Skip to main content

Alibaba's AI Spots Esophageal Cancer on Routine CT Scans

Alibaba DAMO Academy has rolled out its fourth cancer-screening AI model—and this one tackles a notoriously tricky cancer. Called DAMO EAGLE, it was developed with Sichuan Cancer Hospital and Sun Yat-sen University Cancer Center. The model can flag esophageal cancer, including early-stage and precancerous lesions, from plain CT scans. No endoscopy. No contrast agents. Just a routine scan that many patients already get for lung cancer screening.

The research was validated on 80,000 cases across three countries, and the paper published in Nature Medicine on September 22.

Why this matters

China accounts for nearly half of the world's esophageal cancer cases and deaths. Most patients are diagnosed late, when curative treatment is no longer an option. But if caught early, the five-year survival rate can exceed 95% with minimally invasive surgery.

The standard screening tool is upper gastrointestinal endoscopy. It works, but it's invasive, uncomfortable, and hard to scale—especially in regions with limited medical resources. Plain chest CT, on the other hand, is cheap, quick, and widely accepted. Its scan range naturally covers the esophagus. So why not use it to spot esophageal cancer too?

The catch: the esophagus is a long, collapsible tube. Early lesions hide in the epithelial and mucosal layers, and the heart's constant beating blurs the picture. Even skilled radiologists struggle to see anything suspicious on a plain CT.

Image

Cracking the "impossible"

Yao Jiawen, an algorithm expert at DAMO Academy, explained the team's workaround. They matched endoscopic reports—which pinpoint lesions precisely—with enhanced CT images, then mapped those findings onto plain CT scans. This trained the AI to catch early lesions that the human eye misses.

The numbers speak for themselves. In opportunistic screening settings, DAMO EAGLE achieved 90% sensitivity for esophageal cancer. For precancerous lesions, where signals are faint, sensitivity reached 52.5%. In real-world use, specificity hit 99.2%—meaning very few false alarms. And it performs just as well on low-dose CT, so it can slot right into existing lung cancer screening programs without adding any burden to patients.

From detection to action

Wang Qifeng, chief physician at Sichuan Cancer Hospital's Radiotherapy Department, pointed out that western China—especially Sichuan—has a high incidence of esophageal cancer, partly due to diets heavy in hot, salty foods. Endoscopic screening rates lag far behind need. "AI isn't replacing endoscopy," Wang said. "It's identifying high-risk individuals who then go for endoscopy. That makes the exam more targeted and boosts detection rates." More importantly, it pushes prevention earlier, opening a wider window for cure.

A growing portfolio

With models for pancreatic cancer, gastric cancer, colorectal cancer, esophageal cancer, and aortic dissection already released, DAMO Academy has built a "plain CT + AI" multi-check pipeline. The team has now published five papers in Nature Medicine. The long-term vision: a single plain CT scan that screens for China's top seven deadliest cancers.

Key Points:

  • DAMO EAGLE detects esophageal cancer from plain CT scans—no endoscopy or contrast agents needed.
  • Validated on 80,000 cases across three countries; 90% sensitivity, 99.2% specificity.
  • Works on low-dose CT; can be integrated into existing lung cancer screening.
  • AI flags high-risk patients for endoscopy, improving early detection and survival.
  • Part of Alibaba's broader "plain CT + AI" platform, with five Nature Medicine papers to date.