Using High Performance Computing to allow clinical compatible performance for biomechanical Finite Element simulations
We have developed a model to describe inward brain-shift with viscoelastic biomechanical modeling varying three surgery parameters: craniotomy position, cerebrospinal fluid drainage level and intraoperative head position. We propose to develop a real-time estimation of the inward brain-shift displacement with deep learning from only patient-specific MRI without using time-consuming Finite Element building. In this study, we demonstrate that the combination of HPC-generated FEM training data and a U-net approach is a promising solution for clinically compatible performance.