Evaluation of generic EMG-Torque models across two Upper-Limb joints

Area of Science:

• Biomechanics and Motor Control
• Biomedical Engineering
• Neuroscience

Background:

• Subject-specific electromyography-torque (EMG-torque) models require extensive calibration.
• Historically, generic EMG-torque models were assumed to be less accurate than subject-specific ones.

Purpose of the Study:

• To investigate the accuracy of generic EMG-torque models compared to subject-specific models.
• To evaluate the performance of generic models across different joints and degrees of freedom (DoF).

Main Methods:

• Developed generic one DoF EMG-torque models from subject-specific data.
• Evaluated models using elbow (N=64) and hand-wrist (N=9) datasets.
• Compared performance across subjects, DoF, and joints, including simpler Butterworth filter models.

Main Results:

• Subject-specific elbow models showed slightly better performance than generic elbow models (5.79% vs. 6.21% MVT error).
• No significant difference in accuracy between subject-specific and generic hand-wrist models.
• Generic elbow models performed better than generic hand-wrist models when applied across joints.
• Butterworth filter models showed no statistical difference compared to subject-specific models.

Conclusions:

• Generic EMG-torque models offer a simplified approach to training without substantial performance loss.
• These models facilitate potential transfer learning applications between different joints.
• The findings challenge the assumption that subject-specific calibration is always necessary for accurate EMG-torque modeling.