A Deep Learning Model for Diagnosing Gastric Mucosal Lesions Using Endoscopic Images: Development, Validation, and Method Comparison

Abstract

Background: Endoscopic differential diagnoses of gastric mucosal lesions (benign gastric ulcer, early gastric cancer [EGC], and advanced gastric cancer) remain challenging. We aimed to develop and validate convolutional neural network-based artificial intelligence (AI) models: lesion detection (AI-LD), differential diagnosis (AI-DDx), and invasion-depth (AI-ID, pT1a vs. pT1b among EGC) models. Methods: This study included 1,366 consecutive patients with gastric mucosal lesions from two referral centers in Korea. One representative endoscopic image from each patient was used. Histological diagnoses were set as the gold standard. The performances of the AI-DDx (training/internal/external validation set, n=1009/112/245) and AI-ID (training/internal/external validation set, n=620/68/155) were compared with visual diagnoses by independent endoscopists (stratified by novice [ 5 years of experience]) in a prospective manner and by endoscopic ultrasonography (EUS), respectively. Findings: The AI-DDx showed good diagnostic performance for both internal (area under of the receiver operating characteristic curve [AUROC]=0.86) and external validation (AUROC=0.86). The performance of the AI-DDx was better than that of the novice (AUROC=0.82, P=0.01) and intermediate endoscopists (AUROC=0.84, P=0.02), but was comparable to the experts (AUROC=0.89, P=0.12) in the external validation set. The AI-ID showed fair performances in both internal (AUROC=0.78) and external validation sets (AUROC=0.73), which were significantly better than EUS results performed by experts (internal validation: AUROC=0.62, external validation: AUROC=0.56; both P <0.001). Interpretation: The AI-DDx was comparable with experts and outperformed novice and intermediate endoscopists for the differential diagnosis of gastric mucosal lesion. The AI-ID performed better than EUS for the invasion-depth evaluation. Funding Statement: Seoul National University Hospital (No. 04-2020-0860). Declaration of Interests: The authors confirm that this article content has no conflict of interest. Ethics Approval Statement: The present study was approved by the institutional review boards of the two participating institutions (IRB No.: 1706-109-859 [SNUH] and 2020-08-135 [SMC]) and conducted following the ethical guidelines of the World Medical Association Declaration of Helsinki. Informed patient consent was waived by the institutional review board of each institution owing to the retrospective nature of the study.