Histopathological Diagnosis System for Gastritis Using Deep Learning Algorithm.

Area of Science:

• Pathology
• Artificial Intelligence
• Medical Imaging

Background:

• Chronic gastritis is a common gastric condition requiring accurate pathological classification.
• Whole-slide images (WSIs) offer comprehensive views of biopsy specimens but pose challenges for manual analysis.
• Developing automated tools is crucial for efficient and accurate diagnosis.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for the pathological classification of chronic gastritis using WSIs.
• To assess the algorithm's diagnostic performance against expert pathologist consensus.

Main Methods:

• A deep learning model (DeepLab v3 with ResNet-50) was trained on 1,008 WSIs and validated on 100 WSIs.
• The algorithm was tested on an independent set of 142 WSIs, comparing its classifications to pathologist consensus.
• Performance was evaluated using receiver operating characteristic (ROC) curves, area under the curve (AUC), sensitivity, specificity, and accuracy.

Main Results:

• The algorithm achieved high AUCs for chronic superficial gastritis (0.882), chronic active gastritis (0.905), and chronic atrophic gastritis (0.910).
• Excellent sensitivity and specificity were observed for all three gastritis types, with overall accuracies ranging from 0.880 to 0.986.
• The overall predicted accuracy for the three types of gastritis was 0.867.

Conclusions:

• The developed deep learning algorithm demonstrates high accuracy in classifying chronic gastritis from WSIs.
• The algorithm can identify and flag suspicious regions, serving as a valuable auxiliary diagnostic tool for pathologists.
• This AI-driven approach has the potential to enhance diagnostic efficiency and accuracy in gastric pathology.