Assessment of Parkinson's Disease Severity based on Automatic Analysis of Facial Expressions and Motor Activity of the Hands

Assessment of the severity of the disease is an important task in the study of Parkinson's disease. Using non-contact methods for assessing the motor activity of patients, quantitative assessments of motor parameters were obtained, including the assessment of facial expressions and the motor activity of the hands of patients with PD. The study involved 18 patients with PD, whose facial expressions and the motor activity of the hands assessed using the MDS-UPDRS scale by neurologist. In this paper, a regression model was developed that allows to predict the total MDS-UPDRS scores for 3 hand movement exercises with R2 0.781 and RMSE 0.893 based on 5 features of motor activity. To predict the MDS-UPDRS scores, the classification problem is also solved. The patient group was divided into 2 groups according to the severity of the disease based on the fitting of a cut-off value, which is the median value of the MDS-UPDRS scores. The feature space was reduced to 4 using PCA. The best classification result 95% was obtained using logistic regression and support vector machine in a 5-fold cross-validation mode.