EEG-Based Detection of REM Sleep Behaviour Disorder: Towards a Stage-Agnostic Approach

Several recent studies have demonstrated a causal relationship between sleep disorders and neurodegenerative diseases. Remarkably, Idiopathic REM Sleep Behavior Disorder (iRBD) exhibits strong correlation with alpha-synucleinopathies such as Parkinson’s Disease, Dementia with Lewy Bodies, and Multiple System Atrophy. Moreover, recent studies have highlighted the potential role of Slow Wave Sleep (SWS) disorders as a predictive marker of neurodegenerative processes including Parkinson’s Disease and Alzheimer’s dementia. Consequently, the analysis of sleep disorders helps evaluate the risk of developing such conditions. At present, the gold standard for sleep disorder identification is Polysomnography. This method is expensive, time-consuming, and not suitable for widespread screening. Therefore, methods that exploit Artificial Intelligence to enable an automatized sleep analysis, possibly using a reduced electrode setup, could represent a significant step forward. The availability of a low-cost, automatic RBD detection system could be exploited for mass-screening programs and early neurodegeneration risk evaluation. This work proposes an automatized strategy for RBD identification. It aims at simplifying the overall workflow, moving towards a "sleep staging-less" approach based on single-channel Electroencephalography (EEG). EEG signals recorded during both REM and SWS sleep stages are employed. Supervised Machine Learning models, trained and tested in a Leave-One-Out cross-validation framework, achieve accuracy up to 91% and sensitivity up to 94% (RBD class). This consolidates the REM and SWS micro-structures potential as RBD markers and paves the way for a low-cost, stage-agnostic, EEG-based identification of RBD and early evaluation of neurodegeneration risk.