Alibaba Group’s research arm, Damo Academy, has released an open-source AI model that spots nearly 150 abdominal diseases from CT scans. The model, named Damo Radar, works with contrast-enhanced CT images of 18 abdominal organs. It flags a wide range of conditions and abnormalities, among them malignant tumors.
Damo Radar is a vision-language model. It was trained on CT scans matched with clinical reports so it can connect image details to text descriptions. Researchers called it the world’s first expert-level generalist medical imaging model. Unlike tools limited to one disease or organ, this one covers a broad set of findings. The same training approach could later apply to other forms of medical imaging.
Performance Results
Tests used nearly 40,000 real-world examinations. Across 146 clinical findings the model reached an average area under the curve of 0.913. An AUC of 1.0 would mean perfect accuracy.
A separate evaluation involved 26 radiologists from different hospitals. Damo Radar outperformed 23 of them on average. When the radiologists used the model as an assistant, missed diagnoses fell by 10 percent and reading time dropped by more than 30 percent.
Open-Source Release
Making the model open source lets researchers, smaller hospitals, and independent developers examine the code, test it, and build on it. The goal is straightforward: help existing radiology teams interpret complex abdominal scans more effectively.
Damo Academy has earlier work in this area. Its Damo LiON model for liver cancer detected 15 malignant tumors that physicians had missed in a prospective clinical trial.