Quantifying and Removing Free-living Uncertainty for Effective Parkinson’s Disease Diagnosis Using Smart Watch

Wearable technology has played a crucial role in the diagnosis and long-term monitoring of Parkinson’s disease in computer-aided diagnosis. With the development of wearable devices, mobile health detection has become a hot spot. Different from traditional computer-aided diagnosis, this paper is devoted to modeling Parkinson’s disease staging in the field of mobile health using smart watch. First, we analyze the factors that hinder the Parkinson’s disease staging model in the field of mobile health from the three aspects of environment, wearables and wearing. Second, in a free-living environment, the problem of uncertainty in activities of daily living has a fatal impact on machine learning models. On the one hand, the uncertainty of activities caused by patients’ independent completion of activity collection; on the other hand, the uncertainty of staging labels caused by the use of remote video diagnosis by clinicians. Finally, considering many of the above factors, we used smart watch to collect patients’ daily activity data, and built a machine learning model to stage the condition of Parkinson’s patients in free-living environment. Following that, experiments were conducted using our collected dataset of 70 Parkinson’s disease patients to quantify and remove uncertainty in daily activity data in free-living environment. The experimental results show that for Parkinson’s disease staging in free-living environment, the uncertainty of the data is the main factor leading to the instability and low performance of the model. Meanwhile, the anomaly detection and uncertain label detection can alleviate the impact of the uncertainty of the data on the model. In addition, it is verified that the staging framework for Parkinson’s patients in free-living environment constructed using smart watch is feasible.