Isometric Finger Pose Recognition with Sparse Channel SpatioTemporal EMG Imaging.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neuroscience

Background:

• High-fidelity myoelectric control is crucial for restoring function in amputees and individuals with neuromuscular diseases.
• Existing myoelectric control systems often face limitations in accuracy and real-time responsiveness.

Purpose of the Study:

• To investigate the efficacy of spatio-temporal imaging combined with convolutional neural networks (CNNs) for classifying electromyography (EMG) signals.
• To assess the performance of this method using sparse channel EMG data from a consumer-grade device.

Main Methods:

• Generated 10,572 images from 960 EMG samples of isometric finger poses from four subjects.
• Utilized a CNN model to classify these spatio-temporal EMG images.
• Implemented a 250ms continuous overlapping window for real-time classification.

Main Results:

• Achieved over 94% accuracy in classifying EMG samples.
• Successfully demonstrated real-time classification of 12 distinct finger poses.
• Validated the approach using sparse channel data from a consumer-grade EMG device.

Conclusions:

• Spatio-temporal imaging is a viable technique for enhancing EMG signal classification accuracy.
• CNNs can effectively interpret EMG data for high-fidelity prosthetic and orthotic control.
• This method shows promise for improving functional restoration in individuals with limb loss or neuromuscular disorders.