Prosthesis-guided training of pattern recognition-controlled myoelectric prosthesis.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Pattern recognition enhances intuitive control of myoelectric prostheses.
• Screen-guided training (SGT) is the standard for collecting electromyography (EMG) data, requiring users to sync muscle contractions with screen prompts.
• This process is data-intensive and can divert user attention from the prosthesis.

Purpose of the Study:

• To introduce and evaluate Prosthesis-guided training (PGT) as a novel, hardware-free data collection method for myoelectric prostheses.
• To compare the training data and classifier performance resulting from PGT versus traditional SGT.
• To assess the impact of PGT's inclusion of transient EMG signals on classifier accuracy and real-time prosthesis control.

Main Methods:

• Collected EMG data using both SGT and PGT methods from participants.
• Trained pattern recognition classifiers using data from each training method.
• Evaluated user performance on a task using classifiers trained with SGT and PGT data.

Main Results:

• PGT data, including transient EMG signals, led to a decrease in classifier accuracy compared to SGT.
• Despite lower accuracy, subjects using classifiers trained with PGT data completed the performance task significantly faster.
• PGT allows users to maintain focus on the prosthesis during data collection.

Conclusions:

• PGT is a viable, hardware-free alternative for collecting training data for myoelectric prostheses.
• Classifiers trained with PGT data may better represent real-time muscle activity, leading to improved task completion speed.
• Future research should explore optimizing PGT to balance classifier accuracy and real-time performance benefits.