A new meta-analysis of 14 studies found that deep learning (AI) systems can detect early esophageal squamous cell carcinoma (ESCC) more accurately than junior endoscopists and perform on par with experienced specialists. Across the studies, AI achieved 94% sensitivity and 88% specificity, significantly outperforming less experienced physicians while matching senior endoscopists in overall diagnostic accuracy.
The analysis also showed that AI serves as an effective clinical support tool. When junior endoscopists used AI assistance, their diagnostic performance improved substantially, while senior endoscopists also experienced modest gains in sensitivity and accuracy. These findings suggest AI could help standardize detection of subtle early esophageal lesions and reduce variability between operators, particularly in settings with less experienced clinicians.
Researchers noted that AI performed consistently across different lesion sizes and showed particularly strong detection of cancers located in the upper esophagus. However, most of the included studies were single-center, retrospective studies conducted in Asian populations, which may limit how broadly the results apply. Differences in AI model design and validation methods also make direct comparisons challenging.