EAGLE AI detects esophageal cancer on noncontrast chest CT across 12 centers — Nature Medicine

An open-access paper in Nature Medicine dated 22 September 2026 (DOI 10.1038/s41591-026-04656-4) describes EAGLE, an artificial-intelligence model designed to detect esophageal cancer and high-grade intraepithelial neoplasia (HGIN) from routine chest noncontrast computed tomography (CT).

Performance reported

EAGLE was trained on 6,813 patients and validated across 12 centers in three countries, covering 80,612 patients in total. In an external opportunistic screening evaluation across eight centers (n=11,466), the model achieved 98.5% specificity and 90.0% sensitivity for cancer, and 52.5% sensitivity for precancerous lesions. Real-world calibration cut false positives by 72.7%. In a prospective hospital cohort (n=17,446), positive predictive value was 42.2%. On real-world low-dose CT (n=10,959), specificity reached 99.94%. In a reader study of 17 radiologists, AI assistance raised sensitivity from 71.9% to 85.7% and specificity from 79.6% to 91.7%.

Limits stated by the authors

The paper notes that more international validation is needed, especially for esophageal adenocarcinoma and distal disease; that pre-endoscopy triage evidence is partly simulation-based or limited by few prospective positives; that follow-up was under two years in some cohorts; and that endoscopy compliance was suboptimal in places. EAGLE is not a substitute for clinical care. This article is not personal medical advice — talk to a doctor about cancer screening.