Diagnosis of Chest Pneumonia with X-ray Images Based on Graph Reasoning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatrics

Background:

• Pneumonia is a leading infectious cause of death in children globally.
• Early detection of pneumonia in pediatric patients is crucial for effective treatment.
• The World Health Organization reported significant mortality rates due to pneumonia in children under five.

Purpose of the Study:

• To develop and evaluate a deep learning model for binary classification of pneumonia from pediatric chest X-ray images.
• To improve the accuracy and efficiency of pneumonia diagnosis in children.
• To leverage advanced AI techniques for enhanced medical image analysis.

Main Methods:

• A deep learning approach utilizing a convolutional neural network framework for feature extraction.
• Construction of an adjacency matrix to represent image region relevance.
• Application of graph inference for global modeling and classification of pneumonia.
• Experimentation with 6189 pediatric chest X-rays (3319 normal, 2870 pneumonia).

Main Results:

• The proposed pneumonia classification model achieved an accuracy of 89.1%.
• The model demonstrated a high F1-score of 90%.
• Performance was evaluated against 11 common models using 4 metrics on a 20% test dataset.

Conclusions:

• The deep learning model shows significant potential for accurate and effective pneumonia detection in children via chest X-rays.
• The findings highlight the utility of combining convolutional networks and graph inference for medical image classification.
• The developed model offers a promising tool for supporting clinicians in the early diagnosis of pediatric pneumonia.