BoostedNet: A decision support model for the diagnosis of helicobacter pylori from gastric histopathology images

Helicobacter pylori (H. pylori), a bacterium residing in the human stomach, can lead to various gastric diseases, including severe gastric cancer. Microscopic examination of gastric biopsy tissue slides is considered the gold standard method for confirmation of the presence of the H. pylori bacterium. Even so, manual examination of an overwhelming inflow of tissue slides by pathologists is prone to misdiagnosis of tissue slides. A computerized framework for histopathological image analysis is envisioned as a diagnostic tool that alleviates pathologists' encumbrance, enhancing their ability to accurately carry out annotative diagnoses of abnormalities in gastric histopathology. Transfer learning with Convolutional Neural Networks (CNNs) has demonstrated positive prospects in the analysis of histological images. However, interpretability of the decision-making process in transfer learning is challenging, as features learned from the source task may not directly transferable to the target task. Hence, for the pursuits of an indigenous simplistic model portends a promising solution for detection of the H. pylori bacterium. We propose BoostedNet, a novel 6-layer CNN model integrated with an Extreme Gradient Boost algorithm to classify gastric histopathological images as H. pylori-positive or negative. Performance evaluation of the BoostedNet model was appraised using open gastric histopathological image datasets. A remarkable 99% accuracy was achieved in the classification of (H&E)-stained gastric histopathology images. Besides, the proffered generic BoostedNet model demonstrated consistent performance across Giemsa stained dataset, highlighting its reliability and robustness. These findings demonstrate that the recommended BoostedNet model outperformed the baseline CNN model with a (1-10)% performance enhancement over diverse set of datasets.