Effective Severity Assessment of Parkinson's Disease with Wearable Intelligence using Free-living Environment Data

An effective paramedic diagnostic model for the severity of Parkinson's disease (PD) could help hospitals reduce their workload, especially in countries or regions where medical resources are scarce. However, there are still exist two challenges that limit the advance of this work. Firstly, most of the current research is in laboratory settings, as such, the patient's activity data is relatively standardized and many anomalies are ignored. Secondly, not all activity signal segments reflect the disease characteristics, which causes labeling uncertainty. To address above challenges, we collect more practical PD activities signal from free-living environment and propose an effectively and robustness PD severity assessment framework. Specifically, we first collect wearable data from 53 PD patients and 70 health controls (HC) with 16 daily activities based on Unified Parkinson's Disease Rating Scale Part III scale. Data analyses indicate that many anomalies in free-living activities seriously affect the classification performance of model. We propose a novel multi-classification framework for automatic PD diagnosis to eliminate the effect of abnormal data and labelling uncertainty. The results show that the proposed framework can achieve an accuracy of 92.09% ± 0.16 in diagnosing the PD stage in a free-living environment, which can effectively address the anomalies and label uncertainty in the free-living environment.