H. pylori Related Atrophic Gastritis Detection Using Enhanced Convolution Neural Network (CNN) Learner

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Atrophic gastritis (AG), often caused by *Helicobacter pylori* (*H. pylori*) infection, can progress to gastric cancer, a leading cause of cancer mortality.
• Accurate and early detection of AG is critical for preventing severe outcomes.
• Existing deep learning models for *H. pylori*-associated AG detection face challenges with increasing network depth and accuracy.

Purpose of the Study:

• To develop an enhanced deep convolutional neural network (DCNN) model for accurate binary classification of normal versus atrophic gastritis in the gastric antrum.
• To improve diagnostic accuracy and overcome limitations of existing DCNN architectures in detecting *H. pylori*-associated AG.
• To integrate advanced feature extraction and selection techniques for robust disease identification.

Main Methods:

• Utilized a DCNN incorporating pooling and channel shuffle for improved training of deeper networks.
• Employed Canonical Correlation Analysis (CCA) for feature fusion from pre-trained ShuffleNet models.
• Applied ReliefF feature selection and Generalized Additive Model (GAM) for final classification.

Main Results:

• The proposed enhanced DCNN model achieved a testing accuracy of 98.2%.
• The integration of CCA and ReliefF effectively fused and selected relevant features for classification.
• The model demonstrated superior performance in distinguishing between normal and atrophic gastritis.

Conclusions:

• The developed deep learning approach provides a highly accurate method for diagnosing *H. pylori*-associated atrophic gastritis.
• This technique offers a promising upgrade to current diagnostic standards, potentially reducing risks associated with untreated AG.
• The study highlights the potential of advanced AI techniques in early cancer detection and prevention.