Dynamic Time Warping compared to established methods for validation of musculoskeletal models

Area of Science:

• Biomechanics
• Computational Modeling
• Musculoskeletal Simulation

Background:

• Multi-Body musculoskeletal simulation estimates unmeasurable joint forces and moments.
• Validation of these simulations is often qualitative due to limitations in comparing time-dependent signals.
• Quantitative validation methods struggle with phase and amplitude errors.

Purpose of the Study:

• To investigate the Dynamic Time Warping (DTW) algorithm's capability for comparing time-series data in musculoskeletal model validation.
• To quantify phase and amplitude errors using DTW.
• To contrast DTW's sensitivity with established metrics like Pearson correlation, cross-correlation, Geers metric, RMSE, and normalized RMSE.

Main Methods:

• Utilized two datasets for validation: direct validation (clinical gait analysis of trans-femoral amputees) and indirect validation (surface EMG of leg muscles in healthy subjects).
• Compared simulated forces/moments against measured socket-prosthesis data (direct validation).
• Compared simulated muscle activations against measured surface EMG data (indirect validation).

Main Results:

• Direct validation showed a positive linear relationship between RMSE/nRMSE and DTW.
• Indirect validation revealed a negative linear relationship between Pearson correlation and cross-correlation.
• DTW demonstrated good correlation with established methods suitable for specific validation types.

Conclusions:

• The Dynamic Time Warping (DTW) algorithm is proposed for both direct and indirect quantitative validation of musculoskeletal models.
• DTW effectively quantifies phase and amplitude errors, correlating well with task-specific metrics.
• For direct validation, DTW should be used alongside dimensional error metrics for clearer interpretation.