A systematic approach to diagnose Parkinson's disease through kinematic features extracted from handwritten drawings

Parkinson’s disease is a slowly progressing neurodegenerative disorder that is not easy to diagnose at the early stages because of delayed symptoms. The most usual ways to diagnose this disease is either by reviewing the medical history of the patient and looking at the computerized tomography scans along with the magnetic resonance imaging by a neurologist or by analyzing the body movements of the patient by the body movement analysts. However, recent research work indicates that Parkinson’s can be effectively diagnosed at an early stage by measuring the changes in handwriting. In this work, the authors have proposed a Parkinson’s disease diagnosis system by analyzing the kinematic features extracted from the handwritten spirals drawn by patients. The publicly available University of California, Irvine Parkinson’s disease spiral drawings using digitized graphics tablet dataset is used in this study. A total of 29 kinematics features are extracted from the dataset. The class imbalance problem in the dataset is handled by the synthetic minority oversampling technique because the dataset is highly imbalanced. Relevant features are selected using the genetic algorithm and mutual information gain feature selection methods. The performance of four classifiers support vector machine, random forest, AdaBoost and XGBoost are analyzed in terms of accuracy, sensitivity, specificity, precision, F-measure, and area under ROC curve. Tenfold cross-validation method is used for validating the results. The combination of mutual information gain feature selection method with AdaBoost classifiers outperforms with 96.02% accuracy.