MeiBRD: Meta-Learning Intraoperative Biomechanical Residual Deformation
Researchers have developed MeiBRD, a novel framework for intraoperative liver registration that combines biomechanical models with meta-learning. This approach learns a residual deformation function to correct predictions from biomechanical models, utilizing a graph neural diffusion function with geometry-aware attention on a 3D liver mesh. Experiments on a deformable liver phantom dataset show improved accuracy and generalization compared to existing methods, especially for unseen geometries and deformations. AI