Our MedGemma Impact Challenge invited a global community of developers to build prototype applications with Google’s open medical models for real-world clinical needs. Today, we’re announcing the winners.
Yun Liu
Senior Staff Research Scientist, Google Research
Bullet points
- This article announces the winners of Google's MedGemma Impact Challenge, a competition for healthcare AI developers.
- Over 850 teams used Google's open-source MedGemma model to tackle tough healthcare problems.
- Winning projects included demos for critical care, mental health, pathology, and pneumonia management.
- Google is spotlighting these projects to show how AI can bridge healthcare gaps globally.
- Sign up for Google's newsletter to follow how developers are using AI in healthcare.
Summaries were generated by Google AI. Generative AI is experimental.
Your browser does not support the audio element.
Listen to article
This content is generated by Google AI. Generative AI is experimental
[[duration]] minutes
Launched in late 2024, Google’s Health AI Developer Foundations (HAI-DEF) program was created to provide global developers with open weight models, subject to the HAI-DEF Terms of Use, to solve complex healthcare challenges. Following the addition of MedGemma, Google's most capable open model for Health AI development, we introduced MedGemma 1.5 this past January and launched the MedGemma Impact Challenge in collaboration with Kaggle.
The entries from over 850 teams demonstrate HAI-DEF’s potential to transform global health. We saw developers tackle diverse challenges that previously would have required much more resource intensive, ground-up development. Today, we are thrilled to announce the winners.
EpiCast is a mobile-first demo solution built to bridge a critical gap within the Economic Community of West African States. By using a fine-tuned MedGemma model alongside MedSigLIP and HeAR, the system enables community health workers to transform unstructured clinical observations in local languages into structured WHO Integrated Disease Surveillance and Response (IDSR) signals, facilitating the early identification of disease outbreaks.
Sunny is a mobile-first demo designed to help users self-examine and track skin changes for potential signs of cancer. Using a fine-tuned MedGemma model, it generates structured reports from skin photographs while preserving user privacy.
Designed for resource-limited settings, FieldScreen AI demonstrates a novel AI-based tuberculosis screening workflow for community health workers. It uses a fine-tuned MedGemma to analyze chest X-rays and a classifier built based on the HeAR model to detect signs of TB in cough audio. The system runs entirely on-device, using MedASR for voice input and TranslateGemma for local language output.
Tracer demonstrates an AI-driven workflow using MedGemma to help prevent medical errors. By extracting hypotheses from physician notes and reconciling them against incoming test results, the model assigns confidence scores to flag potential discrepancies or incomplete tests for immediate human review.
Special technology winners
These awards honor three specialized themes: agentic workflow improvement, fine-tuning for novel tasks and edge-AI solutions for on-device deployment.
Together, they celebrate teams demonstrating the versatility of HAI-DEF models in driving efficiency, expanding capabilities, and delivering real-world impact.
ClinicDX, an integrated clinical AI demo, uses a custom fine-tuned MedGemma model to bring advanced diagnostics directly into OpenMRS for health centers in sub-Saharan Africa. It operates entirely offline, querying over 160+ WHO and MSF guidelines to inform its outputs.
UniRad3s is a demo that uses a fine-tuned MedGemma model and incorporates MedSAM2 to streamline radiology workflows into three pillars: "Spot" (anomalies), "Segment" (3D lesions) and "Simplify" (patient-friendly automated reporting).
BridgeDX, an offline decision-support demo, is inspired by the healthcare gaps witnessed during the 2015 Nepal earthquake. It anchors its reasoning in guidelines from the WHO and MSF and the Orphanet rare disease database to support community health workers and first responders.
CaseTwin is a clinical decision-support demo that matches acute chest X-rays with historical "twins" and uses an agentic workflow to accelerate referrals, turning an hours-long manual retrieval process into a few minutes in rural hospitals.
BigTB6 is a voice-driven screening demo designed for tuberculosis and anemia screening. It identifies critical biomarkers through cough analysis, chest X-ray and physical pallor assessment to support triage in resource-constrained environments.
Honorable mentions
The selection committee also wishes to acknowledge these entries for their ingenuity:
- Dual Path ICU is a demo designed to help manage high-intensity critical care workflows.
- Sentinel is an on-device mental health monitoring demo for veterans between clinical visits.
- Enso Atlas is a clinical decision support demo designed to assist pathology workflows.
- CAP CDSS is an agentic assistant demo for managing Community-Acquired Pneumonia (CAP) in high-pressure environments.
We were truly humbled by the ingenuity of every participant. We will continue to spotlight the remarkable ways developers use these building blocks to bridge healthcare gaps across the globe. Bookmark goo.gle/hai-def and sign up for our newsletter to follow the journey.
Done. Just one step more.
Check your inbox to confirm your subscription.
You are already subscribed to our newsletter.
You can also subscribe with a