Modeling global and local label correlation with graph convolutional networks for multi-label chest X-ray image...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate diagnosis of chest diseases from radiography is challenging.
• Manual annotation by expert radiologists is time-consuming and subjective.
• Automated analysis methods are crucial for computer-aided diagnosis.

Purpose of the Study:

• To develop an end-to-end multi-label learning framework (GL-MLL) for chest disease diagnosis.
• To jointly model global and local label correlations for improved diagnostic performance.
• To address imbalanced class distribution and capture label-specific features.

Main Methods:

• Proposed GL-MLL framework for multi-label learning.
• Explored label correlation from global static and local adaptive views.
• Incorporated methods to handle imbalanced class distribution.
• Focused on capturing label-specific features in image-level representations.

Main Results:

• Validated the GL-MLL framework on the CheXpert dataset.
• Demonstrated superior performance compared to state-of-the-art approaches.
• Achieved improved accuracy in diagnosing chest diseases.

Conclusions:

• The GL-MLL framework effectively models label correlations for enhanced chest disease diagnosis.
• Automated analysis using GL-MLL offers a promising alternative to manual interpretation.
• The proposed method advances computer-aided diagnosis in medical imaging.