A CNN-LSTM model for six human ankle movements classification on different loads.

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Signal Processing

Background:

• Current robot-assisted rehabilitation struggles with accurate ankle movement decoding from surface electromyogram (sEMG) signals.
• Existing methods face limitations in classifying sufficient movement types and suffer from high computational costs and load variations.
• The direct use of raw sEMG signals in neural networks increases computational burden and reduces classification accuracy.

Purpose of the Study:

• To improve the classification accuracy of six ankle movements for robot-assisted bilateral rehabilitation.
• To reduce the computational cost associated with processing raw sEMG signals.
• To enhance the robustness of ankle movement intention decoding against load variations.

Main Methods:

• A novel approach combining a convolutional neural network (CNN) and long short-term memory (LSTM) model was developed.
• The Boruta algorithm was employed for the first time to select optimal time-domain features from sEMG signals.
• A two-step method was implemented, feeding selected sEMG features into the CNN-LSTM model instead of raw signals.

Main Results:

• The proposed CNN-LSTM model achieved a high classification accuracy of 95.73% for six ankle movements.
• Feeding selected sEMG features into CNN-LSTM, CNN, and LSTM models significantly improved accuracy compared to using raw sEMG.
• The two-step method enhanced overall accuracy from 73.23% to 93.50% in identifying ankle movements under varying loads.
• The CNN-LSTM model demonstrated superior performance over CNN, LSTM, and Support Vector Machine (SVM) models.

Conclusions:

• The proposed feature selection and CNN-LSTM model effectively decode ankle movement intention for enhanced robot-assisted rehabilitation.
• Reducing model parameters by feeding selected features significantly lowers computational cost.
• The method provides a robust and accurate solution for real-time ankle movement classification in rehabilitation settings.