damo-radar README

The project's README from github.com/alibaba-damo-academy/damo-radar. As one markdown file: README.md.

RADAR: An Expert-Level Generalist AI for Abdominal CT Diagnosis

Paper GitHub Zenodo Hugging Face ModelScope License

RADAR is a generalist vision-language model trained on over 400,000 contrast-enhanced abdominal CT examinations with 15 million anatomy-aware image–text pairs, learning directly from clinical reports without manual annotation. RADAR provides a scalable and versatile framework for radiology AI, demonstrating expert-level performance across both routine and complex clinical tasks.

RADAR Overview

Setup

Create a conda environment and install the required dependencies:

conda create -n radar python=3.10
conda activate radar
pip install -r requirements.txt

Model & Data

cd download_scripts
# Download model checkpoints and support files into ckpt/
python download_checkpoints.py
# Download auxiliary data (processed masks)
python download_auxiliary_data.py

Zenodo

Code can also be archived on Zenodo.

Documentation

For detailed instructions, please refer to the following guides:

Guide Description
Training Train RADAR/RADAR+ from scratch or fine-tune on MERLIN data; inference and evaluation are also included.
Inference 1. An inference demo with a radar pre-trained checkpoint on RAD-CT, and 2. Inference and evaluation of radar performance on the external MERLIN test set.
Preprocess Image/mask and radiology report preprocessing code, which can be used to process the MERLIN data or your own custom data.

Acknowledgements

This project is built upon the following open-source projects:

License

This project is released under the CC BY-NC-SA 4.0.

Portions of the code are derived from third-party open-source projects that are distributed under their own licenses (see the Acknowledgements above). Their original license texts are retained in THIRD_PARTY_LICENSES.md.

The Radar model is currently intended for research purposes only. Further improvements and prospective clinical studies are still required before it can be used directly for clinical deployment.

Citation

If you find RADAR useful in your research, please cite our paper:

@article{damo-radar-2026,
    author = {Qi Zhang and Jianpeng Zhang and Weiwei Cao and Zilin Lu and Wanxing Chang and Haonan Ding and Cao Chen and Zhi Li and Xing Xue and Sinuo Wang and Shaoteng Zhang and Yutong Xie and Yong Xia and Qi Wu and Zhongyi Shui and Xi Li and Zhilin Zheng and Yanjie Zhou and Tony C.W. Mok and Yingda Xia and Hongkan Wang and Xianghua Ye and Tao Ma and Jie Peng and Xiaoguang Wang and Jian Ding and Yuming Gao and Huazhen Ye and Yiping Liu and Dongjie Chen and Zhaomin Ni and Jianwen Ning and Wei Zhang and Jian Liu and Chaohui Yu and Shenghong Ju and Jianfeng Zhang and Wenbo Xiao and Ling Zhang and Tingbo Liang },
    title = {An expert-level generalist AI for abdominal CT diagnosis},
    journal = {Science},
    volume = {393},
    number = {6817},
    pages = {eaec6129},
    year = {2026},
    doi = {10.1126/science.aec6129},
    URL = {https://www.science.org/doi/abs/10.1126/science.aec6129}
}

damo-radar (alibaba-damo-academy/damo-radar)

Searching Database…

208 ·GitHub
Damo-radar
What is this?RADAR is a generalist vision-language AI model for abdominal CT diagnosis, trained on over 400,000 contrast-enhanced abdominal CT exams with 15 million anatomy-aware image-text pairs derived directly from clinical reports.
Is this a ready-to-use diagnostic tool or a research model I need to train/fine-tune?It is a research-grade pretrained vision-language model with code for training, fine-tuning, and inference; pre-trained checkpoints are hosted on HuggingFace and ModelScope.
How do I get it running?Create a conda environment and install the required dependencies, then download the pretrained checkpoints and supporting files from HuggingFace (helper scripts are provided under downloadscripts/) before following the Training/Inference/Preprocess guides.
What data and tools does it depend on?It needs a conda environment with PyTorch/CUDA, plus the pretrained checkpoints and demo NIfTI files from HuggingFace; the code builds on LAVIS, nnU-Net, MONAI, and 3D-ResNets-PyTorch.
Can I use this for actual clinical diagnosis?No, the authors explicitly state it is research-only and not ready for clinical deployment until further validation.
Is the license permissive for commercial use?No, it's released under CC BY-NC-SA 4.0, which is non-commercial and share-alike; bundled third-party components keep their own separate licenses.
What test data can I evaluate it on?The docs include guides for inference and evaluation on the RAD-CT dataset using the pretrained checkpoint, as well as evaluation on the external MERLIN test set.
Can I train on my own CT scans?Yes, the Preprocess guide covers image/mask and radiology report preprocessing so you can process MERLIN data or your own custom data for training.
What makes RADAR different from a typical classifier model?It's trained end-to-end from raw clinical radiology reports without manual annotation, learning anatomy-aware image-text alignment across 15 million pairs rather than relying on hand-labeled categories.
What are similar projects?LongCat-Next, Qwen-Drive-1.0, ZDTaichu5.0-9B, OraRL, sam2, pointdit, EasyWAM, vegaml
Which topics is it in?vision-language-models, medical-imaging, multimodal-ai, healthcare, vision, foundation-models, vision-language-model, pretrained-models, fine-tuning, multimodal, transformer, academic-research
Creator
alibaba-damo-academy
Size
511 MB
Stars
208
Updated
CommitSep 22
Age
3mo
Commits · 90 Days