Ankle Joint Torque Prediction Using an NMS Solver Informed-ANN Model and Transfer Learning.

Area of Science:

• Biomechanics
• Computational Neuroscience
• Machine Learning

Background:

• Accurate prediction of joint torque is crucial for understanding human movement and developing assistive technologies.
• Existing models often struggle with inter-subject variability and limited training data.

Purpose of the Study:

• To develop and evaluate a hybrid artificial neural network (ANN) model integrating neuromusculoskeletal (NMS) features for enhanced ankle joint torque prediction.
• To investigate the impact of transfer learning on improving inter-subject prediction accuracy.

Main Methods:

• A hybrid-ANN model was developed, incorporating NMS-derived physical features (muscle force, joint torque) alongside experimental data (joint angles, electromyography).
• The hybrid-ANN was compared against baseline NMS and standard ANN models across various tasks (gait, isokinetic movements) and subject conditions (intra- and inter-subject).
• Transfer learning was implemented to enhance inter-subject prediction by leveraging knowledge from previously trained participants.

Main Results:

• The hybrid-ANN model demonstrated superior ankle joint torque prediction accuracy compared to baseline models.
• Integrating NMS features significantly improved prediction, particularly in inter-subject scenarios.
• Transfer learning further boosted inter-subject prediction accuracy, showing promise for data-limited situations.

Conclusions:

• Combining physics-informed NMS features with ANNs offers a powerful approach for accurate joint torque prediction.
• Transfer learning is a valuable technique for improving the generalizability of predictive models across different individuals.
• This hybrid approach holds potential for applications like designing control strategies for exoskeletons.