Detection of a sleep disorder predicting Parkinson's disease

Idiopathic rapid eye-movement (REM) sleep behavior disorder (iRBD) has been found to be a strong early predictor for later development into Parkinson's disease (PD). iRBD is diagnosed by polysomnography but the manual evaluation is laborious, why the aims of this study are to develop supportive methods for detecting iRBD from electroencephalo-graphic (EEG) signals recorded during REM sleep. This method classified subjects from their EEG similarity with the two classes iRBD patients and control subjects. The feature sets used for classifying subjects were based on the relative powers of the EEG signals in different frequency bands. The classification was based on the fast and classical K-means and Bayesian classifiers. With a subject-specific re-scaling of the feature set and the use of a Bayesian classifier the performance reached 90% in both sensitivity and specificity. For the purpose of reducing the feature count, the features were evaluated with the statistical Smith-Satterthwaite test and by using sequential forward selection a well-performing feature subset was found which contained only five features, while attaining a sensitivity and a specificity of both 80 %.