Improving Computer-Aided Thoracic Disease Diagnosis through Comparative Analysis Using Chest X-ray Images Taken at...

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Thoracic Medicine

Background:

• Thoracic physicians routinely compare serial chest X-rays to detect lung abnormalities.
• Current comparison methods can be time-consuming and prone to misjudgment.
• There is a need for enhanced diagnostic tools to improve efficiency and accuracy.

Purpose of the Study:

• To design a computer-aided diagnosis (CADx) system for efficient comparison of sequential chest X-ray images.
• To improve the accuracy of lesion detection and reduce misjudgments in thoracic imaging.
• To evaluate the performance of deep learning versus traditional methods in key diagnostic stages.

Main Methods:

• Developed a four-component system: segmentation, alignment, comparison, and classification of lung X-ray images.
• Utilized public (NIH Chest X-ray14) and local datasets.
• Compared traditional and deep learning approaches for segmentation and alignment.
• Designed nonlinear transfer functions for difference highlighting and proposed novel network architectures for classification.

Main Results:

• Deep learning methods outperformed traditional methods in segmentation and alignment, showing higher IoU, detection rates, and faster processing.
• Nonlinear transfer functions effectively highlighted differences between sequential images via heat maps.
• Dual-input networks incorporating difference information improved classification accuracy (AUC) by 1.2-1.4%.

Conclusions:

• Deep learning significantly improves efficiency and accuracy in early stages of chest X-ray analysis.
• Incorporating image difference information is crucial for accurate lung disease identification and classification.
• The developed CADx system shows promise for future clinical applications in comparative thoracic imaging analysis.