# RadFM (generalist radiology foundation model)

Source: https://onco.cc/technologies/radfm/  
OnCo record `radfm` (Technology). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

An open generalist model that answers questions about 2D and 3D scans.

## Summary

RadFM is an open generalist radiology foundation model that pairs a visual encoder with a large language model and is trained on interleaved image and text, so users can ask questions about a scan in plain language. The 2023 arXiv paper trained it on the 16M-scan MedMD dataset, and it handles CT, MRI and X-ray in both 2D and 3D with text prompts. It is a research system for groups exploring conversational or multimodal radiology assistants rather than a clinical product. Its accuracy is below specialist models on individual tasks, which is the central trade-off of generalist medical models, and oncology-specific evaluation is limited. For a newcomer: RadFM is an early attempt at a chatbot that can look at many kinds of scan, broad but not yet as good as dedicated tools.

## Fields

- Kind: Technology
- Status: emerging
- Last checked: 2026-09-08
- Tags: foundation-model; radiology
- Principle: RadFM pairs a visual encoder with an LLM trained on interleaved image-text.
- Since: 2023
- Strengths: Modality breadth
- Limitations: Accuracy below specialist models

## Sources

- RadFM (arXiv 2023): https://arxiv.org/abs/2308.02463

## Connected records

- fronts: [AI & Computation](https://onco.cc/fronts/ai-computation/), [Imaging](https://onco.cc/fronts/imaging/)
- technologies: [AI in radiology](https://onco.cc/technologies/radiology-ai-screening/), [Pathology & radiology foundation models](https://onco.cc/technologies/pathology-foundation-model/)
- terms: [Radiology imaging as a data modality (CT, MRI, TCIA)](https://onco.cc/terms/radiology-imaging-modality/)

---
JSON: https://onco.cc/api/v1/entities/radfm.json