Deep learning based estimation of B1 field maps for variable flip angle qT1 mapping

Motivation: Quantitative T1(qT1) is a sensitive brain measure. However, qT1 estimation using variable flip angle methods requires B1 maps to correct the flip angle errors in SPGR images Goal(s): Our goal is to develop a robust deep learning framework for fast and accurate predictions of a B1 maps directly from the SPGR images. Approach: A pipeline was developed to process SPGR images and a U-net model was trained to estimate B1 maps. Results: The deep learning model achieves good bias correction of qT1 with lower mean and standard deviation compared to the baseline. Impact: This method enables retrospective estimation of qT1 from legacy variable flip angle SPGR data acquired without B1 mapping protocols.