EMGCipher: Decoding Electromyography for Upper-limb Gesture Classification with Explainable AI for Resource Optimization.
Journal:
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
PMID:
40040141
Abstract
Assistive limb devices often employ surface electromyography (sEMG) and deep learning (DL) models for gesture classification. While DL models effectively classify diverse upper-limb gestures, their decision-making mechanisms often lack transparency. To address this, we introduce EMGCipher, an interpretable DL framework for upper-limb gesture classification using sEMG. It aims to bridge the gap between interpretability and performance by combining low-level sEMG feature representations with DL model-derived knowledge, quantitatively assessing the probabilistic significance of input sensors and features in gesture classification. Experiments on the Ninapro DB5 dataset demonstrate EMGCipher's effectiveness in sensor-wise and feature-wise interpretation, demonstrating its potential to optimize the usage of sensors and features for improved gesture classification performance and efficiency.