A Parallel-Double-Thread Online EMG Decomposition Approach by Alternating and Updating Motor Unit Separation Vectors

Online electromyogram (EMG) decomposition can be used to extract motor unit (MU) discharge information for accurate decoding of dexterous finger movements. However, the non-stationary MU activities can degrade the performance of MU separation vectors in the blind source separation-based EMG decomposition technique under the real-time condition. In this preliminary study, we developed an improved parallel-double-thread (IPDT) online EMG decomposition approach. Specifically, multiple separation vectors extracted from different neural drive levels were assigned to each MU and were alternated according to the neural drive level during online decomposition. In addition, compared with the previous PDT method, the IPDT method utilized the CNN-based motor unit action potential classification technique to realize the tracking of MUs and then updating of the separation vectors of specific MUs. The IPDT method was tested and compared with two previous methods using the synthetic EMG signals involving both MUAP profile variation and MU recruitment/de-recruitment. The results showed that the proposed IPDT method obtained the best online EMG decomposition performance. The further exploration of our method may provide a robust MU discharge information extraction method for long-term continuous and accurate decoding of dexterous finger movements.