Activity pattern detection in electroneurographic and electromyogram signals through a heteroscedastic change-point method

In this work, we propose a heteroscedastic method in the detection of activity patterns of electroneurographic and electromyogram signals involved in rhythmic activities of nerves and muscles, respectively. The electric behavior observed in such signals is characterized by phases of activity and silence. The beginning and the length of electrically active and electrically silent phases in a signal allow us to quantitatively analyze the changes and the effects on a rhythmic activity produced by experimental changes. In order to distinguish between these two phases, signals are assumed to be a sample of a time-dependent, normally distributed random variable with non-constant variance, and that the determination of the variance at each point allows us to determine in which phase is the signal. The parameters of the model are determined by means of an iterative process which maximizes the log-likelihood under the proposed model. Moreover, we apply our method to the determination of the activity phases and silence phases in sequences of experimental and synthetic electroneurographic and electromyogram signals. The results obtained with synthetic data show that the method performs well in the determination of these activity patterns. Finally, the study of particular signals simulated under a generalized autoregressive conditional heteroscedasticity model suggests the robustness of the method with respect to the assumption of independence.