Objective: To leverage a community of researchers and shared wearable data to develop algorithms to estimate the severity of PD specific symptoms.

Background: People with Parkinson’s disease (PwPD) often experience fluctuations in motor symptom severity. Wearable sensors have the potential to help clinicians monitor symptoms over time, outside the clinic. However, to gather accurate and clinically-relevant measures, there is a need to develop robust algorithms based on clinically-labelled data.

Methods: The Levodopa Response Trial captured three-axis acceleration from two wrist-worn sensors and a smartphone located at the waist from 29 PwPD continuously over 4 days. On day 1, in an in-clinic visit, participants performed clinical assessments and motor tasks on their regular medication regimen. During these visits, a clinician also provided symptom severity scores for tremor, bradykinesia, and dyskinesia. On days 2 & 3, sensor data was collected while participants were at home. On day 4, participants returned to the clinic for the same assessments as day 1, but arrived without having taken their medication for at least 10 hours. Leveraging this dataset, Sage Bionetworks, the Michael J Fox Foundation and the Robert Wood Johnson Foundation launched the PD Digital Biomarker DREAM Challenge which made a subset of the data available to researchers to develop robust and accurate algorithms for the estimation of specific symptoms’ severity.

Results: Teams participating in the challenge used several technical approaches, from signal processing to deep learning. 35 submissions were received for the estimation of action tremor severity. Teams achieved an area under the precision-recall curve (AUPR) of 0.444 to 0.75. As for dyskinesia during movement, 37 submissions were received and the teams achieved an AUPR of 0.175 to 0.477. Finally, 39 submissions were received for the estimation of bradykinesia and the teams achieved an AUPR of 0.413 to 0.95. Null expectations for the testing datasets were 0.432, 0.195, and 0.266, respectively.

Conclusions: Making datasets available to the community leverages the creativity of different groups to develop robust and accurate algorithms for the estimation of PD symptom severity. This will lead to better quality and interpretability of data collected in unsupervised settings within the community.

« Back to 2018 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/the-levodopa-response-trial-and-the-parkinson-disease-digital-biomarker-challenge-monitoring-symptoms-of-parkinsons-disease-in-the-lab-and-home-using-wearable-sensors/