Multi-component CT radiomics-ViT model for VPI prediction in lung adenocarcinoma ≤30 mm

A combined radiomics and Vision Transformer (ViT) model using CT threshold segmentation predicted visceral pleural invasion in lung adenocarcinoma ≤30 mm with AUCs 0.895 training, 0.852 internal test, 0.823 external test, and the highest sensitivity.

• The combined radiomics-ViT model maintained an AUC of 0.823 on external test set (127 patients from a separate center).
• Vision Transformer features were the dominant contributor to the combined model's performance.

Automated summary

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