Vocal test Analysis for Assessing Parkinson's Disease at Early Stage

Parkinson's Disease (PD) is a neurodegenerative disease, worldwide affecting millions of people, which results with speech disorders even at early stages. Here, we developed vocal tests' assessment of PD patients by means of a robust approach based on balanced data and 10-fold cross-validation. In particular, vocal tests consisted in the sustained vowel /e/ and three sentences, from which a number of features were selected by means of audio feature extraction tool. The features were analyzed using different classifiers, such as Multilayer Perceptron (MLP), Support Vector Machine (SVM) with Sequential Minimal Optimization (SMO), and Naive-Bayes. In addition, statistical analysis was performed consisting in vocal tests and classifiers. In particular, from the analysis of one of the sentences, in revealing subjects affected by PD we obtained an accuracy as high as 96.51% (with a p-value of 0.05), among the highest reported in literature. Both Wye -Bayes and SVM-SMO outperformed MLP with a mean accuracy of 94.34% and 93.806% respectively (p-value = 0.05).