PPG-Based Sleep Staging Using SleepPPGNet: Extension to Wearables, Improvements, Limitations.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Traditional polysomnography for sleep disorder diagnosis is resource-intensive and inconvenient for patients.
• Emerging deep learning models offer potential for more accessible sleep staging.
• Photoplethysmography (PPG) is a non-invasive technique for measuring blood volume changes.

Purpose of the Study:

• To evaluate the performance of the SleepPPGNet deep learning model for sleep staging using wrist-worn PPG devices.
• To investigate the impact of incorporating activity counts as additional input to the model.
• To assess the model's efficacy in patients with cardiac arrhythmias.

Main Methods:

• Applied a pre-trained deep learning model (SleepPPGNet) to PPG data collected from wrist-worn devices.
• Augmented the model architecture to include activity count data.
• Compared model performance on datasets with normal cardiac rhythms versus cardiac arrhythmias.

Main Results:

• The model achieved 78% accuracy and a Cohen's kappa of 0.68 using PPG data from wrist-worn devices.
• Incorporating activity counts improved accuracy to 80.0% and Cohen's kappa to 0.69.
• A significant accuracy drop of 10% was observed in patients with cardiac arrhythmia.

Conclusions:

• Wrist-worn PPG devices coupled with deep learning show potential for remote sleep staging.
• Activity counts can enhance the accuracy of PPG-based sleep staging models.
• Further model development is needed to address limitations in patients with cardiac arrhythmias.