RobustPDx: A Structured Web-Based Parkinson’s Assessment Dataset and Benchmark for AI Robustness to Device Type and Handedness
TL;DR
RobustPDx is introduced, a dataset comprising structured motor and cognitive tasks performed remotely via an accessible web-based assessment completed by 261 participants, including 73 individuals with self-reported PD, 33 individuals reporting suspected PD, and 155 non-PD controls, each completing standardized tasks remotely using their own computer.
Abstract
Parkinson’s disease (PD) is a progressive neurodegenerative disorder associated with motor symptoms and cognitive changes. Its increasing global disease burden highlights the need for scalable approaches to support remote assessment and digital biomarker discovery. Digital health platforms offer scalable opportunities to assess motor and cognitive performance metrics relevant to PD diagnosis remotely, minimizing logistical and geographic barriers. We introduce RobustPDx, a dataset comprising structured motor and cognitive tasks performed remotely via an accessible web-based assessment completed by 261 participants, including 73 individuals with self-reported PD, 33 individuals reporting suspected PD, and 155 non-PD controls, each completing standardized tasks remotely using their own computer. The dataset includes raw task-level interaction data and a feature-engineered analytic dataset containing 79 derived features spanning five task categories related to fine motor control and working memory. Each record in the resulting dataset is paired with demographic data, device type, and handedness annotations. Notably, including device type and handedness metadata enables analyses along traditionally underexplored but important axes of variation in remote consumer digital health. Together, these data support the development and benchmarking of AI-driven models for remote PD detection and robust digital biomarker discovery.