Speaker
Description
Gadolinium-based contrast agents improve tumor visualization but pose risks such as toxicity and tissue retention. This study evaluated whether contrast enhancement can be predicted noninvasively from qMRI using deep learning and assessed the added value of QSM and PD.A nnU-Net was trained on 46 datasets and tested on 16 cases. Two models were compared: one using T1, T2, PD, and QSM, and one using only T1 and T2.The four-parameter model performed better, achieving a mean Dice score of 0.60 versus 0.52. It reliably predicted enhancement, correctly identifying non-tumor cases, though false positives occurred in non-enhancing tumors.Multiparametric qMRI improves prediction accuracy and supports Gadolinium-free imaging, highlighting the clinical potential of combining qMRI with deep learning.