Wearable sensors offer exciting opportunities to study Parkinson’s disease (PD) in daily life, but turning real-life, high-frequency sensor data into reliable and meaningful measures can be challenging. With support from MJFF, we developed ParaDigMa - an open-source python toolbox to help researchers tackle this challenge.
ParaDigMa provides validated pipelines to extract digital biomarkers for motor and non-motor signs, based on continuous, wrist-based monitoring. It builds on algorithms we developed and tested under real-world conditions. Our motivation for creating this toolbox was to make these algorithms easily accessible, and to facilitate transparent and reproducible analyses across studies and devices.
What can it do?
ParaDigMa helps you with all processing steps needed to transform raw, passive monitoring data (i.e., wrist accelerometer, gyroscope or photoplethysmography (PPG) signals) into digital biomarkers for studying PD - covering preprocessing, feature extraction, symptom classification, symptom quantification and aggregation. You can interact with the results on different timescales, for example by using week-level digital biomarkers to study long-term disease progression, or by zooming in to second-level predictions to study fluctuations throughout the day.
It currently includes pipelines for:
-
Tremor: detecting rest tremor and quantifying its severity, based on gyroscope data (Timmermans et al. 2025).
-
Reduced arm swing during gait: detecting gait, filtering out gait segments with other arm activities (such as gestures), and quantifying the arm swing range of motion, based on accelerometer and gyroscope data (Post et al. 2025).
-
Pulse rate analysis: evaluating PPG signal quality, accounting for periodic PPG motion artifacts in the PD population (for example introduced by tremor) through fusion with accelerometry, and quantifying pulse rate to enable assessment of cardiac autonomic dysfuntion in PD (Veldkamp et al. 2025).
These pipelines were developed and validated using real-world data from the Personalized Parkinson Project and Parkinson@Home Validation Study. We are currently finalizing publications that demonstrate their ability to capture disease progression in early-stage PD, based on two years of continuous follow-up data from the Personalized Parkinson Project.
Can I use it for my sensor data?
Have you collected raw wrist sensor data through passive monitoring of persons with PD? If so, ParaDigMa might be a suitable option for your analysis. It is designed to be generalizable across different sensor devices, as long as they provide access to raw sensor data with the right configurations (e.g., sampling rate, range). Luckily, there are now various options for collecting raw sensor data in daily life, including more research-oriented devices (e.g., from Axivity, Empatica, and ActiGraph) and consumer smartwatches with the ability to log raw sensor data.
It is good to note that – in addition to these hardware requirements – there are also requirements regarding the context of use, which vary across pipelines. For example, the pulse rate pipeline is only applicable to individuals without specific rhythm disorders, such as atrial fibrillation. Detailed requirements for each pipeline can be found in the documentation.
Getting started
ParaDigMa is available as a Python package via PyPI - so to install, simply run “pip install paradigma”. The documentation includes tutorials for preparing data and using each processing pipeline for extracting digital biomarkers. The API reference provides detailed documentation of all underlying modules and functions.
Need help to get started? We’re happy to help - feel free to reach out to us!
Interested in contributing?
We warmly welcome feedback and new contributions from the community — whether it’s suggestions for additional algorithms, improvements to existing pipelines, or ideas for additional features. The toolbox is designed to be modular, allowing for easy extension with new algorithms and functionalities. If you’d like to contribute, please have a look at our contributing guideline.
We’re excited to hear how you could use it in your research, and we look forward to collaborating with the community to further improve and expand it together.
Best, Luc