U.S. flag An official website of the United States government

U.S. Food and Drug Administration

WearGait-PD Longitudinal: Multi-Session Wearables Dataset for Gait in Parkinson’s Disease 

Catalog of Regulatory Science Tools to Help Assess New Medical Devices 

This open-access longitudinal dataset contains raw inertial measurement unit (IMU) and sensorized insole data from individuals with Parkinson’s Disease (PD) synchronized to a gait walkway reference system over multiple time points.  This dataset can also be used to examine associations between gait measures, PD severity indicators, and PD progression within the study cohort.

Technical Description

This open-access dataset collected over multiple sessions (time points) contains raw inertial measurement unit (IMU) and sensorized insole data from individuals with Parkinson’s Disease (PD) synchronized to a gait walkway reference system.  The minimum follow-up time between each session was six months.   IMU data include 3-degree of freedom (DOF) acceleration, rotational velocity, and magnetic field strength; sensor insole data include absolute pressure from 16 sensors in each insole and 3-DOF acceleration and rotational velocity; walkway data include 2D position and relative pressure for each active sensor during every footfall registered on the walkway.  Frame-by-frame annotation of participant actions during gait and balance tasks was incorporated into the dataset using video cameras synchronized to all systems. To expand the utility of this dataset, all data from each session are associated with demographic information and clinical evaluations (e.g., medications, DBS-status, Movement Disorder Society – Unified Parkinson’s Disease Rating Scale [MDS-UPDRS] scores). 

Intended Purpose 

The WearGait-PD Longitudinal RST dataset described here is intended to support the assessment of algorithms using IMU data and/or sensorized insole pressure data from wearables to derive movement metrics and to facilitate the identification of digitally-derived endpoints relevant to the PD population for potential use in clinical trials and/or home monitoring studies.

Testing

Two clinical sites contributed to the data in this RST: Johns Hopkins Outpatient Center (JHOC) and the Seattle Veterans Administration Center for Limb Loss and Mobility (VA CLiMB). PD participants representative of the PD population were recruited for the study.  Participants performed a series of walking tasks in a controlled lab environment as well as a free walking task around the respective facility to capture data more representative of real-world conditions. 

A quality assurance process was implemented to ensure consistent and accurate data collection and is described in detail in Anderson, Eguren, Gonzalez, et al. (2026) [1] and Gonzalez et al. (2024) [2]. Specifically, a core experimental protocol was developed and reviewed by each researcher in consultation with clinical collaborators. This protocol detailed the steps involved in system-set-up, participant preparation, data collection, and data processing and export. For those modalities that required more manual processing (i.e., walkway data and video annotation data), a second researcher reviewed the primary processing completed by the first researcher to ensure proper footfall identification and conformance with the agreed upon annotation event definitions. 

After final files were created (MAT and CSV files), a data quality control process on those files was implemented through a series of automatic and manual checks that involved visual inspection of data streams by a researcher to identify data cleanliness and validity issues. Specifically, this process identified issues related to:

  • Processing and inclusion of all tasks in the MAT and CSV files
  • Integrity of individual IMU sensor data, including the identification of sensor values outside of expected ranges
  • Unexpected missing columns of data 
  • Unexpected large gaps in data
  • Expectations around the variable type for a given data variable (e.g., annotations contain no numeric data, IMU data does not contain any errant non-numeric data)
  • Video annotation events and expectations surrounding inclusion of specific events for specific tasks
  • Alignment of all data, with a particular focus on walkway pressure and sensor insole force data

If an issue was identified, a second researcher was assigned to review and address the issue. Once an issue was addressed, the relevant files were put through the data quality control process again to confirm that all issues were resolved before the final MAT file and final set of CSV files were generated and included in the dataset.

Limitations

There are several systems from which data are captured to curate this dataset. Synchronization of these systems and accurate data alignment are critical to the overall utility of this tool. While data synchronization between the walkway and IMU system was found to be within 0.02 seconds, alignment between the walkway and sensor insoles was less consistent.  To resolve the observed inconsistency, comprehensive testing was conducted to define the synchronization strategy among the systems. This allowed for fine-tuning the alignment of the sensor insole data with that of the walkway. Despite these enhancements, researchers might consider exercising caution when interpreting data that depends on the precise alignment of sensor insoles with walkway/IMU systems [1].

While the raw data output from the walkway enables calculation of any spatiotemporal metric requiring knowledge of timing and position of footfalls, the use of relative pressure as opposed to absolute pressure may limit accuracy of derived kinetic gait metrics [1].

While tasks were intentionally designed to evoke natural movement within near real-world contexts, it is important to note that data were collected in a controlled lab-based environment. Consequently, the generalizability of findings from these data to real-world scenarios may be somewhat restricted [1].

Supporting Documentation

Data description and access: Wearables for gait in Parkinson's Disease and age-matched controls (WearGait-PD) - syn52540892 - Wiki

  1. Anderson, A.J., Eguren, D., Gonzalez, M.A. et al. “WearGait-PD: An Open-Access Wearables Dataset for Gait in Parkinson’s Disease and Age-Matched Controls.” Sci Data 13, 440 (2026). https://doi.org/10.1038/s41597-026-06806-2 
  2. Gonzalez, M, Anderson A., Eguren D, Muir B. and Kontson K., "Step by Step - Quality Control in a Wearable Sensor Dataset Collected from People with Parkinson's Disease," 2024 IEEE 20th International Conference on Body Sensor Networks (BSN), Chicago, IL, USA, 2024, pp. 1-4. 10.1109/BSN63547.2024.10780535 

Contact

Tool Reference