Statistical Properties of Upper Limb Accelerometer Signals of Patients with Amyotrophic Lateral Sclerosis

Statistical properties of accelerometer (ACC) are useful to determine the appropriate tool to obtain biomedical signal features for each specific aim. It may be applied to evaluate human movement in order to detect and monitor neuromuscular diseases such as amyotrophic lateral sclerosis (ALS). This study aimed to use techniques to determine the degree of stationarity and linearity of ACC to analyze and compare upper limb (UL) in healthy subjects (HS) and ALS. Our dataset contains 10 being HS (age $48.4\pm 4.25$ years) and seven ALS people (age $59.86\pm 16.32$ years) who underwent motion analysis from 16 ACC sensors sampled at 148 Hz for 25 seconds, which were positioned over the UL. In the pre-processing stage, we removed the first five seconds, a low pass filter, data normalization, and Euclidean norm of the 3-axis ACC data. Subsequently, we measured the degree of stationarity (mean, variance, and Kwiatkowski-Phillips-Schmidt-Shin test) and linearity (standard deviation, Brock, Dechert & Scheinkman test, and nonlinear autoregressive exogenous test). Proved by experimental results, ACC data of UL segments evaluated showed nonlinear and nonstationary behavior, mainly in the ALS patients. Our findings provide the first applications of statistical methods to guide ACC analysis from the view of nonlinear and nonstationary properties of ACC signals to extract signal features to guide the therapeutic planning of patients and a better control strategy for assistive technologies.