EMG factorization during walking: does digital filtering influence the accuracy in the evaluation of the muscle synergy number?

This paper aims at verifying if the number of muscle synergies is influenced by the filtering process. Firstly, EMG activity of 12 muscles of the lower limb of a healthy subject was recorded during walking trial. Then, the Wavelet-Independent Component Analysis was used for the separation of the EMG, considered as reference signal, and the noise source from the raw experimental data. To test the effect of different order and cut-off frequencies of the high-pass and low-pass filtering on the numbers of muscle synergies, a Monte Carlo simulation was performed on the EMG signals. In particular, a dataset of 10 5 raw EMG signals were generated by adding different noise signals to the reference signal. The reference dataset was pre-processed with different combination of order and cut-off frequency of low-pass filter for the envelope extraction. The simulated dataset, instead, was pre-processed with different combinations of high and low pass filtering. A non-negative matrix factorization was performed to extract the muscle synergies on both the reference dataset and the simulated ones. For each filtering combination, the percentage of the Number of Synergies NoS within each simulated dataset was computed and compared with the NoS obtained from the reference dataset. The reference dataset produced a number of synergies equal to 4 for all the combinations of the tested low-pass filters, proving that, for the EMG signal without artifact, the number of muscle synergies is robust with respect to different low-pass filtering parameters. Conversely, the simulated dataset generated different numbers of synergies in different filtering conditions, ranging from 3 to 5. However, some particular combinations of filtering parameters produced no variation on Nos. Our results demonstrate the importance of choosing the most appropriate filtering parameters, as it resulted that low order and cut-off frequencies lead to an underestimation of the number of synergies, while high orders and cut-off frequencies cause an overestimation of the number of muscle synergies. In this vein, the present paper offers guidelines for the choice of the best filtering condition in muscle synergy evaluation