A Multisite Fusion-Based Deep Convolutional Neural Network for Classification of Helicobacter pylori Infection Status...

Area of Science:

• Medical Artificial Intelligence
• Gastroenterology Imaging
• Computational Pathology

Background:

• Helicobacter pylori (Hp) infection is a significant global health concern.
• Accurate diagnosis of Hp infection status (uninfected, past, current) is crucial for effective treatment.
• Current diagnostic methods have limitations, necessitating advanced approaches.

Purpose of the Study:

• To develop a deep convolutional neural network (DCNN) for classifying Hp infection status.
• To integrate features from multiple gastric sites for improved diagnostic accuracy.
• To compare the efficacy of single-site versus multisite DCNN models and expert endoscopists.

Main Methods:

• Ten deep learning architectures were trained on 3380 white-light images from 676 subjects across eight centers.
• External validation and testing were performed on separate datasets.
• Single-site and multisite fusion DCNN models were developed and compared, including a voting-based multisite fusion approach.

Main Results:

• The top-performing single-site DCNN model (Wide-ResNet) achieved high accuracy and AUC for all infection categories.
• The voting-based multisite fusion DCNN model demonstrated superior accuracy and AUC, particularly for noninfection and current infection.
• The DCNN models showed enhanced sensitivity, specificity, and precision compared to experienced endoscopists.

Conclusions:

• A voting-based multisite fusion DCNN model effectively classifies Hp infection status.
• The developed DCNN excels in distinguishing between uninfected and currently infected individuals.
• This AI-driven approach offers a promising tool for Hp infection diagnosis.