ALIBABA OPEN-SOURCES AI MODEL DETECTING CANCER, 150 MEDICAL CONDITIONS

Sat 19 September 2026:

Damo Radar analyzes 18 abdominal organs and could help radiologists detect disease faster, while new research shows growing potential for AI-assisted CT diagnosis

Alibaba has open-sourced an artificial intelligence model capable of identifying 146 clinical findings, including cancers and other abnormalities, from abdominal CT scans.

The vision-language model, known as Damo Radar, was developed by Alibaba’s research division, Damo Academy. It analyzes contrast-enhanced CT examinations covering 18 abdominal organs and is designed to identify a broad range of diseases and imaging abnormalities.

Researchers evaluated the system on nearly 40,000 real-world examinations, where it achieved an average area under the curve (AUC) of 0.913 across 146 clinical findings. An AUC of 1.0 represents perfect discrimination between positive and negative cases.

The research team described it as “the world’s first expert-level generalist medical imaging model,” saying the approach could eventually be adapted to other forms of medical imaging.

__________________________________________________________________________

https://whatsapp.com/channel/0029VaAtNxX8fewmiFmN7N22

__________________________________________________________________________

AI Compared With Radiologists

The researchers also assessed Damo Radar alongside 26 radiologists from multiple hospitals.

In the reported comparison, the model’s average performance exceeded that of 23 of the participating radiologists. When radiologists used the AI as an additional diagnostic tool, researchers reported a 10% reduction in missed diagnoses and a more than 30% reduction in diagnosis time.

The model is designed as an aid for clinicians rather than a replacement for radiologists. Its broad coverage is significant because many existing medical AI systems are built to identify a single disease or a narrower group of abnormalities.

Alibaba has also released the RADAR code and pretrained model resources publicly, allowing researchers to examine and build on the technology.

New Research Expands AI-Assisted CT Diagnosis

The Alibaba development comes amid rapid advances in medical imaging AI.

A study published in Nature in 2026 introduced Merlin, a three-dimensional vision-language foundation model designed specifically for abdominal CT interpretation. Researchers trained it using more than 6 million CT images from 15,331 examinations, alongside diagnostic codes and radiology reports.

Merlin was evaluated across hundreds of diagnostic, prognostic and imaging tasks and was tested on more than 44,000 CT examinations from three independent sites and two public datasets. The researchers reported that it outperformed several existing 2D and CT-focused AI models.

Another study published in Nature Communications in August 2026 developed AbdomenNet, a foundation model for detecting 11 acute abdominal conditions using non-contrast CT scans.

The model was trained on more than 103,000 CT examinations. In external testing, it achieved a macro-average AUROC of 0.919 for five emergency conditions. When radiologists received AI assistance, their mean AUROC increased from 0.812 to 0.924, while reading time fell by about 52.5 seconds per case.

AI Still Faces Clinical Limitations

The latest research suggests that AI can improve the speed and consistency of medical-image analysis, but performance in controlled studies does not automatically mean that a system can independently diagnose patients in routine clinical care.

For example, a 2026 Nature Medicine randomized trial involving more than 93,000 chest X-rays found that AI-based prioritization did not significantly reduce the overall time to CT or lung-cancer diagnosis in the studied healthcare pathway. The finding illustrates that even when an AI system performs well on image interpretation, integrating it into real-world clinical workflows can produce different results.

Damo Radar therefore represents part of a broader shift from disease-specific medical AI toward generalist systems capable of examining multiple organs and detecting many different abnormalities in a single scan.

The longer-term challenge will be determining how reliably such models perform across hospitals, populations and imaging equipment, and how they can be incorporated safely into clinical decision-making.

SOURCE: INDEPENDENT PRESS AND NEWS AGENCIES

__________________________________________________________________________

FOLLOW INDEPENDENT PRESS:

WhatsApp CHANNEL
https://whatsapp.com/channel/0029VaAtNxX8fewmiFmN7N22