Automatic separation of awakening from sleep epochs based on bispectrum analysis of electroencephalogram signals

Abstract

Background: Accurate separation of awakening from sleep increases the accuracy of detecting sleep stages and determining sleep efficiency index that is important for medical diagnosis. In this study, 3 new bispectrum-based features were extracted, and combination of them with Bi-Phase correlation was used to detect awakening from sleep. Methods: A gray scale image was made of electroencephalogram bispectrum amounts and converted to binary image with Otsu’s thresholding. Then, 3 features were extracted from it: total numbers of one bits, ratio of one bit in the above of the secondary diagonal to the down of it, and entropy. By combining these features with the average Bi-Phase correlation, an efficient and new way to distinguish the awakening of sleep was proposed. Findings: The accuracy, specificity, and sensitivity were calculated. Awakening intervals could be distinguished from sleep using Bi-Phase correlation feature by the accuracy of 77.53, and using ratio of one bit in the above of the secondary diagonal to the down of it, by the accuracy of 88.12. Finally, combining 3 mentioned features with Bi-Phase correlation gave the ability to separate awakening from sleeping with the accuracy of 92.42, sensitivity of 91.82, and specificity of 93.10. Conclusion: New features have this capability to use in sleep staging. The proposed method in awakening and sleep discrimination, by combining bispectrum image-based features with Bi-Phase correlation, is better than other existing approaches because of high accuracy and low calculation complexity. © 2017, Isfahan University of Medical Sciences(IUMS). All rights reserved.