A Deep Learning Framework for Sleep Apnea Detection

Sleep apnea is a common sleep problem that has a major impact on society. The gold standard, polysomnography, is difficult to obtain, can be unpleasant, and requires a skilled technician to score. Automatic scoring systems based on fewer sensors and automatic categorization algorithms have been suggested and implemented by numerous researchers to overcome these challenges. Database accessibility, newly discovered approaches, the capability of developing machine-made features, and increased computing power allow the algorithms to achieve greater performance than the shallow classifiers, all of which contribute to deep learning’s rising popularity. As a result, there is a lot of focus on deep learning right now in the field of sleep apnea. In this research, a deep learning framework using a convolutional neural network is introduced for an automated, highly accurate detection of obstructive sleep apnea (OSA). To diagnose obstructive sleep apnea, the suggested work creates a system that analyzes single-lead electrocardiography signals from individuals. The outcomes demonstrate the efficacy of the proposed strategy in comparison with the state-of-the-art alternatives. The current technique can identify OSA without the use of any additional algorithms for feature extraction or classification. The suggested network can be used for supervised feature learning as well as feature classification. Although it is computationally expensive, the approach has been shown to outperform all other published literature by a margin of more than 9%. The approach also exhibits strong resistance to signal contamination by noise. Existing approaches cannot compete with the present method, even when a very low signal-to-noise ratio is used.