Classification of the stages of Parkinson’s disease using novel higher-order statistical features of EEG signals

Abstract

Parkinson’s disease (PD) is the second most common neurodegenerative disease after Alzheimer’s disease. This progressive disorder is mostly known by its signature symptoms of bradykinesia and rest tremor. The progression of PD is divided into five stages by the Hoehn and Yahr (H&Y) rating scale. The diagnosis of PD and its progress is of utmost importance in providing an effective treatment plan and to manage the disease and maintain quality of life. The goal of this study is to introduce several new higher-order features and use them with a combination of existing lower and higher-order statistical features of EEG signals to diagnose the stages of PD. In this paper, we use the total of 30 higher-order and six lower-order (first and second-order) statistical features extracted from EEG signals of 20 PD patients, along with 20 age-matched healthy control subjects, recorded during rest state, to perform the classification. We use a range of classification methods, and although other methods may have higher overall accuracy, we see that due to our unbalanced dataset, RUS Boosted trees ensemble has the highest sensitivity for this application (highest performance for the late PD class). It is also shown that the accuracy of the classification is improved significantly by using the newly developed features.