Efficient Deep Learning Based Hybrid Model to Detect Obstructive Sleep Apnea.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) is a growing concern with limited awareness.
• Current diagnostic standard, polysomnography, is complex, expensive, and inaccessible to many.
• Existing machine learning methods using single-lead signals lack accuracy and reliability.

Purpose of the Study:

• To develop an accurate and accessible method for obstructive sleep apnea detection.
• To evaluate the efficacy of deep learning models, specifically MobileNet V1 and its convergence with recurrent neural networks (LSTM, GRU), for OSA diagnosis.

Main Methods:

• Implementation of MobileNet V1 architecture for analyzing single-lead physiological signals.
• Integration of MobileNet V1 with Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks.
• Validation using the PhysioNet Apnea-Electrocardiogram database, comprising authentic patient data.

Main Results:

• MobileNet V1 achieved an accuracy of 89.5% in detecting obstructive sleep apnea.
• The convergence of MobileNet V1 with LSTM and GRU demonstrated superior performance, reaching 90% and 90.29% accuracy, respectively.
• The proposed deep learning models significantly outperformed existing state-of-the-art methods.

Conclusions:

• The developed deep learning paradigms, particularly MobileNet V1 combined with recurrent neural networks, offer a highly accurate and reliable solution for obstructive sleep apnea detection.
• A prototype wearable device was designed for real-time ECG monitoring and secure cloud transmission, demonstrating practical clinical application.
• This approach presents a promising, accessible alternative to polysomnography for widespread obstructive sleep apnea screening and diagnosis.