Helicobacter pylori (Hp) infection remains the leading cause of chronic gastritis and gastric cancer worldwide. Asian studies described specific gastric mucosal patterns observed by electronic chromoendoscopy, the “spotty” and “cracked” patterns, which are associated with active and past Hp infection, respectively (1). However, no prospective European study evaluated their diagnostic performance. This study aimed to assess the reproducibility and diagnostic accuracy of mucosal pattern recognition by Blue Light Imaging (BLI) for the identification of active or past Hp infection.

This was a prospective, single-center study conducted between 2023 and 2025. Adult outpatients (≥18 years) undergoing gastroscopy for dyspepsia, gastroesophageal reflux, post-eradication follow-up (≥6 months after therapy), or first-degree family history of gastric cancer were enrolled. An expert endoscopist performed endoscopic examinations. For each patient, five non-magnified BLI images (two antrum, one angulus, two corpus) were prospectively selected and anonymized. Three experienced endoscopists independently evaluated all images, classifying each as normal, spotty, or cracked. A fourth expert adjudicated discordant evaluations. For diagnostic performance, the final pattern assigned to each image was defined when all three or at least two endoscopists agreed; in cases of complete disagreement, the final classification was established after discussion with the fourth expert. Histological assessment, following the updated Sydney System, served as the reference standard. Analyses were performed at the image level. Interobserver agreement was quantified by Fleiss’ κ and pairwise Cohen’s κ. Diagnostic performance was calculated for each mucosal pattern, overall, and stratified by gastric region and proton pump inhibitor (PPI) status. Comparisons between diagnostic performance proportions were conducted using two-sided z-tests.

A total of 120 patients (mean age 54.2 ± 17.0 years; 65.0% female) were enrolled, yielding 600 BLI images, of which 29 were excluded after consensus for insufficient quality, resulting in 571 images. Overall interobserver agreement among the three endoscopists was substantial (Fleiss’ κ = 0.74; p < 0.001), with comparable values in the antrum (κ = 0.73) and corpus (κ = 0.75). Pairwise κ values ranged from 0.72 to 0.77, and the overall proportion agreement was 0.79, confirming high reproducibility across endoscopists. Using histology as the reference, the spotty and cracked patterns showed very high specificities of 98.9% (95% CI 97.5–99.6) and 87.9% (95% CI 84.3–91.0), diagnostic accuracies of 96.1% (95% CI 94.2–97.6) and 86.0% (95% CI 82.9–88.8), and sensitivities of 43.5% (95% CI 34.3–53.0) and 50.6% (95% CI 43.1–58.0), respectively. Conversely, the normal pattern achieved high sensitivity of 95.7% (95% CI 92.5–97.9) but a moderate specificity of 58.5% (95% CI 52.8–64.0), with a diagnostic accuracy of 75.3% (95% CI 71.6–78.8). Regional analyses revealed complementary diagnostic behavior across gastric sites: the normal pattern was more accurately identified in the antrum (72.6%; 95% CI 67.5–77.3) than in the corpus (42.4%; 95% CI 36.0–49.1; p < 0.0001), while the cracked pattern performed better in the corpus (90.8%; 95% CI 86.4–94.1) than in the antrum (83.1%; 95% CI 78.7–87.0; p = 0.0089). When stratified by PPI use, diagnostic performance remained stable for the spotty and cracked patterns, but the normal pattern showed lower accuracy in patients receiving PPIs (53.1%; 95% CI 46.7–59.5) compared with those not receiving PPIs (64.8%; 95% CI 59.3–70.0; p = 0.0048).

This first European prospective study showed substantial interobserver agreement and consistent diagnostic accuracy in BLI-based assessment of gastric mucosal patterns associated with active and past Hp infection. Electronic chromoendoscopy may contribute to defining the infection status, although awareness of PPI use remains important to avoid potential misinterpretation of mucosal patterns.