PDMove

Abstract

The medicine adherence in Parkinson's disease (PD) treatment has attracted tremendous attention due to the critical consequences it can lead to otherwise. As a result, clinics need to ensure that the medicine intake is performed on time. Existing approaches, such as self-report, family reminder, and pill counts, heavily rely on the patients themselves to log the medicine intake (hereafter, patient involvement). Unfortunately, PD patients usually suffer from impaired cognition or memory loss, which leads to the so-called medication non-adherence, including missed doses, extra doses, and mistimed doses. These instances can nullify the treatment or even harm the patients. In this paper, we present PDMove, a smartphone-based passive sensing system to facilitate medication adherence monitoring without the need for patient involvement. Specifically, PDMove builds on the fact that PD patients will present gait abnormality if they do not follow medication treatment. To begin with, PDMove passively collects gait data while putting the smartphone in the pocket. Afterward, the gait preprocessor helps extract gait cycle containing the Parkinsonism-related biomarkers. Finally, the medicine intake detector consisting of a multi-view convolutional neural network predicts the medicine intake. In this way, PDMove enables the medication adherence monitoring. To evaluate PDMove, we enroll 247 participants with PD and collect more than 100,000 gait cycle samples. Our results show that smartphone-based gait assessment is a feasible approach to the AI-care strategy to monitor the medication adherence of PD patients.