EMGCipher: Decoding Electromyography for Upper-limb Gesture Classification with Explainable AI for Resource...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Surface electromyography (sEMG) and deep learning (DL) are crucial for assistive limb devices.
• Current DL models for gesture classification lack transparency in their decision-making processes.

Purpose of the Study:

• Introduce EMGCipher, an interpretable DL framework for sEMG-based upper-limb gesture classification.
• Enhance transparency in DL models by quantifying the significance of sensors and features.
• Improve gesture classification performance and efficiency through optimized sensor and feature utilization.

Main Methods:

• Developed EMGCipher, a DL framework integrating low-level sEMG features with DL model insights.
• Quantitatively assessed the probabilistic significance of input sensors and features.
• Validated the framework on the Ninapro DB5 dataset for upper-limb gesture classification.

Main Results:

• EMGCipher demonstrated effective sensor-wise and feature-wise interpretation.
• The framework successfully bridged the gap between interpretability and performance.
• Probabilistic significance assessment provided insights into model decision-making.

Conclusions:

• EMGCipher offers a transparent approach to sEMG-based gesture classification.
• The framework has the potential to optimize sensor and feature selection for enhanced performance.
• This interpretability can lead to more efficient and reliable assistive limb devices.