Can I use DAMO RADAR commercially?

Not the weights. DAMO RADAR splits its licensing: the code on GitHub is Apache-2.0, but the checkpoints on Hugging Face are CC BY-NC-SA 4.0, which allows research and requires share-alike attribution while ruling out commercial deployment. A company wanting to ship RADAR inside a product would need a separate agreement with Alibaba DAMO Academy.

Is DAMO RADAR cleared for clinical use?

No. DAMO RADAR has no clearance under the Software as a Medical Device pathway, so it is a research release rather than a diagnostic product. The Science study measures how it performs against radiologists on retrospective exams; using it on live patients would require regulatory approval in each market plus local validation on that hospital's own scanner and patient mix.

How does DAMO RADAR compare with a human radiologist?

In the reader study, 26 radiologists from 14 centers read the same cases, and DAMO RADAR scored higher than 23 of them. The more practical result is the assisted one: when radiologists read with RADAR's output in front of them, their diagnostic sensitivity rose by roughly 10%, which points at a second-reader role rather than a replacement.

What scans does DAMO RADAR actually handle?

DAMO RADAR is built for contrast-enhanced abdominal CT only, covering 18 anatomical structures. It was not trained on acute abdominal presentations, yet still reached 0.904 AUC on them. On the four cancers confirmed by pathology — liver, pancreas, stomach and colorectum — it scored between 0.891 and 0.984 AUC.

Where do I get the weights and how big is the download?

The RADAR checkpoints live on Hugging Face under the radar-generalist account, alongside a RADAR-auxiliary-data dataset and English and Chinese BERT tokenizers. The repository ships download_checkpoints.py and download_auxiliary_data.py so you do not fetch them by hand; the pre-trained RADAR checkpoint plus RADAR+ from-scratch and fine-tuned variants are all published there.

Alibaba DAMO Academy · 2026-09-18 · seismic

DAMO RADAR — Alibaba's open CT model beats most radiologists

DAMO RADAR is an open vision-language model that reads contrast-enhanced abdominal CT scans and flags 146 findings across 18 organs. It averaged 0.913 AUC on about 40,000 real-world exams.

GitHub repository card for alibaba-damo-academy/damo-radar

An open abdominal-CT model that names 146 findings in one pass, published in Science with weights on Hugging Face.

Key specs

Mean auc	0.913
Exams evaluated	~40,000

Quick facts

Maker	Alibaba DAMO Academy
Code license	Apache-2.0
Weights license	CC BY-NC-SA 4.0 (research only)
Input	Contrast-enhanced abdominal CT
Coverage	146 findings across 18 organs
Training data	420,000 exams, 15M image-text pairs
Availability	GitHub + Hugging Face

What is it?

DAMO RADAR reads a contrast-enhanced abdominal CT scan and flags 146 different findings across 18 organs in a single pass, including liver, pancreas, stomach and colorectal cancers. Alibaba's DAMO Academy published the work in Science and put the code on GitHub under Apache-2.0 the next day. Most medical imaging models handle one disease or one organ; this one is a generalist.

How does it work?

Instead of hand-labelled scans, the model learned from 420,000 contrast-enhanced abdominal CT examinations paired with the radiology reports doctors had already written. DAMO Academy turned those reports into 15 million anatomy-aware image-text pairs, so the vision-language model ties each phrase to the organ it describes without manual annotation. That is what lets one model cover 146 findings rather than a narrow set.

Why does it matter?

Radiology departments are short-staffed and abdominal CT is one of the highest-volume studies. In a reader study with 26 radiologists from 14 centers, RADAR outperformed 23 of them, and having RADAR assist raised the readers' diagnostic sensitivity by about 10%. Because the weights are public, hospitals and researchers can test that claim on their own data instead of taking a vendor's word for it.

Who is it for?

medical imaging researchers and clinical AI teams

Frequently asked questions

Can I use DAMO RADAR commercially?: Not the weights. DAMO RADAR splits its licensing: the code on GitHub is Apache-2.0, but the checkpoints on Hugging Face are CC BY-NC-SA 4.0, which allows research and requires share-alike attribution while ruling out commercial deployment. A company wanting to ship RADAR inside a product would need a separate agreement with Alibaba DAMO Academy.
Is DAMO RADAR cleared for clinical use?: No. DAMO RADAR has no clearance under the Software as a Medical Device pathway, so it is a research release rather than a diagnostic product. The Science study measures how it performs against radiologists on retrospective exams; using it on live patients would require regulatory approval in each market plus local validation on that hospital's own scanner and patient mix.
How does DAMO RADAR compare with a human radiologist?: In the reader study, 26 radiologists from 14 centers read the same cases, and DAMO RADAR scored higher than 23 of them. The more practical result is the assisted one: when radiologists read with RADAR's output in front of them, their diagnostic sensitivity rose by roughly 10%, which points at a second-reader role rather than a replacement.
What scans does DAMO RADAR actually handle?: DAMO RADAR is built for contrast-enhanced abdominal CT only, covering 18 anatomical structures. It was not trained on acute abdominal presentations, yet still reached 0.904 AUC on them. On the four cancers confirmed by pathology — liver, pancreas, stomach and colorectum — it scored between 0.891 and 0.984 AUC.
Where do I get the weights and how big is the download?: The RADAR checkpoints live on Hugging Face under the radar-generalist account, alongside a RADAR-auxiliary-data dataset and English and Chinese BERT tokenizers. The repository ships download_checkpoints.py and download_auxiliary_data.py so you do not fetch them by hand; the pre-trained RADAR checkpoint plus RADAR+ from-scratch and fine-tuned variants are all published there.

Try it

git clone https://github.com/alibaba-damo-academy/damo-radar && cd download_scripts && python download_checkpoints.py