Launched in January of 2020, and ending in May, the BEAT-PD Dream Challenge was designed to understand whether passive data collected from sensors could track symptom severity and disease progression in patients with Parkinson’s disease. Since the data would be collected during daily life, not specific activities, researchers were unsure of how that information would translate into a medical standpoint.
Hosted by the Michael J. Fox Foundation (MJFF), Sage Bionetworks, BRAIN Commons, Radboud University Medical Center, Evidation Health, and Northwestern University, the contest boasted a $25,000 prize.
BEAT-PD follows a previous contest and challenge. In the prior challenge, researchers determined that data collected during physician-monitored tasks could predict disease progression.
Ultimately, 43 teams competed. Using accelerometers and gyroscopes, teams collected data and estimated patients’ treatment status, symptom severity, and Parkinson’s disease progression.
Can researchers use data gathered from wearable sensors to learn more about Parkinson’s disease, symptoms, and disease progression? Ultimately, the four winners found effective ways to monitor patient data during daily life.
- Team dbmi
- This team consisted of Yidi Huang, Brett Beaulieu-Jones, Mark Keller, and Mohammed Saqib. Team dbmi performed data analysis and tests to determine the efficacy of wearable sensors on modeling and predicting disease progression. The team notes:
We found that the patient-specific, windowed models we tried outperformed full-dataset, full-measurement models. We hypothesize that this is due to the self-labeled nature of the dataset, as symptom severity may have been reported differently across patients.
- See their full findings and data here.
- Team ROC BEAT-PD
- This team consisted of Alex Page, Greta Smith, Robbie Zielinski, Monica Javidnia, and Charles Venuto. According to the team, their method was:
Using smartwatch and smartphone sensor data, we have developed algorithms to predict patient self-reported medication status, dyskinesia severity, and tremor severity from two separate clinical study cohorts. Our approach to divide recordings into 30-second snippets allowed for analysis of data at small granularity, which could then be aggregated over larger timescales.
- See their findings here.
- Yuanfang Guan
- Hailing from the University of Michigan, Yuanfang Guan created a core method of:
1-D convolutional network model with spatial and time augmentation.
- By using a deep-learning model, Guan was able to further track and develop data. See the method here.
- Team HaProzdor
- From Bar-Ilan University comes Team HaProzdor, consisting of Ayala Matzner, Yuval El-Hanany, and Izhar Bar-Gad. Currently, findings and data from this team are not available. However, the team is said to have applied signal processing techniques to smartphone sensor data, much like ROC BEAT-PD and dbmi.
Moving forward, all four teams may collaborate, create optimized testing models, and co-author a paper on the results.
Read the source article here.