Quantitative assessment of Parkinson disease using wearable sensing system

1. BackgroundParkinson’s Disease (PD) is the second most common neurodegenerative disorder. It presents with characteristic and disabling motor symptoms, such as tremors, muscular rigidity, postural instability, Bradykinesia and Hypokinesia, caused by a loss of brain dopaminergic neurons. Currently, for the assessment of movement disorders, a neurologist uses a visual examination of motor tasks and semi-quantitative rating scales, such as the Hoehn-Yahr (HY) Scale and the Movement Disorder Society -Unified Parkinsons Disease Rating Scale(MDSUPDRS). The most accurate objective testing for PD consists of specific brain scanning techniques (e.g. SPECT DATSCAN) that can measure the dopamine level and brain metabolism. 2. Method Although there is clinical demand for new technology that can accurately measure Parkinsonian tremors, automatic scoring of Parkinsonian tremors using machine-learning approaches has not yet been employed. This study aims to fill this gap by proposing measurement system which is capable of objective and quantitativeanalysis of movements of the upper and lower limbs using an Inertial Measurement Unit (IMU). This system is also used machine learning algorithms as a way to predict the UPDRS, which mimic how neurologists rate scores in actual clinical practice. In this study, the tremor signals of 20 patients with PD are measured using a wrist-watch-type wearable device consisting of an accelerometer, a gyroscope and a magnetometer. The displacement and angle signals are calculated from the measured acceleration and angular velocity, andthe acceleration, angular velocity, displacement, and angle signals are used for analysis. We also aim to differentiate PD patients from healthy controls, on the basis of features derived from plantar vertical ground reaction force (VGRF) data during walking at normal pace using dataset collected 29 PD patients. Nineteen features are selected to be extracted from each signal, and the pairwise correlation strategy is used to reduce the number of feature dimensions. With the selected features, a decision tree (DT), support vectormachine (SVM), discriminant analysis (DA), random forest (RF), and k-nearest-neighbor (kNN) algorithm are explored for automatic scoring of the Parkinsonian tremor severity and UPDRS estimation. The performance of the employed classifiers is analyzed using accuracy, recall, and precision, and compared to other findings in similar studies. 3. Results Our results show that SVM with cubic kernel outperformed other classifiers with an accuracy of 93.6%, the sensitivity of 93.1%, and specificity of 94.1%. In comparison to other studies, considering same dataset, our designed prediction system improves the classification performance by approximately 10%. 4. Conclusions Implementation of the proposed solution in this project investigates the accuracy of disease detection and the subtypes over different machine learning algorithms where the results show a 73% accuracy of diseasedetection and 85.55% accurate subtype detection.