Early detection of COPD based on graph convolutional network and small and weakly labeled data.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Chronic obstructive pulmonary disease (COPD) presents significant morbidity and mortality.
• Early COPD detection is vital but challenging due to subtle early symptoms.
• Current diagnostic methods often miss early-stage disease.

Purpose of the Study:

• To propose a novel Graph Convolutional Network (GCN) method for early detection of COPD.
• To leverage weakly labeled chest computed tomography (CT) image data for COPD diagnosis.
• To improve the accuracy and efficiency of early COPD identification.

Main Methods:

• Utilized chest CT images from the Danish Lung Cancer Screening Trial database.
• Constructed graphs from randomly selected regions of interest in segmented lung parenchyma.
• Applied a GCN model to extract both feature and topological structure information for COPD detection.

Main Results:

• Achieved an accuracy of 0.77 and an Area Under the Curve (AUC) of 0.81 for early COPD detection.
• Demonstrated superior performance compared to previous studies on the same dataset.
• Outperformed several state-of-the-art methods in COPD detection tasks.

Conclusions:

• The proposed GCN model offers a promising approach for the early detection of COPD.
• Graph-based analysis of CT images can effectively capture disease-specific patterns.
• This study represents the first application of GCN on this dataset for COPD detection.