Machine Learning-Based Gait Mode Prediction for Hybrid Knee Prosthesis Control.

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Machine Learning in Prosthetics

Background:

• Hybrid prosthetic knees integrate passive and active mechanisms for transfemoral amputees.
• Seamless transitions between passive and active modes require accurate gait mode prediction.
• Data imbalance poses a challenge for classifier performance in gait mode prediction.

Purpose of the Study:

• To evaluate machine learning classifiers for predicting passive and active gait modes in hybrid prosthetic knees.
• To investigate the use of generative adversarial networks (GANs) for augmenting data and improving classification performance.
• To identify optimal classifiers for real-time gait mode prediction in transfemoral amputees.

Main Methods:

• Collected gait data from nine unilateral transfemoral amputees during various ambulation tasks (level ground, inclines, stairs).
• Evaluated multiple machine learning classifiers (e.g., linear discriminant analysis, random forest) for gait mode prediction.
• Developed and applied a generative adversarial network (GAN) to generate synthetic data for classifier training.

Main Results:

• Linear discriminant analysis showed high sensitivity for the active mode (stair ascent).
• Random forest achieved the highest overall classification accuracy.
• Training classifiers with GAN-generated synthetic data significantly improved classifier sensitivity.

Conclusions:

• Machine learning, particularly linear discriminant analysis and random forest, shows promise for gait mode prediction in hybrid prosthetic knees.
• Generative adversarial networks effectively address data imbalance, enhancing classifier performance.
• Improved gait mode prediction can lead to more intuitive and responsive prosthetic knee control for transfemoral amputees.