Personalized Parameter Setting in Musculoskeletal Models Through Multitrajectory Optimization

Musculoskeletal models are indispensable tools in biomechanics, offering insights into muscle dynamics and joint mechanics. However, the parameters of a personalized musculoskeletal model are nonidentifiable when multiple parameters compensate for each other to produce similar force outputs, posing challenges to model accuracy and reliability. This study introduces a multitrajectory optimization framework integrated with subject-specific modeling to address this issue. By incorporating diverse movement tasks within a simple biceps curl context, the proposed approach narrows the parameter space, introducing constraints that can enhance model identifiability and robustness under specific conditions. Unlike traditional single-task optimization, this framework employs a dual-stage process: a global search using particle swarm optimization (PSO) to explore the solution space, followed by local refinement via Pattern Search to achieve precise parameter estimates. Applied to biceps curl tasks, this method reduced optimization convergence error by 97.9% and validation error by 99.2% on an unseen movement task compared to single-task optimization. These results highlight the framework's effectiveness in improving parameter estimation accuracy and suggest generalizability across the tested movement conditions. The integration of optimization techniques provides a promising approach for addressing challenges in musculoskeletal modeling. By improving model reliability and precision under simplified conditions, this work offers preliminary insights for potential applications in clinical rehabilitation, sports science, and ergonomic design. Future efforts will refine neuromuscular control representations and integrate dynamic subject-specific data to extend this framework's applicability beyond joint angle estimation to more complex movements and musculoskeletal outputs.