Parkinson’s Disease Detection by Using Machine Learning Algorithms and Hand Movement Signal from LeapMotion Sensor

This work is devoted to the detection of Parkinson's disease (PD) by the kinematic parameters of hand movements using machine learning methods. Hand movements of PD patients (N16) and control group (N16) were recorded using a Leap Motion sensor. Three motor tasks were chosen based on MDS-UPDRS part 3: finger tapping (FT), pronation- supination of the hand (PS), opening-closing hand movements (OC). For the signal received from the sensor, 25 kinematic parameters were calculated by key points. The key point determination was carried out with maximums and minimums finder algorithm, as well as manual marking, using a specially designed user application. For the binary classification (PD or non-PD), for each motor task separately and for three combined, various feature extraction options were used. Four classifiers: kNN, SVM, Decision Tree (DT), Random Forest (RF) were trained. Testing was carried out in the 8-fold cross-validation mode. The best results were obtained using the combination of the most significant features of both hands. The results for each task were the following: for FT 95.3%, for OC 90.6%, for PS 93.8%. The combined features result of all motor tasks was 98.4%.