Data structures influence the number of muscle synergies and reconstruction effect across trials

Muscle synergy analysis featured with low dimensions was implemented extensively to operate neural control and rehabilitation assessment. Several researches suggested that the synergy number could be used as a physiological marker of motion deficiency, like the stroke and cerebral palsy children. However, it was still vague that the impact of data structure on synergy dimensionality, which was a fundamental question to be addressed and to help us better use the synergy number as an assessment metric. Therefore, we extracted synergies from three structures of electromyogram (EMG) using Nonnegative Matrix Factorization Algorithm (NNMF), including single-trial EMGs (SIN), average (AVE) and concatenation of all-trial EMGs (CON). Results indicated the significant impact of data structures on synergy numbers. Further, we also calculated the reconstruction effect of EMGs across trials to examine ability of the three structures to capture trial-based features. It suggested that synergies extracted by SIN captured more features of the original trial than other trials, and CON and AVE could both invariably and adequately reconstruct EMGs of four trials but with higher extent of CON than AVE. These results were beneficial in guiding synergy-based researches in rehabilitation assessment and control to deal with EMGs of multiple trials.