¹⁸F-FDG-PET/CT Whole-Body Imaging Lung Tumor Diagnostic Model: An Ensemble E-ResNet-NRC with Divided Sample Space

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Deep learning for medical diagnosis

Background:

• Convolutional Neural Networks (CNNs) face challenges like network degradation and high-dimensional features in lung tumor diagnosis using 18F-FDG-PET/CT.
• Multimodal imaging data (CT, PET, PET/CT) presents opportunities and complexities for feature extraction and classification.

Purpose of the Study:

• To propose an Ensemble ResNet nonnegative representation classifier (E-ResNet-NRC) model to address CNN training challenges in lung tumor diagnosis.
• To improve the classification performance, robustness, and generalization ability of lung tumor detection using multimodal imaging.

Main Methods:

• Utilized transfer learning to initialize a pretrained ResNet model.
• Divided samples into distinct spaces (CT, PET, PET/CT) and extracted Region of Interest (ROI) features.
• Developed an individual ResNet-NRC classifier and an ensemble E-ResNet-NRC classifier using a relative majority voting method.

Main Results:

• The E-ResNet-NRC model demonstrated superior overall classification performance compared to individual ResNet-NRC.
• Achieved higher specificity and sensitivity in lung tumor classification.
• Exhibited enhanced robustness and generalization capabilities.

Conclusions:

• The proposed E-ResNet-NRC model effectively overcomes CNN training limitations for lung tumor diagnosis.
• Ensemble learning significantly boosts diagnostic accuracy and reliability.
• The model shows promise for improved clinical application in multimodal medical imaging analysis.