Surrogate Simulation of Subject-Specific Lateral Pinch via Deep Learning.
Journal:
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
PMID:
40039080
Abstract
Musculoskeletal modeling and simulation is often a lengthy and computationally expensive process, particularly when developing and using personalized models. We present a deep learning-based adaptive surrogate model for lateral pinch, which accepts both musculoskeletal parameters and muscle activations as input for personalization and simulation. This model matches traditional OpenSim forward dynamics with an average root-mean-squared error (RMSE) of 2.27 N, within standard errors of experimental measurements, while demonstrating sensitivity to both categories of input and performing thousand of simulations in seconds (10-1000x faster than traditional multi-body simulations). In addition to direct use as a surrogate, the differentiable nature of the model may support future use in optimization problems, while its flexibility may support adaptation to modeling of experimental data.