Alibaba open-sources AI model that detects cancer across 150 abdominal conditions

Alibaba ↗ 4.33% 's research division, Damo Academy, has released an open-source artificial intelligence model capable of identifying nearly 150 abdominal conditions from CT scans, including malignant tumors, in a development that could reshape how hospitals approach radiology diagnostics worldwide.

A Generalist Model for Abdominal Imaging

The model, formally called RADAR — Rapid Abdominal Diagnosis with AI and Radiology — was detailed in a study published in Science on Wednesday. Unlike most medical AI tools that target a single condition, RADAR evaluates 146 clinical findings across 18 abdominal organs simultaneously, analyzing contrast-enhanced CT scans by breaking them into individual anatomical structures and linking them to corresponding diagnostic descriptions.eurekalert

Trained on 424,911 examinations containing more than 1.5 million image-text pairs and over 15 million anatomy-specific pairs, the model uses a vision-language contrastive-learning approach to connect imaging findings with diagnoses. In nearly 40,000 real-world examinations, RADAR achieved a mean area under the curve score of 0.913, compared with 0.776 for the best competing vision-language model. It also performed well in emergency settings it was never trained on, reaching an AUC of 0.904 across more than 27,000 emergency CT cases.eurekalert

Outperforming Most Radiologists

In a study involving 26 radiologists from multiple hospitals, RADAR outperformed 23 on average, according to the South China Morning Post. When radiologists used the model as an assistant, missed diagnoses dropped by 10 percent and diagnosis time fell by more than 30 percent. Testing across eight external medical centers showed the model maintained high accuracy with an AUC of 0.895, demonstrating generalization across diverse patients and clinical protocols.scmp+2

The research team described RADAR as "the world's first expert-level generalist medical imaging model" and said its training method could be extended to other types of medical imaging.scmp

Open Access and Clinical Implications

By open-sourcing the model, Alibaba allows hospitals, researchers, and developers globally to inspect and build upon the technology without licensing fees — a departure from the typical approach of guarding diagnostic algorithms as proprietary assets. The release is aimed at integration into existing radiology departments, particularly in under-resourced settings where specialist radiologists are scarce. Damo Academy has been expanding its medical AI portfolio in recent months, having also released DAMO LiON, a liver cancer diagnostic model, in August, and COCA, a colorectal cancer detection tool, earlier this year.yenisafak+3