A Case Series in Position-Aware Myoelectric Prosthesis Control Using Recurrent Convolutional Neural Network...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Rehabilitation Technology

Background:

• Myoelectric prosthesis control demands extensive user training.
• Transfer Learning (TL) offers a potential solution to reduce this training burden.
• Recurrent Convolutional Neural Networks (RCNNs) show promise with TL in offline analyses.

Purpose of the Study:

• To evaluate a real-time RCNN-based controller with TL (RCNN-TL) for myoelectric prostheses.
• To assess if RCNN-TL reduces user training time and improves functional task performance.
• To investigate RCNN-TL's effectiveness in mitigating the limb position effect.

Main Methods:

• Pre-trained an RCNN-TL model using forearm muscle signals from 19 participants.
• 8 participants tested the RCNN-TL and a Linear Discriminant Analysis (LDA-Baseline) controller in real-time.
• Collected data on training burden and functional task performance metrics.

Main Results:

• TL demonstrably reduced user training burden.
• RCNN-TL showed improved task performance durations during Grasp and Release phases compared to LDA-Baseline.
• Certain performance metrics for RCNN-TL were not improved or worsened.

Conclusions:

• TL is effective in reducing training requirements for myoelectric prosthesis control.
• RCNN-TL offers benefits in specific functional tasks but requires further optimization.
• Understanding training conditions and comprehensive metrics is crucial for advancing position-aware prosthesis control.