Lung Disease Detection Using U-Net Feature Extractor Cascaded by Graph Convolutional Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Graph Neural Networks

Background:

• Computed tomography (CT) scans are crucial for rapid lung disease diagnosis.
• Existing deep learning methods often overlook spatial connectivity in CT image analysis.

Purpose of the Study:

• To develop an accurate method for COVID-19 diagnosis using graph convolutional networks (GCNs).
• To enhance feature extraction from CT scans by incorporating spatial connectivity patterns.

Main Methods:

• Utilized U-Net for image segmentation and feature extraction.
• Employed GCNs to capture spatial connectivity from extracted deep features, forming an adjacency matrix.
• Integrated original image graph, largest kernel graph, and feature-extracted graph for input to GCN.
• Applied a dropout layer to mitigate overfitting.

Main Results:

• The proposed feature-extracted graph convolutional network (FGCN) demonstrated superior performance in lung disease identification.
• FGCN outperformed existing deep learning architectures not based on graph representations.
• The model also surpassed common transfer learning models used in medical diagnosis.

Conclusions:

• Graph representation offers significant advantages over traditional methods for medical image analysis.
• The FGCN framework provides a powerful tool for accurate and efficient lung disease diagnosis from CT scans.