Automated classification of cyclic alternating pattern sleep phases in healthy and sleep-disordered subjects using convolutional neural network

• Published in: Computers in biology and medicine Vol. 146; p. 105594
• Main Authors: Murarka, Shruti, Wadichar, Aditya, Bhurane, Ankit, Sharma, Manish, Acharya, U. Rajendra
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.07.2022; Elsevier Limited
• Subjects: Artificial neural networks; Automation; Classification; Convolutional neural network (CNN); Cyclic alternating patterns (CAP); Datasets; Deep learning; Deep Learning, Electroencephalogram (EEG); EEG; Electroencephalography; Epilepsy; Frontal lobe; Human error; Insomnia; Internal Medicine; Machine learning; Medical personnel; Monitoring; Narcolepsy; Neural networks; Other; Phase A and phase B Detection; Phases; REM sleep; Respiration; Sleep apnea; Sleep disorders; Telemedicine; Well being; Sleep disorders; Cyclic alternating patterns (CAP); Phase A and phase B Detection; Convolutional neural network (CNN); Deep Learning, Electroencephalogram (EEG)
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2022.105594

Sleep contributes to more than a third of a person's life, making sleep monitoring essential for overall well-being. Cyclic alternating patterns (CAP) are crucial in monitoring sleep quality and associated illnesses such as insomnia, nocturnal frontal lobe epilepsy (NFLE), narcolepsy, etc. However, traditionally medical specialists practice manual division techniques of CAP phases which are sensitive to human weariness and inaccuracies. This might result in a false sleep stage diagnosis. This study proposes an automated approach using a deep learning model based on a 1-dimensional convolutional neural network for classifying CAP phases (A and B). The proposed model uses single-channel standardized electroencephalogram (EEG) recordings provided by the CAP sleep database. The model was created with the help of healthy participants and patients suffering from five distinct sleep disorders, which includes narcolepsy, rapid eye movement behaviour disorder (RBD), periodic leg movement disorder (PLM), NFLE, and insomnia. The developed model has achieved the highest automated classification accuracy of 78.84% for the healthy dataset and 82.21%, 79.48%, 78.73%, 76.68%, and 70.88% for narcolepsy, RBD, PLM, NFLE, and insomnia subjects, respectively in categorizing phases A and B. The proposed approach can help medical professionals monitor sleep and examine a person's brain stability. •A 1-D CNN is used for CAP phase classification of healthy and disordered patients.•Proposed model requires no pre-processing & uses only one EEG signal.•Obtained maximum accuracy of 78.84% for CAP classification of healthy subjects.•Model obtained an accuracy of 82.21% for narcolepsy.•RBD, PLM, NFLE & INS have reported accuracy of 79%, 78%, 76% & 70%, respectively.