Improving classification system to detect cyclic alternation pattern using sleep EEG
Abstract
Cyclic alternative patterns (CAP) became an important tool to diagnose sleep disorders. This study aims to have a better understanding of CAP, as the clinical applications remain limited as CAP analysis is a time-consuming activity; the goal of this thesis research is to improve the automatic classification system used to detect CAP in sleep. To determine the highest accuracy at detecting CAPs, MATLAB's classification learner app trains and tests the extracted features’ characteristics against different classifiers. The combination of these selected characteristics and proposed classifiers can effectively measure the degree of change in brain state between non-CAPs and Caps. Time domain characteristics are computed from the signal amplitude values like Tsallis entropy, Renyi entropy, and Shannon entropy. In this study, the Hilbert-Huang Transform (HHT) and FFT characteristics provide information about the frequency characteristics of CAP phases. A cross-validation procedure is used in MATLAB model validation to estimate the performance of the classifier. CAP detection in healthy patients was more effective using time-based entropy features and KNN classifiers than frequency-based ones. With higher accuracy up to 90%, time-based entropy features performed better for insomnia patients.