Integrating physiological signals for enhanced sleep apnea diagnosis with SleepNet

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Sleep apnea is a common respiratory disorder linked to serious health issues.
• Current diagnostic methods like polysomnography are effective but costly and inconvenient.
• There is a need for accessible and accurate sleep apnea detection methods.

Purpose of the Study:

• To introduce SleepNet, a novel multimodal deep learning framework for precise sleep apnea detection.
• To evaluate the performance of SleepNet using electrocardiogram (ECG) data alone and in combination with other physiological signals.
• To compare the efficacy of the multimodal approach against unimodal methods.

Main Methods:

• Developed a deep learning model fusing one-dimensional convolutional neural networks (1D-CNN) and bidirectional gated recurrent units (Bi-GRU).
• Analyzed single-lead ECG recordings for sleep apnea detection.
• Integrated additional physiological signals (nasal airflow, abdominal respiratory effort) to assess performance enhancement.

Main Results:

• The SleepNet model achieved 95.08% accuracy using only ECG data.
• Incorporating nasal airflow and abdominal respiratory effort modestly increased accuracy to 95.19%.
• The multimodal approach demonstrated superior sensitivity (96.12%) and specificity (93.45%) compared to unimodal methods.

Conclusions:

• SleepNet represents a significant advancement in sleep apnea diagnostic efficacy.
• Integrating multiple data streams via deep learning holds transformative potential for sleep apnea detection.
• The study provides a strong foundation for future research in AI-driven respiratory diagnostics.