Hybrid ResNet-50–Vision Transformer with radiomics predicts lung cancer TNM stage from CT

A hybrid ResNet-50–Vision Transformer model with radiomic feature fusion achieved 97.62% accuracy for benign vs. malignant lung nodule classification on CT (LIDC-IDRI). It also independently predicts T, N, and M stage components.

• The model combined ResNet-50 for local morphological features and Vision Transformer for global spatial dependencies, with handcrafted radiomic features for TNM prediction.
• Explainability analysis using Grad-CAM identified critical tumor regions to support clinician confidence.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.