Graph neural network model using radiomics for lung CT image segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate lung segmentation in CT images is crucial for diagnosing lung cancer, COVID-19, and other respiratory diseases.
• Existing segmentation methods face challenges due to overlapping structures, complex features, and intricate tissue morphology, limiting accuracy.
• Early detection through precise segmentation is key to improving patient treatment outcomes.

Purpose of the Study:

• To introduce GEANet, a novel framework for enhancing lung segmentation accuracy in CT images.
• To address limitations in current segmentation techniques by incorporating advanced deep learning and graph-based approaches.
• To improve the detection and delineation of lung abnormalities, including tumors.

Main Methods:

• GEANet employs an encoder-decoder architecture enhanced with radiomics features.
• Graph Neural Network (GNN) modules are integrated to capture tumor heterogeneity.
• A boundary refinement module and a hybrid loss function (Focal Loss and IoU Loss) are utilized for improved accuracy and robustness.

Main Results:

• GEANet demonstrated superior performance compared to eight state-of-the-art methods on benchmark datasets.
• The framework achieved higher segmentation accuracy across various evaluation metrics.
• GEANet maintained computational efficiency while delivering enhanced segmentation results.

Conclusions:

• GEANet offers a significant advancement in automatic lung segmentation for CT images.
• The proposed framework effectively handles complex anatomical structures and tumor heterogeneity.
• GEANet shows strong potential for clinical application in early lung disease diagnosis and management.