Analysis of surface EMG signal morphology in Parkinson's disease

Abstract

A novel approach is presented for the analysis of surface electromyogram (EMG) morphology in Parkinson's disease (PD). The method is based on histogram and crossing rate (CR) analysis of the EMG signal. In the method, histograms and CR values are used as high-dimensional feature vectors. The dimensionality of them is then reduced using the Karhunen-Loève transform (KLT). Finally, the discriminant analysis of feature vectors is performed in low-dimensional eigenspace. Histograms and CR values were chosen for analysis, because Parkinsonian EMG signals typically involve patterns of EMG bursts. Traditional methods of EMG amplitude and spectral analysis are not effective in analyzing impulse-like signals. The method, which was tested with EMG signals measured from 25 patients with PD and 22 healthy controls, was promising for discriminating between these two groups of subjects. The ratio of correct discrimination by augmented KLT was 86% for the control group and 72% for the patient group. On the basis of these results, further studies are suggested in order to evaluate the usability of this method in early stage diagnostics of PD.