Objective:

To evaluate the performance of endoscopic image-based classification models for H. pylori status in patients with established gastric cancer.

Approach:

• Study Design: Retrospective analysis of 602 endoscopic images from 337 gastric cancer patients, with 576 images retained for model development and evaluation.
• H. pylori Status Definition: Defined using routine clinical testing, including serum anti-H. pylori IgG and/or the 13C urea breath test.
• Model Evaluation: Endoscopic image classification models were evaluated using repeated stratified train, validation, and test splits, with the primary performance metric being the area under the receiver operating characteristic curve (AUC).

Key Findings:

• The best-performing model achieved a median test AUC of 0.6255.
• Sensitivity was 0.7308, specificity was 0.3714, accuracy was 0.6092, and F1 score was 0.6800.
• Fine-tuning showed a higher median AUC than retraining, but discrimination remained limited.

Interpretation:

Endoscopic image-based classification showed limited ability to discriminate H. pylori status in gastric cancer patients, indicating challenges in image-based assessment.

Limitations:

• The study reflects the heterogeneity of routine clinical endoscopic imaging rather than standardized research image acquisition.
• Device-level acquisition parameters and image enhancement settings could not be systematically reconstructed.

Conclusion:

Image-based assessment of H. pylori status in gastric cancer patients remains challenging and may require more rigorous phenotyping, larger datasets, and validation across clinically diverse populations.