AI Can Detect Esophageal Cancer From Chest CT Scans

A new artificial intelligence system, named EAGLE, has demonstrated the ability to detect signs of esophageal cancer from chest CT scans, even when the scans were not initially intended to look for this specific disease. Published in Nature Medicine, the research suggests that AI could unlock valuable health information from millions of existing CT scans, potentially identifying other illnesses without requiring new tests or additional radiation exposure.

Radiologists often encounter "incidental findings" during scans performed for other reasons. For example, a low-dose CT scan for lung cancer screening can also reveal coronary artery calcification, bone density changes, or fatty liver disease. The EAGLE system builds on this concept, applying AI to identify esophageal cancer, a disease notoriously difficult to detect early.

Developed by researchers, EAGLE first locates the esophagus in a 3D CT scan and then analyzes it for potential malignant lesions. In external validation tests involving over 11,500 patients, EAGLE achieved a 90% detection rate for esophageal cancer with 98.5% specificity. While its sensitivity for early-stage cancers (around 60%) and pre-cancerous lesions (around 52.5%) is lower, the AI significantly improved radiologists' performance when used as an assistive tool, increasing average sensitivity from 71.9% to 85.7% and specificity from 79.6% to 91.7%.

The system was also adapted for low-dose lung cancer screening CTs, achieving 88.4% sensitivity and 99% specificity. Given that smoking is a risk factor for both lung and esophageal cancers, integrating EAGLE into existing screening programs could be a practical step. The researchers noted that this approach could potentially serve as a "casual screening" method, identifying esophageal cancer in patients undergoing scans for other conditions.

However, significant caveats remain, particularly for populations outside of East Asia where the AI was trained. Esophageal cancer subtypes and prevalence differ between East Asia and Western countries like Israel. Dr. Arnon Makori, head of imaging at Assuta Medical Centers, emphasized that the system's effectiveness in Western populations, where the disease is rarer and tumor types vary, needs further validation. Additional research is required to confirm the clinical utility of EAGLE for widespread use, especially considering its better performance in men than women and the need for longer follow-up periods in real-world trials.