Chaotic biomedical time signal analysis via wavelet scattering transform

Abstract

Non-linear techniques are useful to analyze and understand complex systems, especially biological systems. This paper presents a novel procedure to study the dynamics of biomedical signals. The procedure uses features of a wavelet scattering transform to classify signal segments as either chaotic or non-chaotic. To this end, in the paper is also definition of a chaos measure. Classification is made using a model trained on a dataset consisting of signals generated by systems with known characteristics. Using an example PPG signal, this paper demonstrates the usefulness of the wavelet scattering transform for the analysis of biomedical signals, and shows the importance of correctly preparing the training set. Next, the trained models were used for the classification of other biomedical signals. The eye movement signal and the human joint angles data during gait were selected for the tests. The present work is an extended version of the conference paper (Szczęsna et al., 2022). The aim of this extended version is to demonstrate the generality and usefulness of the proposed method of biomedical time series analysis.