A forward dynamics framework for parameter optimization of the EMG-driven musculoskeletal model.

Area of Science:

• Biomechanics
• Musculoskeletal Modeling
• Computational Physiology

Background:

• Accurate estimation of muscle force and joint torque in subject-specific musculoskeletal models is hindered by challenges in determining muscle parameters.
• This study addresses methodological questions for an electromyography (EMG)-driven model integrated with a genetic algorithm and OpenSim API.

Purpose of the Study:

• To estimate muscle force and knee joint torque using a subject-specific EMG-driven musculoskeletal model.
• To validate the model's accuracy by comparing predicted torques with dynamometer measurements.

Main Methods:

• A Hill muscle model was employed, using filtered EMG data as input to calculate knee torque.
• The model incorporated parameters such as optimal fiber length and tendon slack length, tuned via a genetic-simulated annealing algorithm to minimize torque prediction error.
• Surface EMG data from quadriceps muscles were recorded during isometric knee tasks at various joint angles from eight participants.

Main Results:

• The proposed EMG-driven model achieved high accuracy, with an overall mean Root Mean Square (RMS) error of 3.7 Nm and an R-squared value of 0.97.
• Simulated torque curves closely matched measured torque curves when key muscle parameters were simultaneously adjusted.

Conclusions:

• Subject-specific, well-calibrated musculoskeletal models significantly enhance the prediction accuracy of muscle force and knee joint torque.
• The developed method demonstrates feasibility for generating personalized muscle-tendon unit (MTU) parameters for the knee with high precision and minimal error.