Deep learning alone rivals multimodal MRI ensemble for meningioma grade prediction

A multimodal MRI ensemble predicted meningioma WHO grade with external AUC 0.742, but did not significantly outperform nnU-Net (AUC 0.732; ΔAUC +0.011, p=0.737).

• External AUCs: nnU-Net 0.732, DINOv3 0.703, clinical 0.645, radiomics 0.628.
• Radiomics-containing models lost external transportability despite strong development performance.
• Multimodal ensemble improved reclassification over clinical model (IDI +0.109, p<0.001).

Automated summary

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