Developing a deep learning model for sleep stage prediction in obstructive sleep apnea cohort using 60 GHz...

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Sleep quality significantly impacts overall health.
• Early detection of sleep disorders is crucial for timely intervention.
• Polysomnography (PSG) is the gold standard but can be invasive and costly.

Purpose of the Study:

• To evaluate the efficacy of radar technology for non-invasive sleep stage classification.
• To develop and validate an Attention Bi-LSTM model for predicting sleep stages using FMCW radar.
• To assess the impact of using multiple radar sensors and OSA severity on classification performance.

Main Methods:

• Utilized 60 GHz FMCW radar to monitor 78 participants from an obstructive sleep apnea (OSA) cohort.
• Collected simultaneous PSG and radar data during overnight sleep.
• Applied an Attention Bi-LSTM model for classifying wakefulness, REM, and NREM sleep stages.

Main Results:

• Achieved an overall accuracy of 85.2% and a Cohen's kappa of 0.746 in sleep stage classification.
• Models using data from two radars (Radar 1 and Radar 2) outperformed those using only Radar 1.
• Radar 2 data improved classification accuracy, especially with increasing OSA severity.

Conclusions:

• Radar technology shows significant potential as a non-invasive and cost-effective screening tool for sleep stage classification.
• Multi-radar systems enhance the accuracy of sleep stage prediction.
• This approach could facilitate earlier detection and management of sleep disorders.