Modeling Spatial Synchronization of Sleep for Detection of Idiopathic REM Sleep Behavior Disorder

Idiopathic REM sleep behavioral disorder (iRBD) is a parasomnia and is considered a prodromal stage of Parkinson's Disease (PD). The golden standard of diagnosis is through visual analysis of a polysomnography (PSG) recording, which is resource-intensive. There is a need for better markers for severity evaluation and sub-typing. We developed a method to investigate and quantify the synchronization of brain activity as measured by electroencephalogram (EEG) during sleep. Using hypnodensities, an automatically derived sleep stage probability distribution, we evaluated sleep synchronization across pairs of electrodes using a single EEG-based sleep staging model. Sleep stage synchronization was evaluated using various epoch sizes (1, 3, 5, 15, 30 seconds) using epoch-wise correlation and dynamic time warping (DTW) to measure the similarity between hypnodensities. Furthermore, we also evaluated EEG coherence in frequency bands of interest. Using all features, gradient-boosted decision trees achieved an AUC of 0.82. Impurity-based feature importance analysis revealed that the model relied mainly on increased coherence of gamma activity during REM sleep but also a lower DTW distance of N1 probability in iRBD.Clinical Relevance-Our results highlight that brain activity synchronization during sleep is abnormal in iRBD, as measured both by EEG coherence and hypnodensity similarity. These could be used as additional markers using our approach in EEG-based diagnostic models of RBD or prognostic models of phenoconversion to PD.