Evaluating EMG Feature and Classifier Selection for Application to Partial-Hand Prosthesis Control.

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Prosthetics and Orthotics

Background:

• Pattern recognition-based myoelectric control offers multi-degree-of-freedom control for upper-limb prostheses.
• Effectiveness for partial-hand amputations, particularly with retained wrist function, is less studied.
• Classification accuracy of hand motions across various wrist positions requires further investigation.

Purpose of the Study:

• Evaluate linear and non-linear pattern recognition algorithms for myoelectric control in partial-hand amputees.
• Assess the performance of optimal electromyography (EMG) feature subsets for classifying hand motions.
• Investigate the impact of different wrist positions on classification accuracy.

Main Methods:

• Compared linear discriminant analysis, linear, and non-linear artificial neural networks against quadratic discriminant analysis.
• Evaluated optimal EMG feature subsets against standard time-domain and time-domain/autoregressive feature sets.
• Tested algorithms on 16 non-amputees and 4 partial-hand amputees across various hand motions and wrist positions.

Main Results:

• Linear discriminant analysis and artificial neural networks significantly outperformed quadratic discriminant analysis.
• Incorporating data from multiple wrist positions substantially reduced classification error for amputees.
• An optimal feature selection method significantly outperformed traditional feature sets for classification accuracy.

Conclusions:

• Advanced pattern recognition algorithms and optimal feature selection enhance myoelectric control for partial-hand amputees.
• Utilizing multiple wrist positions improves the classification of hand movements, crucial for functional prosthetics.
• The proposed feature selection method serves as an effective filter for optimizing EMG-based control systems.