Personalization of Finite Element Head Models by SynthMorph-Based Mesh Morphing

Sammanfattning

Subject-specific finite element (FE) head models are increasingly used for simulating accurate biomechanical responses in traumatic brain injuries (TBI). Such models are commonly generated via mesh morphing, which is to deform a validated baseline model into individual mesh geometries obtained through medical imaging. While traditional iterative image registration algorithms have been widely used for mesh morphing, they are often multi-step, require manual intervention, and must balance registration accuracy with element mesh quality. This study proposes an integrated computational pipeline utilizing the deep learning framework SynthMorph to generate subject-specific FE head models. A subset of subjects from the Human Connectome Project (HCP) was used, and the Anatomically Detailed And Personalizable head Trauma (ADAPT) head model based on the ICBM152 template served as the baseline mesh. The performance of this approach was systematically evaluated and compared against traditional image registration-based methods (Demons+DRAMMS) using registration accuracy metrics (Dice Similarity Coefficient, 95th percentile Hausdorff Distance) and mesh quality indicators (Jacobian, Aspect Ratio, Solid Distortion Index). Results indicate that the SynthMorph-based pipeline maintained comparable registration accuracy, while preserving positive Jacobian values and maintaining mesh quality close to that of the baseline model. Overall, this work demonstrates that SynthMorph-based registration pipeline could be considered a highly efficient and robust alternative for subject-specific biomechanical modeling.