Facial muscle mapping and expression prediction using a conformal surface-electromyography platform

Facial muscles are uniquely attached to the skin, densely innervated, and exhibit complex co-activation patterns enabling fine motor control. Facial surface Electromyography (sEMG) effectively assesses muscle function, yet traditional setups require precise electrode placement and limit mobility due to mechanical artifacts. Signal extraction is hindered by noise and cross-talk from adjacent muscles, making it challenging to associate facial muscle activity with expressions. We leverage a novel 16-channel conformal sEMG system to extract meaningful electrophysiological data from 32 healthy individuals. By applying denoising and source separation techniques, we extracted independent components, clustered them spatially, and built a facial muscle atlas. Furthermore, we established a functional mapping between these clusters and specific muscle units, providing a framework for understanding facial muscle activation. Using this foundation, we demonstrated a deep-learning model to predict facial expressions. This approach enables precise, participant-specific monitoring with applications in medical rehabilitation and psychological research.