A novel fusion algorithm for benign-malignant lung nodule classification on CT images

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Accurate lung nodule recognition on CT scans is vital for early lung cancer detection and reducing mortality.
• Deep learning, particularly Convolutional Neural Networks (CNNs), shows promise in medical image analysis.
• Radiomics and Graph Convolutional Networks offer advanced methods for feature extraction and contextual understanding.

Purpose of the Study:

• To propose a novel fusion algorithm (RGD) for benign-malignant lung nodule classification.
• To enhance feature representation by integrating Radiomics and Graph learning with multiple Deep CNNs.
• To improve the robustness and accuracy of lung nodule classification for better clinical decision-making.

Main Methods:

• Developed a fusion algorithm (RGD) combining Radiomics, Graph learning, and multiple Deep CNNs.
• Utilized a 10-fold cross-validation on the LIDC-IDRI dataset for performance evaluation.
• Ensembled predictions from the integrated models for robust decision-making.

Main Results:

• Achieved an average accuracy of 93.25% on the LIDC-IDRI dataset.
• Reported high performance metrics including sensitivity (89.22%), specificity (95.82%), precision (92.46%), F1 Score (0.9114), and AUC (0.9629).
• Demonstrated superior performance compared to existing state-of-the-art methods, confirmed by ablation studies.

Conclusions:

• The RGD fusion model significantly improves benign-malignant lung nodule classification accuracy.
• The integration of Radiomics and Graph learning enhances feature distinctiveness and classification robustness.
• The model shows potential for increasing confidence in clinical lung cancer diagnosis.