Self-applied single-channel mastoid ExG for fully automated detection of REM sleep behaviour disorder

Isolated REM sleep behaviour disorder (RBD) is the strongest prodromal marker of Parkinson's disease (PD) and dementia with Lewy bodies, yet diagnosis requires video-polysomnography with assisted montage and expert scoring and does not scale to screening or trial enrichment. We developed a fully automated, self-applied system that detects RBD from a pair of electrodes placed behind the ears, with no manual scoring at any stage. A novel bipolar mastoid ExG derivation enables both sleep staging and quantification of REM sleep without atonia (RWA). A fine-tuned deep-learning 1-channel model staged sleep at a Cohen's kappa of 0.65 in PD, iRBD and controls ({kappa} = 0.73 for the 2-channel model). Automated mastoid RWA correlated strongly with expert chin SINBAR scoring (r = 0.82). In self-applied home recordings from 76 participants, the 1-channel system detected RBD with an AUC of 0.95 (sensitivity 94%, specificity 86%), reproduced on in-lab polysomnographies (AUC 0.93, n = 378). In RBD, between-night RWA variability warrants repeated nights for prognostic monitoring. The system provides a scalable tool for RBD detection and a continuous RWA measure for longitudinal studies of neurodegeneration.