DeepChestGNN: A Comprehensive Framework for Enhanced Lung Disease Identification through Advanced Graphical Deep...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Lung diseases cause millions of deaths globally, posing a significant public health challenge.
• Classifying lung diseases from chest X-rays is difficult due to visual similarities and data complexities like noise and imbalances.
• Existing diagnostic methods require improvement for precision and efficiency.

Purpose of the Study:

• To develop an automated system for classifying ten different lung diseases from chest X-ray images.
• To address challenges in large-scale medical image datasets, including noise, annotations, class imbalance, and redundancy.
• To create a flexible and accurate model for precise lung disease identification.

Main Methods:

• Compiled a large dataset of 71,096 chest X-ray images from 17 sources.
• Applied image pre-processing (resizing, de-annotation, CLAHE, filtering) and elastic deformation augmentation.
• Developed DeepChestGNN, a novel model combining deep convolutional neural networks (DCNNs) and graph neural networks (GNNs) for feature extraction and classification.

Main Results:

• The DeepChestGNN model achieved 99.74% accuracy in extensive trials.
• The model effectively extracted 100 significant deep features for disease indication.
• The GNN integration provided flexibility for accurate classification using graph data.

Conclusions:

• DeepChestGNN offers a precise and efficient framework for lung disease diagnosis and classification.
• Combining advanced AI with clinical applications significantly impacts diagnosing respiratory illnesses.
• The developed automated system demonstrates high potential for clinical adoption in radiology.