A novel solution of deep learning for sleep apnea detection: enhancement of SC and elimination of GVICS

• Published in: Multimedia tools and applications Vol. 81; no. 27; pp. 38569 - 38592
• Main Authors: Limbu, Narayan, Alsadoon, Abeer, Prasad, P.W.C., Abdullah, Salma, Rashid, Tarik A., Alsadoon, Omar Hisham, Jerew, Oday D., Alrubaie, Ahmad
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.11.2022; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial neural networks; Classification; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Deep learning; Machine learning; Multimedia Information Systems; Neural networks; Segmentation; Sleep apnea; Special Purpose and Application-Based Systems; Time dependence; Gradient vanishing; Internal covariate shift; ECG signal segmentation; Sleep apnea; Convolutional neural network
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-022-13142-1

Convolutional neural network (CNN) classification has not achieved a medically-satisfied level of accuracy in sleep apnea detection due to the negative effect of the ECG data segmentation process, Gradient vanishing issues, and internal Covariate shift. This research aims to enhance segmentation, classification accuracy, and eliminate gradient vanishing and internal covariate shift to increase the sleep apnea detection classification accuracy. The proposed system consists of a novel enhancement of segmentation and classification technique (Enhancement of SC) and elimination of gradient vanishing and internal covariate shift (Elimination of GVICS). The enhancement of SC considers the time-dependent nature of apnea events during the classification process and the elimination of GVICS reduces the instability of the neural network due to inactive neurons and improves classification accuracy by removing large input distribution of data during the training phases . The results show that the proposed system achieves better classification performance in four different datasets tested and given accuracy 98.9% against the current accuracy of 94.5% processing time per sample 3 milliseconds against the current processing time 5 milliseconds. Besides, the model has achieved its optimal accuracy in fewer epochs 8 against the current optimal accuracy time of 11 epochs. The proposed solution has improved the accuracy of sleep apnea classification and reduced time to achieve optimal accuracy of the neural network by using the ESCEGCs algorithm; the study has solved the issues of gradient vanishing and internal covariate shift and enhances the overall classification accuracy.