Using recurrent artificial neural network model to estimate voluntary elbow torque in dynamic situations

Area of Science:

• Biomechanics
• Neuroscience
• Artificial Intelligence

Background:

• Muscle modeling is crucial for analyzing body movement.
• Previous research primarily focused on static conditions, leaving dynamic voluntary movements understudied.
• The relationship between electromyographic (EMG) signals and joint torque in dynamic situations requires further investigation.

Purpose of the Study:

• To evaluate a recurrent artificial neural network (RANN) for estimating elbow complex joint torque during voluntary dynamic movements.
• To compare the predictive accuracy of models using EMG and kinematic data versus EMG data alone.
• To determine the contribution of joint angle and angular velocity to torque prediction accuracy.

Main Methods:

• EMG signals and kinematic data (joint angle, angular velocity) were collected from six healthy subjects.
• A RANN was trained and tested using data from dynamic elbow movements under varying loads and velocities.
• Two models were compared: one with EMG and kinematic inputs, and another with EMG inputs only.

Main Results:

• The model incorporating EMG and kinematic data achieved lower root mean squared errors (RMSE) (0.17 ± 0.03 Nm training, 0.35 ± 0.06 Nm testing) compared to the EMG-only model (0.57 ± 0.07 Nm training, 0.73 ± 0.11 Nm testing).
• Kinematic data, specifically joint angle and angular velocity, significantly enhanced the accuracy of torque prediction.
• The RANN demonstrated effective performance in estimating joint torque during dynamic voluntary movements.

Conclusions:

• Integrating EMG signals with kinematic data provides superior performance for joint torque prediction in dynamic movements.
• Joint angle and angular velocity are vital inputs for accurate torque estimation during voluntary dynamic actions.
• This approach advances muscle modeling and body segmental motion analysis in dynamic scenarios.