AI-ASSISTED DETECTION OF GASTRIC INTESTINAL METAPLASIA: DESIGN AND VALIDATION OF THE INTELLIMETA ALGORITHM

Background:Gastric intestinal metaplasia(GIM) is a potential precancerous lesion that significantly increases the risk of gastric cancer. Its accurate detection requires expertise in digestive pathology and remains challenging due to histological complexity and interobserver variability. Artificial intelligence (AI) represents a promising tool to support early and precise diagnosis.

Methods: We developed IntelliMeta, an AI-based algorithm designed to automatically detect GIM on digitized gastric biopsy slides. A dataset of 229 histological slides (173 normal, 56 with GIM) collected at the Hassan II Regional Hospital of Agadir was digitized using an APERIO LV1 scanner. After expert annotation, a total of 902 histological images were processed. The algorithm, based on a Visual Geometry Group (VGG) transfer learning model, was trained and validated using data preprocessing, augmentation, and cross-validation. Key functionalities include automatic segmentation, multi-region quantification, and binary classification (focal vs diffuse GIM).