Artificial intelligence-based assessment of gastric mucosal exposure and its association with lesion detection: a multicenter study

Abstract

Completely and clearly visualizing the mucosa during esophagogastroduodenoscopy (EGD) improves high-risk lesion detection. Here we propose a new EGD quality control indicator and present the optimal gastric mucosal exposure area (OGMEA) system to automatically assess mucosal exposure and examine its association with the detection of early gastric cancer and other gastric lesions. A segmentation model was developed to assess mucosal exposure. Expert endoscopists defined a standard for 19 gastric anatomical sites. Images exceeding the OGMEA threshold were considered adequately-exposed. The OGMEA was validated using a multi-center dataset from a randomized controlled trial involving 25,320 participants from 24 centers. Associations between OGMEA and lesion detection were assessed at the anatomical-site and patient levels. Patient-level associations were evaluated using center-clustered multivariable logistic generalized estimating equation models adjusted for trial allocation, the number of screened standard anatomical sites, and inspection time. Of assessed sites, 46.36% and 53.64% were adequately and inadequately exposed, respectively. Detection rates of EGC, gastric neoplasms, precancerous conditions, and focal lesions were significantly higher in adequately-exposed sites than in inadequately-exposed sites ( P <0.001). At the patient level, the number of adequately-exposed sites was positively correlated with lesion detection across all lesion categories (ρ = 0.830-1.000, all P <0.01). In adjusted analyses, each additional adequately exposed site was associated with higher odds of detecting EGC (aOR 1.081, 95% CI 1.021–1.146, P = 0.008), precancerous conditions (aOR 1.056, 95% CI 1.017–1.096, P = 0.004), and focal lesions (aOR 1.059, 95% CI 1.006–1.114, P = 0.028). The association with neoplastic lesions was similar in direction but did not reach statistical significance (aOR 1.038, 95% CI 0.998–1.080; P = 0.060). Effect estimates were materially consistent in endoscopist-clustered sensitivity analyses. OGMEA may serve as a practical quality indicator for EGD, with greater mucosal exposure associated with higher lesion detection.