Skip to main content

Future Roadmap

There are several critical areas of future investment for the MEDS project. These areas are outlined below, and we welcome contributions from the community to help us achieve these goals. To contribute, please file a GitHub Issue in the core MEDS repository. Note that this is just a smattering of the issues we are thinking about; if you have an idea that is not listed below, don't hesitate to reach out to us!

Immediate Future​

MEDS is currently undergoing significant expansion and development. We are actively working on the following aspects:

  1. Improved documentation and tutorials. In addition, writing up formal, finalized papers for the MEDS ecosystem and MEDS-DEV benchmarking system.
  2. Additional tools, datasets, models, ETLs -- as driven by community need.
  3. Resources to make it easier to build MEDS tools, datasets, and models in a tested, validable manner. E.g., synthetic or demo datasets for automated testing of any new models or tools. See this GitHub issue to contribute to that discussion.
  4. Expanding the set of covered tasks significantly and engendering community buy-in on defining and collaboarating on tasks in this way.

Near Future to Longer-term​

Multi-modality​

The MEDS working group believes strongly that multi-modal data is essential for health AI. We are actively working on integrating multi-modal data into the MEDS standard, and we welcome contributions from the community towards this front! See this GitHub Issue to add to the discussion!

Data Visualization and Understanding​

We believe that one of the key things that is missing from the Health AI ecosystem, especially for junior students and researchers, are tools to help visualize, understand, interact with, and improve the quality of health data. Given how paramount understanding the realities of healthcare and the data generative process, it is essential that we have tools to help us understand the data we are working with. To help ideate and build efforts in this space, get in touch with us and we can loop you in on further discussions on this issue internally, or start building a tool for this yourself and let us know how we can help you!

Support tooling-based harmonization utilities​

While MEDS (intentionally) prioritizes data standardization over harmonization, we recognize that in a number of cases data harmonization or data linkage with external ontologies can be very valuable. There are preliminary efforts underway to enable such efforts; if you are interested, please see this GitHub pull request.

Better metadata integration in MEDS-DEV​

Right now, it is up to the community's best efforts to appropriately link tasks, datasets, and models together when running specific experiments, despite the fact that, in some cases, various combinations of these three should not go together -- e.g., we cannot study a primary care task on the MIMIC ICU datasets. We are currently brainstorming on ways to structure this sort of dataset-task-model metadata in a way that is easy to use and understand, and we welcome contributions to this discussion.