Automated detection of sleep stages using energy-localized orthogonal wavelet filter banks

Sleep is an integral part of human life which provides the body with much-needed rest which facilitates recovery and promotes health. Sleep disorders, however, lead to a reduced quality of sleep and as a result, affect the standard of human life. It is important to classify sleep stages in order to...

Full description

Saved in:
Bibliographic Details
Main Authors: Sharma, Manish, Patel, Sohamkumar, Choudhary, Siddhant, Acharya, U. Rajendra
Format: Article
Published: Springer Heidelberg 2020
Subjects:
Online Access:http://eprints.um.edu.my/37252/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Sleep is an integral part of human life which provides the body with much-needed rest which facilitates recovery and promotes health. Sleep disorders, however, lead to a reduced quality of sleep and as a result, affect the standard of human life. It is important to classify sleep stages in order to detect sleep disorders. Electroencephalogram (EEG) signals are obtained from patients under observation. But, classifying these EEG signals into various sleep stages is an arduous task. It becomes more difficult when one tries to classify EEG signals visually. Even sleep specialists struggle to classify the EEG signals into different sleep stages by visual inspection. Several approaches have been adopted by scientists across the world to mitigate these errors by using EEG and polysomnogram signals. In this paper, an automated method has been proposed for scoring various sleep stages employing EEG signals. We have employed a two-band energy-localized filter in the time-frequency domain, which decomposed six sub-bands using five-level wavelet decomposition. Subsequently, we compute discriminatory features namely fuzzy entropy and log energy from the decomposed coefficients. The extracted features are fed to various supervised machine learning classifiers. Our proposed approach yielded an accuracy of 91.5% and 88.5% for six-class classification task using small and large datasets, respectively.