Differential Evolution: A Global Optimization Framework For Marker Registration and Scaling Musculoskeletal Models
Abstract
This thesis introduces an optimization framework for the scaling and registering of musculoskeletal models. Leveraging evolutionary algorithms, the framework navigates a highly dimensional and non-convex parameter space, exclusively using subject-specific motion-capture marker data. Monte Carlo simulations tested the framework's sensitivity to initial marker locations, demonstrating its robustness. The framework achieved an average RMS error of 0.77 cm in normal gait trials, representing a 44% improvement in RMS error over previous optimization methods (1.38 cm) and a 70% improvement over manual scaling (2.58 cm). This approach significantly reduces the need for expert intervention, enhances computational efficiency, and increases both accuracy and reproducibility for scaling musculoskeletal models.