“…Also, EMG is prone to background noise and especially spurious spikes from intramuscular EMG, because of a few surrounding muscle fibers that react to the presence of the needles. To effectively account for it, several methods were compared (integrated profile, sample entropy, Bayesian changepoint analysis) and the generalized likelihood ratio (GLR) method [17] was the most effective to abstract from the noise, as it allowed us to model the signals. It was computed on a 100ms sliding window as recommended by the original paper [32], and with an exponential probability density function to represent the TKEO-transformed raw EMG [28].…”