Classification of Biomedical Signals using the Dynamics of the False Nearest Neighbours (DFNN) Algorithm

Abstract

Accurate and efficient analysis of biomedical signals can be facilitated by proper identification based on their dominant dynamic characteristics (deterministic, chaotic or random). Specific analysis techniques exist to study the dynamics of each of these three categories of signals. However, comprehensive and yet adequately simple screening tools to appropriately classify an unknown incoming biomedical signal are still lacking. This study is aimed at presenting an efficient and simple method to classify model signals into the three categories of deterministic, random or chaotic, using the dynamics of the False Nearest Neighbours (DFNN) algorithm, and then to utilize the developed classification method to assess how some specific biomedical signals position with respect to these categories. Model deterministic, chaotic and random signals were subjected to state space decomposition, followed by specific wavelet and statistical analysis aiming at deriving a comprehensive plot representing the three signal categories in clearly defined clusters. Previously recorded electrogastrographic (EGG) signals subjected to controlled, surgically-invoked uncoupling were submitted to the proposed algorithm, and were classified as chaotic. Although computationally intensive, the developed methodology was found to be extremely useful and convenient to use.