The Open Accelerator Hackathon (Winner)

The Open Accelerator Healthcare Hackathon — Track 1: Federated Medical Imaging Search

Find where relevant pediatric imaging data lives across a network of hospitals — without moving it, without exposing PII, and without granting access to anyone who isn't authorized.

The Problem

A child being treated for a brain tumor gets scanned every few months so clinicians can tell whether it's growing, stable, or responding to treatment. Building models that help specialists measure these changes reliably requires examples from many children, scanners, hospitals, and disease stages — but pediatric brain tumors are rare, so no single hospital has enough cases on its own.

The scans researchers need may already exist somewhere in a network of children's hospitals. Today there's no safe way to find out, and even less clarity on how to access what's found without exposing sensitive patient health information. Researchers currently approach hospitals one at a time, spending months on contacts and approvals before even learning whether enough suitable data exists.

MedFind answers, for a participating network: where might relevant data exist, how much is available, who is authorized to see it, and how can it be securely retrieved?

Our Solution

MedFind is a distributed search network: a central gateway takes a researcher's natural-language query, expands it into the clinical vocabulary each hospital actually uses in its reports, and forwards the query to every hospital node in parallel. Each hospital node runs its own database and evaluates the query — and, separately, any retrieval request — against its own local access policy. The gateway never stores patient data and never overrides a hospital's access decision; it only carries the researcher's identity token and aggregates what the hospitals are willing to disclose.

Each hospital node is an independent FastAPI service holding its own synthetic dataset in memory. It decides, per role, whether a caller may retrieve a de-identified study and different hospitals are allowed to (and do) disagree about the same researcher.

https://github.com/pjsk02/TOA-HealthHack