Predicting Helicobacter pylori Antibiotic Resistance from Routine Hematoxylin and Eosin Histopathology Using a...

Area of Science:

• Computational pathology
• Medical artificial intelligence
• Infectious disease diagnostics

Background:

• Helicobacter pylori (H. pylori) eradication is increasingly challenged by antibiotic resistance, particularly to clarithromycin (CLA) and fluoroquinolones (FQs).
• Current susceptibility testing methods are resource-intensive and not widely accessible.
• The potential for H&E histopathology to reveal indirect markers of H. pylori resistance remains unexplored.

Purpose of the Study:

• To develop and evaluate a deep learning framework for predicting H. pylori antibiotic resistance directly from H&E stained gastric biopsy whole-slide images (WSIs).
• To assess the feasibility of using routine histopathology as a surrogate for susceptibility testing.

Main Methods:

• A weakly supervised deep learning framework was created to predict H. pylori phenotypic resistance from WSIs.
• Vision-language models (VLMs) filtered image patches for quality and tissue type (gastric surface/gland-neck epithelium).
• Retained patches were encoded using a histopathology-specific vision foundation model (VIRCHOW2) and aggregated for slide-level prediction, with PCR-confirmed resistance phenotypes as ground truth.

Main Results:

• The framework achieved high accuracy in an independent test set: AUC of 0.903 for CLA resistance and 0.959 for FQs resistance.
• At a default threshold, sensitivity was 85.2% for CLA and 98.1% for FQs, indicating strong screening potential.
• Optimized thresholds improved specificity to 95.9% for CLA and 93.9% for FQs, enabling risk stratification.

Conclusions:

• Routine H&E stained gastric biopsies contain morphological features indicative of H. pylori resistance phenotypes that can be leveraged by deep learning.
• This AI-driven histopathology approach offers a complementary tool for resistance risk stratification, particularly beneficial in resource-limited settings.
• Further prospective validation is recommended.