Alibaba DAMO Academy Unveils AI Model for Early Esophageal Cancer Detection

Alibaba DAMO Academy, in collaboration with medical institutions, developed DAMO EAGLE, an AI model that identifies early esophageal cancer and precancerous lesions from non-contrast plain CT scans. Published in Nature Medicine, the model was validated on more than 80,000 cases across three countries without requiring invasive endoscopy. Early-stage esophageal cancer carries a 5-year survival rate exceeding 95%, yet standard screening via endoscopy is invasive and difficult to deploy broadly. By enabling non-invasive opportunistic screening during routine low-dose CT scans, such as those used for lung cancer, DAMO EAGLE allows early risk detection without extra patient burden or procedural risk. DAMO EAGLE achieved 90% sensitivity for esophageal cancer and 52.5% sensitivity for subtle precancerous lesions, along with a 99.2% specificity in real-world settings. To overcome the technical challenge of detecting mucosal lesions in a collapsed organ, researchers trained the AI by pairing detailed endoscopy reports and contrast-enhanced CT images with non-contrast CT scans.