Parkinson’s disease detection based on dysphonia measurements

Abstract

Assessing dysphonic symptoms is a noninvasive and effective approach to detect Parkinson’s disease (PD) in patients. The main purpose of this study is to investigate the effect of different dysphonia measurements on PD detection by support vector machine (SVM). Seven categories of dysphonia measurements are considered. Experimental results from ten-fold cross-validation technique demonstrate that vocal fundamental frequency statistics yield the highest accuracy of 88%±0.04. When all dysphonia measurements are employed, the SVM classifier achieves 94%±0.03 accuracy. A refinement of the original patterns space by removing dysphonia measurements with similar variation across healthy and PD subjects allows achieving 97.03%±0.03 accuracy. The latter performance is larger than what is reported in the literature on the same dataset with ten-fold cross-validation technique. Finally, it was found that measures of ratio of noise to tonal components in the voice are the most suitable dysphonic symptoms to detect PD subjects as they achieve 99.64%±0.01 specificity. This finding is highly promising for understanding PD symptoms.