Intuitive real-time control strategy for high-density myoelectric hand prosthesis using deep and transfer learning.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Artificial Intelligence in Medicine

Background:

• Myoelectric hand prostheses aim to restore function for upper-limb amputees but suffer from low user acceptance due to non-intuitive control interfaces.
• Existing systems often require complex calibration and lack responsiveness, hindering seamless integration into daily activities.

Purpose of the Study:

• To develop and evaluate a highly intuitive, responsive, and reliable real-time myoelectric hand prosthesis control strategy.
• To demonstrate the efficacy of a novel human-machine interface for improved prosthesis usability and user acceptance.

Main Methods:

• Utilized surface high-density electromyography (HD-EMG) for muscle signal acquisition.
• Implemented a convolutional neural network (CNN) for adaptive, user-specific gesture recognition.
• Employed a transfer learning approach to minimize training time and simplify calibration.

Main Results:

• Achieved high real-time performance with mean and median positive predictive values (PPV) of 93.43% and 100% for 6 grip modes.
• Demonstrated instant accessibility between gesture states, eliminating mode switching for natural control.
• System latency for correct prediction was under 116 ms.
• Transfer learning reduced setup time to under 10 minutes, an 89.4% improvement over traditional methods.

Conclusions:

• The proposed HD-EMG and CNN-based control strategy offers an intuitive and efficient solution for myoelectric hand prostheses.
• The system's reduced setup time and enhanced responsiveness significantly improve potential user acceptance and rehabilitation outcomes.
• This approach paves the way for more natural and seamless control of prosthetic devices.