Multimodal Neural Operators for Real-Time TBI Biomechanical Modeling

A study evaluates multimodal neural operator architectures for real-time biomechanical modeling of traumatic brain injury (TBI). The research frames TBI modeling as a multimodal operator learning problem, integrating volumetric neuroimaging, demographic parameters, and acquisition metadata. Two fusion strategies are tested: field projection for Fourier Neural Operator (FNO) architectures and branch decomposition for Deep Operator Networks (DeepONet). Four models—FNO, Factorized FNO, and others—are evaluated for predicting full-field brain displacement from MRE data. The goal is to overcome the computational expense of finite element solvers, enabling clinical deployment. The work is published on arXiv (2510.03248) and addresses the underexplored integration of volumetric imaging with scalar metadata for biomechanical predictions.