How are Muscle Synergies Affected by Electromyography Pre-Processing?

Muscle synergies have been used for decades to explain a variety of motor behaviors, both in humans and animals and, more recently, to steer rehabilitation strategies. However, many sources of variability such as factorization algorithms, criteria for dimensionality reduction and data pre-processing constitute a major obstacle to the successful comparison of the results obtained by different research groups. Starting from the canonical EMG processing we determined how variations in filter cut-off frequencies and normalization methods, commonly found in literature, affect synergy weights and inter-subject similarity (ISS) using experimental data related to a 15-muscles upper-limb reaching task. Synergy weights were not significantly altered by either normalization (maximum voluntary contraction – MVC – or maximum amplitude of the signal - SELF) or band-pass filter ([20–500 Hz] or [50–500] Hz). Normalization did, however, alter the amount of variance explained by a set of synergies, which is a criterion often used for model order selection. Comparing different low-pass (LP) filters (0.5 Hz, 4 Hz, 10 Hz, 20 Hz cut-offs) we showed that increasing the low pass filter cut-off had the effect of decreasing the variance accounted for by a set number of synergies and affected individual muscle contributions. Extreme smoothing (i.e., LP cut-off 0.5 Hz) enhanced the contrast between active and inactive muscles but had an unpredictable effect on the ISS. The results presented here constitute a further step towards a thoughtful EMG pre-processing for the extraction of muscle synergies.