A Deep Learning Convolutional Neural Network Can Recognize Common Patterns of Injury in Gastric Pathology

Area of Science:

• Gastrointestinal Pathology
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Deep learning (DL) applications in pathology predominantly focus on neoplastic diseases, leaving inflammatory conditions under-researched.
• Nonneoplastic gastric pathologies, such as H. pylori gastritis and reactive gastropathy, represent a significant diagnostic challenge.

Purpose of the Study:

• To evaluate the efficacy of deep learning algorithms in diagnosing nonneoplastic conditions in gastric biopsies.
• To assess the performance of a convolutional neural network (CNN) as a diagnostic aid for H. pylori gastritis and reactive gastropathy.

Main Methods:

• A deep learning model was trained and tested on digitized gastric biopsy images, including normal, H. pylori-infected, and reactive gastropathy cases.
• Phase 1 involved 300 classic cases with pathologist-established gold standard diagnoses, while Phase 2 tested 106 consecutive nonclassical cases.
• Area distribution (AD) percentages were used to correlate DL algorithm output with diagnoses, analyzed via receiver operating curves (ROC).

Main Results:

• Phase 1 demonstrated high diagnostic accuracy, with Area Under the Curve (AUC) values of 99.7% for normal, 100% for H. pylori, and 99.9% for reactive gastropathy at optimal AD cutoffs.
• Phase 2 showed slightly reduced but still strong performance, with AUCs of 91.9% (normal), 100% (H. pylori), and 94.0% (reactive gastropathy).
• High sensitivity and specificity were achieved for H. pylori gastritis in both phases, indicating robust detection capabilities.

Conclusions:

• A convolutional neural network serves as an effective screening tool and diagnostic aid for H. pylori gastritis.
• Deep learning holds significant potential for improving the accuracy and efficiency of diagnosing nonneoplastic gastric pathologies.