CNN-ViT fusion model on MRI stratifies survival in high-grade glioma after chemotherapy

A deep learning model fusing CNN and Vision Transformer features on MRI achieved external validation AUC 0.810 (95% CI 0.710–0.911) for stratifying high-grade glioma survival after temozolomide, in a multicenter study of 507 patients.

• Fusion model outperformed standalone CNN and ViT, with external-validation AUC 0.810 (95% CI 0.710–0.911), internal-test AUC 0.828, and training AUC 0.950.
• Transcriptomic analysis linked the imaging-based low-risk group to upregulation of cell cycle and DNA damage response genes (CHEK1, CDK1, FEN1).

Automated summary

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