Contactless real-time detection of REM eye movements with a bedside infrared camera, recorded in parallel with forehead EEG: a feasibility study
Michael Paul Coder
Zenodo (CERN European Organization for Nuclear Research) November 3, 2026 preprint DOI: 10.5281/zenodo.23912374 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Case study Case report |
|---|---|
| Sample size | 1 |
| Population | The author's own sleep, recorded over two nights at home |
| Duration | Two nights |
| Measures | Two-channel forehead EEG (CGX Patch), ezscore ez6 automatic staging, kappa agreement, correlation (r) |
| Key findings | On one night the camera's eye-movement counter tracked EEG-defined REM moderately well (r = 0.75 in 10-minute windows, kappa 0.41 epoch by epoch), but on a second night agreement was weak (r = 0.20, kappa −0.03). Of 13 cue triggers fired during sleep, 10 fell inside EEG REM periods, and the device triggered in 2 of 6 REM periods (exact 95% CI 4% to 78%). |
Abstract
Most devices that detect REM sleep at home are worn on the face or head. This study tests the lucidcode INSPEC, a bedside infrared camera that detects the rapid eye movements of REM sleep in real time and can cue the sleeper to induce lucid dreams. I recorded two nights of my own sleep at home with the camera and a two-channel forehead EEG (CGX Patch) running at the same time. The EEG was staged automatically by ezscore ez6. On night 1, the camera's eye-movement counter correlated with REM in the EEG (r = 0.75 in 10-minute windows) and agreed moderately with ez6 epoch by epoch at the app's REM level (kappa 0.41). From sleep onset, 10 of the 11 REM bins in the app's hypnogram contained EEG REM (kappa 0.71), and they covered 3 of the night's 4 REM periods. On night 2, where ez6 found no REM in the first five hours, the counter followed REM only weakly (r = 0.20, kappa −0.03), and the EEG showed eye-movement transients through the main REM period that the camera did not register. On both nights, the counter's rises in REM coincided with eye-movement-like transients in the forehead EEG more often than chance. Across both nights, the device fired 13 cue triggers during sleep, 10 of them inside an ez6 REM period, and triggered in 2 of the 6 REM periods (exact 95% CI 4% to 78%). The data, logs and snapshots are openly available, and the analysis code is available on request.