Background Gastric intestinal metaplasia (GIM) is a well-established precancerous lesion and key biomarker for assessing the risk of gastric cancer. Histological classification into complete and incomplete types, along with the Operative Link on Gastric Intestinal Metaplasia (OLGIM), provides valuable prognostic information, with stage III–IV strongly associated with cancer development. Therefore, an accurate diagnosis and appropriate surveillance of GIM are essential. Summary This review highlights the epidemiology, histological subtypes, and staging systems of GIM, and evaluates the performance of endoscopic modalities, including conventional white-light endoscopy (WLE), chromoendoscopy, and image-enhanced endoscopy (IEE). Conventional WLE has limited sensitivity and specificity for detecting GIM. In contrast, IEE techniques and scoring systems, such as the Endoscopic Grading of Gastric Intestinal Metaplasia (EGGIM), improve detection, demonstrate a strong correlation with histology, and enable endoscopic risk stratification. Although these methods increase the diagnostic yield, their accuracy remains dependent on operator experience and training. To address this limitation, artificial intelligence (AI) has attracted significant attention. Recent studies have reported that AI-assisted endoscopy achieves sensitivities and specificities exceeding 90%, often outperforming human endoscopists. AI applications extend beyond detection to real-time segmentation, automated EGGIM scoring, and pathological image analyses. However, challenges remain, including the need for external validation, management of data heterogeneity, generalizability beyond Asian cohorts, and development of explainable AI. Key Messages (i) GIM is a critical biomarker for gastric cancer prevention that requires accurate diagnosis and surveillance. (ii) IEE techniques substantially improve endoscopic detection compared to WLE, although interobserver variability remains. (iii) AI demonstrates excellent diagnostic accuracy for GIM and may help overcome the human limitations in detection and grading. (iv) Future directions include prospective multicenter validation, real-time video-based AI, and integration of explainable and ethically sound AI systems into clinical practice.
Ahn et al. (Thu,) studied this question.