Comparison of automated methods forREM sleep without atonia detection

Objectives/Introduction:REM sleep without atonia (RSWA) detection is a prerequisite for REM sleep behaviour disorder (RBD)diagnosis. Since current visual methods for RSWA detection are sub-jective and tedious, several automated methods have been proposed.This study aims to compare their accuracies in identifying RSWA and to analyse the influence of respiration and arousal‐related movements in such accuracies. Methods:In a cohort including 27 healthy control subjects (C), 25 Parkinson's disease (PD) patients without RBD (PD‐RBD), 29 PD patients with RBD (PD+RBD), 29 idiopathic RBD patients and 36 patients with periodic limb movement disorder (PLMD), the following indices were calculated: The REM atonia index (RAI), the supra‐threshold‐REM‐activity metric (STREAM), the Frandsen index (FRI),the short/long muscle activity indices (sMAI/lMAI) and the Kempfner index (KEI). The indices were calculated in various cases: 1) considering all muscle activities; 2) excluding the ones related to arousals; 3)excluding the ones during apnea events; 4) excluding the ones before and after apnea events; 5) combining cases 2 and 3; and 6)combining cases 2 and 4. In each case, each index was used to train and test a logistic regression model with 10‐fold cross‐validationscheme to calculate accuracies in the followingcomparisons:1(C, PD-RBD, PLMD) vs (PD+RBD, RBD);2C vs RBD;3PLMD vs RBD;4C vs PD-RBD;5C vs PLMD;6PD-RBD vs PD+RBD; and7C vs PLMD vs RBD.Kruskal‐Wallis tests were used to assess whether the accuracies for each index and comparison were varying significantly across the cases.Results:The indices showed varying performances across cases and comparisons, making it impossible to identify one index as significantly better than the others. In all comparisons and cases, the average test accuracy of each index was lower than 80%. Moreover, apnea and arousal‐related movements did not influence significantly the performance of the algorithms in group distinctions.Conclusions:None of the automated method can be elected asthe optimal one for RSWA detection and the accuracies suggest the need of improvements. Automated methods seem robust towards respiration and arousal‐related movements.