A signal-based approach for assessing the accuracy of high-density surface EMG decomposition

This study introduces a novel and computationally efficient metric for assessment of accuracy of decomposition of high-density surface EMG signals. The metric, so called Pulse-to-Noise Ratio (PNR), builds on the results of the previously published Convolution Kernel Compensation (CKC) decomposition technique and is applied to every identified motor unit (MU), without any significant computational or experimental cost. As validated on both synthetic and experimental signals with different spatial supports, the proposed PNR metrics correlates significantly with both the sensitivity and the false alarm rate of the identified MU discharges. In our study, all the MUs identified with PNR larger than 30 dB exhibited sensitivity larger than 90 % and false alarm rate below 1 %, so that 30 dB can be used as a practical threshold in PNR for assuring highly accurate MU spike train identification.