Our paper “Deform Any Liver in Real-Time” was presented at IEEE ISBI 2025 (International Symposium on Biomedical Imaging) in Houston, United States.
This work introduces DAL (Deform Any Liver), a real-time surrogate model that accurately predicts liver deformations given a set of applied forces. The key innovation lies in the use of a hypernetwork to condition a U-Net architecture on the liver shape, while the U-Net receives external forces as input to predict the resulting displacement field. This design enables the model to generalize across various patient geometries, overcoming the limitations of existing patient-specific approaches.
The model was trained, validated, and tested using real human liver geometries, with finite element method (FEM) simulations generating realistic deformation data. Results demonstrate the model’s effectiveness in accurately predicting deformations across diverse liver shapes in real-time.
Authors: Sidaty El Hadramy, Nicolas Padoy, and Stéphane Cotin
Read the paper: hal-04834198
