Wearable Sensors for Prodromal Motor Assessment of Parkinson's Disease using Supervised Learning

Abstract

Parkinson's disease (PD) is a common neurodegenerative disorder characterized by disabling motor and non-motor symptoms. Idiopathic hyposmia (IH), a reduced olfactory sensitivity, is a preclinical marker for the pathology and affects >95% of PD patients. In this paper, SensHand V1 and SensFoot V2, two inertial wearable sensors for upper and lower limbs, were developed to acquire motion data in ten tasks of the MDS-UPDRS III. Fifteen healthy subjects of control, 15 IH people, and 15 PD patients were enrolled. Seventy-one parameters per side were computed by spatiotemporal and frequency data analysis, and the most significant were selected to distinguish among the different classes. Performances of supervised learning algorithms (i.e., Support Vector Machine (SVM), and Random Forest (RF)) were compared on two-group and three-group classification and considering upper and lower limbs separately or together as a full system. Excellent results were obtained for healthy vs. patients classification (accuracy 1.00 for RF, and 0.97 for SVM), and good results were achieved by including IH subjects (0.92 F-measure with RF) within a three-group classification. Overall, the best performances were obtained using the full system with an RF classifier. The system is, thus, suitable to support an objective PD diagnosis. Furthermore, combining motion analysis with a validated olfactory screening test, people at risk for PD can be appropriately analyzed, and subtle changes in motor performance that characterize the prodromal phase and the early PD onset can be identified.