Deep Learning for Robust Decomposition of High-Density Surface EMG Signals

Blind source separation (BSS) algorithms, such as gradient convolution kernel compensation (gCKC), can efficiently and accurately decompose high-density surface electromyography (HD-sEMG) signals into constituent motor unit (MU) action potential trains. Once the matrix is blindly estimated on a signal interval, it also possible to apply same subsequent segments. Nonetheless, trained matrices are sub-optimal in noisy conditions require that incoming data undergo computationally expensive whitening. One unexplored alternative instead use paired HD-sEMG BSS output train model predict MU activations within supervised learning framework. A gated recurrent (GRU) network was both simulated experimental unwhitened using of gCKC algorithm. The results were validated by comparison with decomposition concurrently recorded intramuscular EMG signals. GRU outperformed at low signal-to-noise ratios, proving superior performance generalising new data. Using 12 seconds per recording, performed similarly gCKC, rates agreement 92.5% (84.5%-97.5%) 94.9% (88.8%-100.0%) respectively for against matched sources.