Measurement of Early Gastric Cancer Size Using a Computer Vision-Assisted Virtual Ruler: A Prospective Clinical Study

BACKGROUND: Accurate measurement of early gastric cancer (EGC) size is important for clinical management. We evaluated a computer vision-assisted virtual ruler (VR) for measuring lesion size during endoscopy. METHODS: This prospective clinical trial included 78 patients who underwent endoscopic submucosal dissection (ESD) from January to December 2024. The maximum diameter of the post-ESD specimen, measured with a physical ruler, was used as the reference. VR measurements obtained during ESD and preoperative visual estimates were compared with the reference using mean absolute error (MAE), percentage relative error (PRE), Bland-Altman analysis, and the intraclass correlation coefficient (ICC). Lesion size was also compared between patients with high-grade intraepithelial neoplasia (HGIN) and adenocarcinoma. RESULTS: Visual estimation underestimated lesion size in the overall cohort and in the HGIN and adenocarcinoma groups, whereas VR measurements did not differ significantly from the reference. MAE was lower with VR than with visual estimation (1.00 vs. 3.00 mm, P=0.001), as was PRE (3.28% vs. 7.89%, P=0.001). The difference was most apparent for flat lesions and adenocarcinoma. VR also showed narrower Bland-Altman limits of agreement (-4.23 to 4.19 mm vs. -3.85 to 9.49 mm) and a higher ICC (0.979 vs. 0.909, P<0.001). CONCLUSION: The computer vision-assisted VR showed better agreement with the reference measurement and less error than visual estimation for EGC lesion size.