Bayesian-Optimized Deep Learning for Helicobacter pylori Detection in Egyptian Gastric Histopathology: A Multi-Architecture Study

Helicobacterpylori (H. pylori) infection is a significant health issue. It is classified by the World Health Organization as a Group I carcinogen. Egypt has unusually high prevalence rates of 64.6-90.3%, significantly above the global averages. Histopathological diagnosis is tedious and time-consuming and has high interobserver variation, especially with small, scattered bacterial colonies. This study represents the first comprehensive application of artificial intelligence to this diagnostic task. We compared eight state-of-the-art architectures. The models were trained for 50 epochs and assessed on 100 histopathological whole-slide images using metrics including accuracy, sensitivity, specificity, and AUC-ROC. MobileNetV2 and Xception performed perfectly with 100% accuracy, sensitivity, and specificity, an AUC of 1.0, and zero misclassifications. MobileNetV2 was trained 2.3 times faster, offering an efficiency advantage relevant to deployment in resource-limited pathology laboratories, pending external validation. Another four models (DenseNet-201, InceptionV3, VGG-16, and VGG-19) recorded 100% test accuracy, while Vision Transformer recorded 96.97%. These results represent retrospective patch-level classification performance and do not constitute clinical diagnostic validation. This study not only establishes the first validated artificial intelligence classification system of H. pylori in Egypt but also shows high performance on Egyptian samples and offers a solid methodological groundwork for future medical AI applications in resource-constrained contexts.