ⟨hal-04460035⟩