AI Model Spots Esophageal Cancer On Routine Chest CT Scans
A new artificial intelligence model published in Nature Medicine on September 22, 2026 can spot esophageal cancer on routine chest CT scans with 90.0% sensitivity and 98.5% specificity, using imaging most patients already have on file.
Researchers trained an AI tool called EAGLE on CT scans from 6,813 patients at two Chinese cancer centers, then validated it across 12 centers in three countries covering 80,612 patients. In an external test group of 11,466 patients across 8 centers, EAGLE detected 90.0% of esophageal cancers and 52.5% of precancerous lesions, called high-grade intraepithelial neoplasia, with 98.5% specificity. Performance was similar on low-dose CT used in lung-cancer screening, tested in 1,607 patients. After recalibration on real-world data from 35,402 patients, false positives fell 72.7%, from 432 to 118 cases, without losing sensitivity. In prospective testing of 17,446 hospital patients, a positive AI flag matched true cancer 42.2% of the time, and specificity reached 99.94% in a real-world low-dose screening group of 10,959 patients.
EAGLE AI Model: 90% Sensitivity vs 98.5% Specificity
Esophageal cancer usually causes no symptoms until a tumor narrows the swallowing passage, and standard endoscopy is invasive, so most people are never screened. The esophagus is a narrow, collapsible tube that moves with each heartbeat, making early lesions nearly invisible on a plain chest CT. The authors report that the model was trained on CT scans with expert-annotated lesion masks and confirmed diagnoses, teaching it to detect subtle lesions that are easily missed on standard chest CT. Because chest CT already accounts for 40% of all CT scans, a validated tool like this could add cancer detection to scans people already get for lung or heart reasons, without new radiation or procedures.
All cohorts came from hospitals in China, with added testing in the Czech Republic and Australia; performance in other populations has not been established. A positive AI flag was confirmed as true cancer only 42.2% of the time in prospective testing, so more than half of flagged patients still needed follow-up imaging or endoscopy, and whether earlier AI detection improves survival has not been reported.
Data Panel
- Who
- 6,813 patients trained the model at two Chinese cancer centers; validation covered 80,612 patients across 12 centers in three countries, including an external test group across 8 centers, n = 11,466 patients.
- Design
- Multicenter AI model development and validation study using noncontrast chest CT, with internal and external testing, low-dose CT validation, real-world recalibration and prospective hospital testing.
- Dose
- Not applicable; this is an AI diagnostic model applied to standard noncontrast chest CT and low-dose CT scans, not a drug or supplement.
- Primary result
- In external testing across 8 centers with 11,466 patients, EAGLE detected 90% of esophageal cancers and 52.5% of precancerous lesions with 98.5% specificity.
- Secondary
- False positives fell 72.7% (432 to 118) after recalibration on 35,402 real-world patients; prospective validation in 17,446 patients showed 42.2% positive predictive value; a real-world low-dose screening group of 10,959 patients reached 99.94% specificity.
- Funding / conflicts
- Not reported in the available text; several authors are affiliated with Alibaba Group's DAMO Academy, which helped develop the AI system.
Dr. Axe's Take
I like that this AI reads scans people already have on file, because esophageal cancer almost never gets caught early any other way. Reflux, excess weight and heavy alcohol use are the drivers I see most often behind this cancer, and all three respond to lifestyle changes you can start now. If you deal with chronic heartburn, work on losing excess weight, eat lighter dinners earlier in the evening, and cut back on alcohol, since all three lower the acid exposure that damages the esophagus over time. Ask your doctor whether AI-enhanced CT screening is available anywhere you already get chest imaging, and keep taking any reflux medication as prescribed while you work on the lifestyle side.
— Dr. Axe