Future Perspective of Artificial Intelligence Diagnostics for Early Barrett's Neoplasia.

Abstract

Background: Barrett's esophagus (BE) represents the only established precursor to esophageal adenocarcinoma. While endoscopic surveillance is a cornerstone of early detection, it remains limited by interobserver variability, sampling error, and variability in diagnostic yield. In recent years, artificial intelligence (AI) has emerged as a promising tool to improve the detection and characterization of neoplastic lesions in BE.

Summary: This review outlines the current landscape and future potential of AI applications in the endoscopic management of BE. Diagnostic systems employing convolutional neural networks and transformer-based architectures have achieved high performance for both lesion detection (CADe) and characterization (CADx), with several models externally validated in multicenter cohorts. The first CE-certified commercial system, CADU™, has further marked the entry of AI into clinical use. Emerging developments include AI tools for infiltration depth estimation, vessel detection during endoscopic submucosal dissection, post-therapeutic surveillance, and procedural quality assessment. Challenges related to generalizability, human-AI interaction, ethical implementation, and regulatory compliance are discussed in the context of clinical translation.

Key messages: (1) AI systems demonstrate high diagnostic accuracy and enable real-time assistance in BE surveillance. (2) In-domain pretrained models and transformer-based systems may improve robustness and adaptability. (3) Clinical applications are expanding beyond diagnostics to therapeutic guidance and posttreatment monitoring. (4) Successful implementation depends on rigorous validation, explainability, and integration into clinical workflows.