Open Source/ Radiology AI/ Community
Open radiology AI, built by radiologists.
A decentralized hub for radiology AI — open source, community driven and self-hosted. Connecting people, knowledge and technology to build better imaging tools together.
Why it is different
Better tools. Stronger together.
Open source
The code is the argument. Anyone can read it, run it, audit it and change it — including the parts that decide what a model does with a study.
Self-hosted
Runs inside your own network. Studies stay where they belong, pseudonymisation happens at the point of ingest, and nothing leaves the house unless you send it.
Community validated
Built and reviewed by people who read images for a living. Claims are measured against real cases, and disagreements are published, not hidden.
The mark
Four ideas in one drawing.
Medical imaging
Concentric rings read as an axial slice — the shape a radiologist sees first.
Decentralized knowledge
A hub with peers, not a centre that owns everything. Gaps in the ring are deliberate.
Community collaboration
Radiologists, developers and hospitals in one network — the amber node is the human one.
Self-hosted open source
Freedom and control in the same sentence: your hardware, your data, your build.
Status
What exists, and what does not.
A radiologist is building this alongside clinical work, so the honest version is the only useful version. Nothing below is a promise.
- Working Proof of concept. Pseudonymisation before storage, a DICOM archive and web viewer, and a reporting layer with a local language model — all self-hosted on a single workstation.
- In progress Imaging analysis. Segmentation and measurement models on a MONAI foundation, plus terminology and translation support for reports in German and Dutch.
- Not yet Clinical validation and certification. No CE marking, no multi-site deployment, no published benchmark. Anyone claiming otherwise is ahead of the code.
Run it yourself
No cloud required.
The proof of concept is a Docker Compose stack: MONAI for imaging, Orthanc for the archive, OHIF for viewing, and a local model for the reporting layer. It is designed to sit on hardware you already own.
# the repository is being prepared for public release
git clone https://github.com/hounsfieldhub/hounsfieldhub
cd hounsfieldhub
cp .env.example .env # set your data path
docker compose up -dSetup notes are being written up as the documentation lands in the repository.