DistillSleep: Real-Time, On-Device, Interpretable Sleep Staging from Single-Channel EEG

Published in SLEEP, 2025

Polysomnography is the gold standard for sleep staging, but its high cost, laboratory equipment, and lengthy manual scoring limit patient access. DistillSleep replaces the typical 12-20-sensor setup with a single-channel EEG and performs inference in <10 ms per epoch on a Raspberry Pi, Jetson orin nano, or Coral dev board. Trained and tested on >10,000 overnight recordings from six independent cohorts, it matches expert accuracy (Macro-F1 up to 80%) and supplies clinicians with frequency-band saliency, inter-epoch context, and well-calibrated confidence scores. By combining interpretability, millisecond-level latency, and an open-source code release, DistillSleep supports point-of-care diagnostics, same-night CPAP titration, and large-scale home monitoring, substantially broadening the reach of sleep medicine.

Recommended citation: Park, K., Hong, J., Lee, W., Shin, H.W. and Kim, H.S., 2025. DistillSleep: Real-Time, On-Device, Interpretable Sleep Staging from Single-Channel EEG. SLEEPJ, p.zsaf240.
Download Paper | Download Bibtex