Reconstruction-based approach for chest X-ray image segmentation and enhanced multi-label chest disease...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• U-Net models often fail to accurately segment pathological areas in chest X-rays.
• Precise segmentation is crucial for extracting relevant radiomic features for disease classification.

Purpose of the Study:

• To develop a novel approach for precise medical image segmentation and mask generation of pathological areas in chest X-rays.
• To improve the classification accuracy of chest diseases by incorporating pathological information.

Main Methods:

• Utilized CycleGAN for enhanced segmentation of pathological regions in chest X-rays.
• Implemented a feature selection strategy to identify significant radiomic features.
• Employed XGBoost for multi-label classification of effusion and infiltration.

Main Results:

• Achieved 92.05% average accuracy and 89.48% average AUC for effusion and infiltration classification.
• Attained an average AUC of 83.12% for classifying 14 diseases in the ChestX-ray14 dataset, surpassing prior research.
• Demonstrated effective pathological mask generation and feature selection for disease classification.

Conclusions:

• The proposed pipeline significantly improves chest disease classification accuracy.
• Effective pathological mask generation and feature selection are vital for accurate chest disease classification.
• The methodology shows potential for broader applications in medical image analysis and disease classification.