Large-Scale Validation of a Dual Cross-Attention Network for Automated Sleep Staging Using Wearable...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Polysomnography (PSG) is the gold standard for sleep staging but is invasive and impractical for home use.
• Photoplethysmography (PPG) offers a non-invasive, wearable alternative for sleep monitoring, but faces challenges with signal noise and variability.
• Accurate sleep staging is crucial for diagnosing sleep disorders and guiding treatment.

Purpose of the Study:

• To develop and validate a deep learning framework (DCA-Sleep) for accurate sleep staging using single-channel PPG.
• To address data scarcity challenges through a cross-modality transfer learning strategy.
• To evaluate the performance of DCA-Sleep on a large, diverse cohort of subjects.

Main Methods:

• Developed DCA-Sleep, a deep learning model featuring a Dual Cross-Attention (DCA) mechanism for analyzing temporal dependencies in PPG signals.
• Implemented a cross-modality transfer learning approach, pre-training on six electrocardiogram (ECG) datasets.
• Validated the model on a combined dataset of 9738 subjects across nine public datasets, including MESA and CFS.

Main Results:

• DCA-Sleep achieved robust performance, with an average F1-score of 0.731 and Cohen's Kappa of 0.652 on the MESA dataset.
• The model significantly outperformed existing state-of-the-art methods.
• Demonstrated high sensitivity in identifying critical sleep stages like Wake and Deep Sleep.

Conclusions:

• DCA-Sleep provides a reliable, non-invasive method for long-term sleep monitoring using PPG.
• The study validates the scalability and clinical utility of PPG-based sleep staging tools.
• This technology can enhance clinical screening and management of sleep disorders.