Toward Hand Gesture Recognition Using a Channel-Wise Cumulative Spike Train Image-Driven Model.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Natural human-machine interaction relies on recognizing hand gestures from neural signals, crucial for prosthetics and rehabilitation.
• Current methods face challenges in associating motor unit neural control signals with specific gestures.

Purpose of the Study:

• To introduce a novel channel-wise cumulative spike train (cw-CST) image-driven model (cwCST-CNN) for enhanced hand gesture recognition.
• To leverage spatial activation patterns of motor unit firings for improved distinction of motor intentions.

Main Methods:

• Decomposition of motor unit cw-CSTs from high-density surface electromyography using spatial spike detection.
• Reconstruction of cw-CSTs into images based on spatial recording positions.
• Classification of gestures using a customized convolutional neural network (cwCST-CNN).

Main Results:

• The proposed cwCST-CNN achieved a high classification accuracy of 96.92% ± 1.77% across 10 gestures and 10 subjects.
• Comparison with RMS-based and cw-CST discharge rate methods showed superior performance of cwCST-CNN.
• Analysis indicated better gesture separability and consistency for cw-CST features compared to RMS features.

Conclusions:

• The cwCST-CNN model offers a significant advancement in hand gesture recognition accuracy using neural drive signals.
• This approach provides a new solution for natural human-machine interaction, particularly in advanced prosthetic control and rehabilitation applications.