Dear all,
I am writing to introduce myself to the community and also introduce PDkit, a Python toolkit we created a few years ago with support from MJFF, for the development of end-to-end machine learning pipelines to generate digital biomarkers using data collected by mobile apps and wearables. The PDkit is open source software and we invite everyone in the community to use and contribute. The code and documentation is also available as the newest repository on the MJFF Research Community GitHub.
There is a lot of information available here, including reasonably complete documentation, links to papers describing how the toolkit works, as well as results from clinical studies where it has been employed. Our purpose in developing this software was to help address the current lack of algorithmic and model transparency in this area. Our approach is to facilitate open sharing of common methods employed in studies with the view to enable comparison of results across multiple studies, data sets, hardware and methods.
The current version of the toolkit supports a number of apps developed for the assessment of PD symptoms including mPower, the OPDC app used by Oxford Parkinson’s Disease Centre, HopkinsPD and cloudUPDRS, our own app. There is a separate (less well-developed) real-time version of the toolkit specifically for consuming continuous streaming data from wearables using Apache Beam. Few studies have taken this approach due to the burden on batteries, but this is also an option. The toolkit is extensible and we would be more than happy to work with anyone that would like to add new data sources.
The toolkit provides over 400 biomarkers for tremor, tapping, reaction, gait and voice tests as well as the ability to produce MDS-UPDRS scores if there is relevant training data. Use cases and sample data for testing from our own studies are available on GitHub via the link above. Our current work relates to improvements to voice biomarkers, to match the state-of-the-art in the literature.
Although cloudUPDRS is not open source (due to the contract with Innovate UK which funded that work), we would be delighted to make the app available to anyone who might find it useful for their work.
We would also be delighted to support any members of the community interested in using the PDkit in their work. We are also very keen to receive contributions of new biomarkers, data sources, or anything else that may be of interest.
I hope this is of interest to many of you and look forward to sharing and furthering the work.
All the best,
George