Correlation analysis of electromyogram signals for multiuser myoelectric interfaces.

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Signal Processing

Background:

• Myoelectric interfaces face challenges adapting to individual user movement styles and limb differences.
• Electromyogram (EMG) signal variations between individuals hinder practical application of current myoelectric interfaces.

Purpose of the Study:

• To develop a multiuser myoelectric interface that easily adapts to novel users and maintains high movement recognition performance.
• To implement a style-independent feature transformation framework using Canonical Correlation Analysis (CCA).

Main Methods:

• A CCA-based mapping projects diverse users' data onto a unified-style space for style-independent feature extraction.
• A classifier is trained on these style-independent features, followed by calibration for new users.
• Novel user features are projected into the unified-style space for classification.

Main Results:

• The proposed method achieved >83% accuracy across multiple users with diverse movement styles.
• The framework demonstrated >82% average accuracy when trained on able-bodied subjects and tested on amputees after calibration.
• Successfully overcame individual differences in EMG signals for improved myoelectric control.

Conclusions:

• The CCA-based framework offers a robust solution for style-independent myoelectric control, enhancing adaptability and performance.
• This approach significantly improves the practical applicability of myoelectric interfaces for both able-bodied users and amputees.
• The developed method paves the way for more intuitive and reliable prosthetic limb control.