Transmit radiofrequency field maps can be predicted from standard brain images using machine learning.

Motivation: Transmit radiofrequency field (B1+) maps are typically needed for quantitative relaxation mapping but are not normally acquired in clinical studies. Goal(s): We developed a machine learning model to predict B1+ maps from standard clinical brain images. Approach: A Generative Adversarial Network (GAN) was trained on combinations of T1, T2 and proton density (PD) weighted images and used to predict brain B1+ maps in comparison to measured B1+ maps in a large dataset. Results: The predicted B1+ maps closely matched the ground truth in both regional and global analyses, with best agreement in central regions. Impact: Prediction of B1+ maps from standard clinical brain images will enable more widespread use of quantitative MRI in clinical settings where B1+ maps are not typically acquired.