A peel-off convolution kernel compensation method for surface electromyography decomposition

Objective: Estimating discharge patterns of motor units by electromyography (EMG) decomposition has shown promising perspectives in neurophysiologic investigations and human-machine interfaces. However, the number of decoded motor units is still limited, especially under high excitation and noisy conditions, which is a major barrier to the broad application of EMG decomposition. In this work, we applied a peel-off decomposition strategy to the convolution kernel compensation (CKC) algorithm to extract more motor units non-invasively. Methods: EMG signals were firstly decomposed into motor unit spike trains (MUSTs) using the CKC. After each decomposition, the motor unit action potentials (MUAPs) were reconstructed and subtracted from the EMG signals. Then the same CKC decomposition was applied to the residual signals. This peel-off procedure was repeated until no valid motor units were identified. The proposed decomposition strategy was validated on synthetic EMG signals by convolving simulated MUSTs and experimentally extracted MUAPs under multiple excitations and noise conditions. Then the decomposition performance was evaluated on experimental data. Main results: Compared with the classic CKC method, the number of the identified motor units from synthetic signals was significantly increased by 7 to 43 in each decomposition. Moreover, the average agreement between identified MUST and ground truth was 0.80 +/- 0.20, indicating a high decomposition accuracy. From experimental EMG signals, the peel-off method could identify more motor units (>20) with high confidence than the classic method (<5) across three excitation levels. Conclusion and Significance: These results demonstrate the efficiency of the proposed method in identifying more motor units from EMG signals, extending the potential applications of surface EMG decomposition for neural decoding.