A Novel Framework for Cross-User Open-Set Myoelectric Pattern Recognition.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Myoelectric control systems translate muscle electrical activity into device commands.
• Cross-user variability and interference from outlier motions pose significant challenges to current myoelectric pattern recognition methods.
• Robustness and usability of myoelectric gestural interfaces require advanced pattern recognition techniques.

Purpose of the Study:

• To develop a robust myoelectric pattern recognition method.
• To simultaneously alleviate cross-user variability and outlier motion interference.
• To enhance the accuracy and reliability of myoelectric control for prosthetic devices.

Main Methods:

• A convolutional neural network (CNN)-based feature extractor was pre-trained on existing user data.
• Model transfer and adaptation were performed using limited labeled data from new users.
• A Euclidean metric-based prototypical loss function was employed for improved class separability and compactness.
• Inlier and outlier motions were identified using a prototype matching procedure.

Main Results:

• The proposed method achieved an average accuracy of 82.37 ± 1.21% for inlier motion recognition.
• The method demonstrated an average accuracy of 97.21 ± 2.65% for outlier motion rejection.
• Performance significantly outperformed existing methods (p < 0.05) in cross-user, open-set scenarios.

Conclusions:

• The developed method shows excellent performance in cross-user open-set myoelectric pattern recognition.
• A short and simple calibration routine is sufficient for effective model adaptation.
• This research provides a valuable solution for improving the robustness and usability of myoelectric gestural interfaces.