Alibaba Group’s DAMO Academy has presented two artificial intelligence models for medical imaging, backed by peer-reviewed studies in Nature Medicine and Science.

One system is designed to identify esophageal cancer and malignant precancerous lesions from routine noncontrast chest computed tomography scans. The other is a generalist model that evaluates contrast-enhanced abdominal CT images for 146 findings across 18 anatomical structures.

The research indicates how AI could extract more diagnostic value from scans already performed in hospitals and screening programs. The results do not amount to regulatory approval or a clinical deployment, however, and both systems still require medical oversight and further evaluation.

EAGLE analyzes routine chest CT scans

The Esophageal AI-Guided malignant Lesion Evaluation model, or EAGLE, was developed to flag high-risk patients using noncontrast CT images that include the esophagus. Early malignant changes can be difficult to see because the organ can collapse and is affected by motion from nearby structures.

According to the Nature Medicine study, EAGLE was trained on scans from 6,813 patients at two centers and validated across 12 centers in China, the Czech Republic and Australia. The full validation program involved 80,612 patients in hospital, low-dose lung-screening and population-screening settings.

In external tests covering 11,466 patients at eight centers, the model recorded 98.5 percent specificity, 90 percent sensitivity for cancer and 52.5 percent sensitivity for malignant precancerous lesions. A separate low-dose CT evaluation produced comparable results, supporting the possibility of adding esophageal risk screening to existing lung-cancer screening workflows without another imaging procedure.

The researchers also calibrated an enhanced version, EAGLE-Plus, against a real-world cohort of 35,402 patients. In a prospective hospital cohort of 17,446 scans, the study reported a positive predictive value of 42.2 percent. That measure is important because even a model with high specificity can produce false positives when used to screen a population in which the disease is uncommon.

Radiologists improved with AI assistance

A reader study involving 17 radiologists found that EAGLE assistance increased average sensitivity from 71.9 percent to 85.7 percent and specificity from 79.6 percent to 91.7 percent. The improvement was especially marked for difficult early-stage cases.

The authors said EAGLE could be used as a risk-stratification layer to identify people who should receive an endoscopic examination, rather than replacing endoscopy or a clinician’s diagnosis. The study was funded partly by DAMO Academy and included researchers from Alibaba, Chinese cancer centers and partner institutions overseas.

The model’s underlying code is protected by patents and has not been released publicly. The paper documents the methods and provides a de-identified sample dataset and interactive portal, but the absence of public code limits independent reproduction of the complete system.

RADAR covers 146 abdominal findings

DAMO also introduced Rapid Abdominal Diagnosis with AI and Radiology, or RADAR, a vision-language model trained on more than 400,000 contrast-enhanced abdominal CT examinations and 15 million anatomy-aware image-text pairs.

The Science study reported that RADAR learned from clinical radiology reports without manual pixel-level annotation. Across internal and external evaluations, it assessed 18 anatomical structures and 146 findings. In a reader study, assistance from the model improved the diagnostic sensitivity of 26 radiologists by about 10 percent.

DAMO has made RADAR resources available through an official GitHub repository, but its documentation labels the system as research-only and says further prospective clinical studies are needed before direct deployment.

Medical AI still faces a deployment gap

The two studies add to Alibaba’s work on AI for pancreatic, gastric, colorectal and liver-cancer imaging. They also reflect a wider push across Asia to use AI to extend specialist capacity and find disease earlier, an area examined in a TNGlobal INSIDER contribution on AI and cancer detection across the region.

For health systems, the potential value lies in using existing scans to prioritize follow-up care. Whether that promise translates into routine practice will depend on prospective validation, local regulatory review, workflow integration and evidence that performance holds across different populations and equipment.

Convergence in the blood: How AI and biotech are rewriting cancer detection across Asia