A CACDSS for automatic detection of Parkinson's disease using EEG signals

The advancement from new-born to old-age results in physical and psychological growth of human-being. The number of neurons also begins to die or become impaired with advancing age. These dying or impaired neurons result in declination for the generation of dopamine which is the prime reason for Parkinson's disease (PD). Though PD is incurable, early detection, proper diagnosis, and timely medication may help PD patients to perform their routine tasks. Electroencephalogram (EEG) signals are one such medium for automatic detection of PD. But the nature of EEG signals is complex, non-linear, and non-stationary making its analysis difficult. Therefore, this paper presents a computer-aided clinical decision support system (CACDSS). The CACDSS consists of automatic signal analysis and classification techniques combining automated variational mode decomposition (AOVMD) and automated extreme learning machine (AOELM) classifier. AOVMD selects the decomposition parameters adaptively using the arithmetic optimization algorithm by extracting representative modes and minimizing reconstruction error. The modes are further used to compute features which are fed to AOELM classifier to classify normal controls (NC) versus off medication PD EEG records (SFPD) and NC versus on medication PD EEG records (SOPD). The highest accuracy of 98.91% and 98.55% is obtained in classifying NC versus SOPD and NC versus SFPD, respectively.