Investigation of Acoustic Features and Machine Learning for Early Detection of Parkinson's Disease

Abstract

Parkinson's disease (PD) is a neurodegenerative disease that afflicts millions of people. The early detection of the disease is crucial. According to recent research, the level of dysarthria is a good indicator for computer-assisted diagnosis and remote monitoring of patients in the early phases. Despite the significance of articulatory deficits in dysarthria among individuals with PD, automatic speech performance evaluation methods mainly concentrate on assessing dysphonia. In this study, our objective was to classify the phonation, articulation et diadochokinetic features by machine learning (ML) algorithms with feature selection technique. Using Italian data, a Lasso-cross validation feature selection algorithm was used to select the voice features extracted from three vocal tasks, followed by three classifiers (Artificial Neural Network (ANN), Random Forest (RF), and Support Vector Machine (SVM)) to detect the disorder. The highest overall classification score achieved an 100% accuracy rate in discriminating between PD and control participants. More interest, articulatory features were found to be the most powerful indicator of PD-related dysarthria than phonation and DDK features among all the classifying algorithms.