Framework for enhanced respiratory disease identification with clinical handcrafted features

Area of Science:

• Medical Imaging and Artificial Intelligence
• Computational Pathology
• Respiratory Disease Diagnostics

Background:

• Respiratory disorders are a leading cause of global mortality, necessitating early and accurate detection.
• Chest X-ray interpretation requires specialized expertise, highlighting the need for computational support.
• Existing diagnostic methods can be enhanced by advanced machine learning techniques for improved accuracy.

Purpose of the Study:

• To develop and validate an automated system for classifying respiratory diseases using chest X-ray images.
• To leverage Graph Neural Networks (GNNs) for enhanced feature analysis and disease classification.
• To improve diagnostic accuracy and efficiency in respiratory healthcare through artificial intelligence.

Main Methods:

• Utilized a dataset of 18,000 chest X-ray images with preprocessing, augmentation, and region of interest extraction.
• Employed handcrafted feature extraction and K-nearest neighbors (KNN) graph construction for tabular data transformation.
• Developed and tested the Chest X-ray Graph Neural Network (CHXGNN) model, incorporating GNNExplainer for validation.

Main Results:

• The CHXGNN model achieved a high accuracy of 99.56% in classifying respiratory diseases.
• Feature analysis and GNNExplainer successfully identified critical attributes influencing classification decisions.
• The system demonstrated robust performance across diverse datasets and demonstrated effectiveness in disease detection.

Conclusions:

• The proposed CHXGNN model offers a highly accurate and efficient automated solution for respiratory disease classification.
• This AI-driven approach has significant potential to assist medical professionals and improve patient outcomes.
• The study underscores the value of GNNs in medical image analysis for enhanced diagnostic capabilities.