Track #2 - Medical AI Assistant (MAIA)
Problem Statement
Imagine having a digital health record that you control, similar to your email or retirement accounts. A complete and accurate health record that you can easily share, in part or entirely, with your healthcare providers, and that works seamlessly with new AI-based medical assistants as they are developed.
Currently, the role of medical AI in the patient journey is largely driven by hospitals and institutions, rather than by physicians and patients. The problem with a hospital-driven patient journey is that hospitals do not have a fiduciary relationship with the patient as a doctor does.
This challenge seeks to create a decentralized alternative that allows for separately chosen applications for health records and medical AI assistants.
In this decentralized system, patients and physicians must have the freedom to choose both the health record that manages data and the medical AI assistant that interprets and learns from that data. To achieve this, a standard is needed to connect health records with AI assistants. The recently completed Grant Negotiation and Authorization Protocol (GNAP) by the Internet Engineering Task Force (IETF) provides the necessary standard to secure these connections. This challenge focuses on innovating Medical AI Assistants (MAIA) while supporting patient-centered health records and GNAP-controlled access as open-source software.
For more details, please watch this video
Eligibility
This challenge is open to developers, AI researchers, software engineers, and innovators interested in advancing patient-centered care through AI and secure health data management.
Individuals or teams from anywhere in the world may particip
Challenge Requirement
HIE of One will provide open-source software on GitHub for health records, GNAP, and an example MAIA. Participants are encouraged to extend or completely replace the MAIA component and submit a video demonstrating their proposed innovation for the patient journey.
To adhere to judging criteria and open-source principles, submissions must be based on one or more of the synthetic patient records available. While testing MAIA on actual patient records is allowed, submissions containing identifiable real patient data will be disqualified.
The example MAIA provided uses OpenAI GPT-4o, but participants are encouraged to explore other large language models, local language models, or even split or federated learning models.
Rules
All designs submitted should be open source. This means that your design files, prototypes and any related resources should be made publicly available under a suitable open-source license.
Prize Money
Winner - $3000
1st Runner-up - $600
2nd Runner-up - $400
All participants/teams will receive a certificate of participation in recognition of their efforts.
Submission Guidelines
Format : Submission videos (5 minutes or less) of MAIA must be uploaded to YouTube for judging and must be based on open-source code in GitHub.
Submission Platform: Participants must register with an individual or team name with email on the Challenge website. Registered participants will provide links to their YouTube video and GitHub repository to initiate the judging process. Scoring will be based on a 5-minute video that should combine elements of a demonstration with business and sustainability plans. Judges may contact participants with specific questions about submissions.
Deadline: All submissions must be received by September 30, 2024. Late submissions will not be considered.
Frequently Asked Questions
-
Open-source software for the EHR and GNAP components, along with an example MAIA software demo is provided.
-
Software provided can be modified at the participant's discretion. HIE of One software is licensed under Affero GPL3, and accepted improvements will be made available to all participants. MAIA example software is MIT licensed.
-
There is no licensing requirement for the MAIA component submitted. However, open-source submissions will score higher.
-
Standards for signing MAIA documents to be added to the health record are encouraged but not required.
-
A 5-minute video and corresponding source code on GitHub, along with a sample synthetic patient dataset.
-
Yes. Be sure to list all team members in your submission.
