Time frequency distribution and deep neural network for automated identification of insomnia using single channel EEG-signals

Abstract

It is essential to have enough sleep for a healthy life; otherwise, it may lead to sleep disorders such as apnea, narcolepsy, insomnia, and periodic leg movements. A polysomnogram (PSG) is typically used to analyze sleep and identify different sleep disorders. This work proposes a novel convolutional neural network (CNN)-based technique for insomnia detection using single-channel electroencephalogram (EEG) signals instead of complex PSG. Morlet wavelet-based continuous wavelet transforms and smoothed pseudo-Wigner-Ville distribution (SPWVD) are explored in the proposed method to obtain scalograms of EEG signals of duration 1s along with convolutional layers for features extraction and image classification. The Morlet transform is found to be a better time-frequency distribution. We have developed Morlet wavelet-based CNN (MWTCNNet) for the classification of healthy and insomniac patients using cyclic alternating pattern (CAP) and sleep disorder research centre (SDRC) databases with C4-A1 single-channel EEG derivation. We have used multiple cohorts/settings of the CAP and SDRC databases to analyse the performance of proposed model. The proposed MWTCNNet achieved an accuracy, sensitivity, and specificity of 98.9%, 99.03%, and 98.66%, respectively, using the CAP database, and 99.03%, 99.20%, and 98.87%, respectively, with the SDRC database. Our proposed model performs better than existing state-of-the-art models and can be tested on a vast, diverse database before being installed for clinical application.