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

Alibaba DAMO Academy researchers released RADAR, a generalist vision-language model for abdominal CT diagnosis trained on more than 400,000 contrast-enhanced abdominal CT examinations and 15 million anatomy-aware image-text pairs, with the work published in Science (volume 393, number 6817, pages eaec6129). RADAR learns directly from clinical reports without manual annotation and is reported to deliver expert-level performance across routine and complex clinical tasks. Pre-trained checkpoints and supporting files are available on HuggingFace, with code archived on Zenodo, and the project is released under the Apache License 2.0.

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. Create a conda environment and install the required dependencies: conda create -n radar python=3.10 conda activate radar pip install -r requirements.txt - The pre-trained checkpoints and supporting files are available on HuggingFace https://huggingface.co/radar-generalist . - For convenience, we have provided the demo nifty, and predicted results in CSV format in this repo. The supporting files required for the inference demo and training can be downloaded from HuggingFace. - Download via scripts: We provide two helper scripts under download scripts/ to fetch the required files from HuggingFace: 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 Code can also be archived on Zenodo https://zenodo.org/records/21271172 . For detailed instructions, please refer to the following guides: | Guide | Description | |---|---| | Training https://github.com/alibaba-damo-academy/damo-radar/blob/main/docs/TRAINING.md | Train RADAR/RADAR+ from scratch or fine-tune on MERLIN data; inference and evaluation are also included. | | Inference https://github.com/alibaba-damo-academy/damo-radar/blob/main/docs/INFERENCE.md | 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 https://github.com/alibaba-damo-academy/damo-radar/blob/main/docs/PREPROCESS.md | Image/mask and radiology report preprocessing code, which can be used to process the MERLIN data or your own custom data. | This project is built upon the following open-source projects: - LAVIS https://github.com/salesforce/LAVIS BSD 3-Clause License - nnU-Net https://github.com/MIC-DKFZ/nnUNet Apache License 2.0 - MONAI https://github.com/Project-MONAI/MONAI Apache License 2.0 - 3D-ResNets-PyTorch https://github.com/kenshohara/3D-ResNets-PyTorch MIT License This project is released under the Apache License 2.0 https://github.com/alibaba-damo-academy/damo-radar/blob/main/LICENSE . Portions of the code are derived from third-party open-source projects that are distributed under their own licenses see the Acknowledgements acknowledgements above . Their original license texts are retained in THIRD PARTY LICENSES.md https://github.com/alibaba-damo-academy/damo-radar/blob/main/THIRD PARTY LICENSES.md . 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} }