A multi-task learning model using RR intervals and respiratory effort to assess sleep disordered breathing.

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence in Healthcare

Background:

• Sleep-disordered breathing (SDB) necessitates accessible diagnostic and monitoring tools.
• Current methods require specialized equipment, limiting widespread use.
• Exploiting cardiac and respiratory signals offers a promising avenue for unobtrusive SDB assessment.

Purpose of the Study:

• To introduce a novel multi-task model for simultaneous sleep-wake classification and SDB event detection.
• To automatically estimate the apnea-hypopnea index (AHI) for SDB severity assessment.
• To evaluate the model's performance using readily available physiological signals.

Main Methods:

• A multi-task model combining convolutional and recurrent neural networks was developed.
• The model was trained using RR intervals from electrocardiograms and respiratory effort signals.
• Performance was validated against polysomnography (PSG) recordings from 198 patients.

Main Results:

• The model achieved a Cohen's kappa of 0.70 for sleep-wake classification and a Spearman's R of 0.830 for total sleep time estimation.
• An R of 0.891 was obtained for estimated versus reference AHI.
• The multi-task model outperformed a single-task model, particularly for patients with low sleep efficiency (R=0.861 vs. R=0.746).

Conclusions:

• The multi-task model automatically and accurately estimates AHI and SDB severity using RR intervals and respiratory effort.
• This technology holds potential for improved SDB screening with unobtrusive sensors.
• The method is effective even for individuals with low sleep efficiency, without requiring additional sleep-monitoring sensors.