Artificial intelligence for endoscopic grading of gastric intestinal metaplasia: advancing risk stratification for gastric cancer

Background and study aims. Endoscopic Grading of Gastric Intestinal Metaplasia (EGGIM) correlates with histological assessment of GIM and enables stratification of risk to gastric cancer. We aimed at developing and evaluating an Artificial Intelligence (AI) approach for EGGIM estimation. Materials and methods. Two datasets (A and B) with 1280 Narrow-Band Imaging images were used for per-image analysis. Still images with manually selected patches of 224x224, annotated by experts, were used. Dataset-A was retrospectively collected from clinical routine, while Dataset-B (used for per-patient analysis) was prospectively collected and included 65 fully documented patients. To mimic clinical practice, a Deep Neural Network classified image patches into 3 EGGIM classes (0,1,2) and calculated overall EGGIM per-patient (0–10). Results. On per-image analysis, a balanced accuracy of 87% (95% confidence interval [CI], 71-100%) was obtained. Per-patient EGGIM estimation had an average error of 1.15 (out of 10) and showed 88% (95%CI, 80–96%) accurate clinical decisions for surveillance (EGGIM≥5), with 85% (95%CI, 75–94%) specificity but 0% false negatives rate. The positive and negative predictive values of this estimation were 62% (95%CI, 32–92%) and 100% (95%CI, 100–100%), respectively. Conclusions. We demonstrated the potential of applying AI tools in endoscopic image analyses, by estimating EGGIM with high accuracy. This opens the possibility for automated assessment of EGGIM, providing a greener strategy for gastric cancer risk stratification but also for prospective studies and interventional trials.

Publication History

Received: 24 February 2025

Accepted after revision: 27 May 2025

Accepted Manuscript online:
17 July 2025