Improving FLAIR SAR efficiency by predicting B1-maps at 7T from a standard localizer scan using deep convolutional neural networks

Ultra-high-field (7T) instrumentation offers the possibility of acquiring FLAIR images at an improved resolution when challenges such as efficient B1 calibration and SAR reductions can be realized. Instead of acquiring a separate B1-map, we propose to predict B1-maps based on the implicit B1 inhomogeneity field present in an AutoAlign localizer using deep convolutional neural networks. We show that a 34% reduction in SAR can be achieved by adjusting the power of FLAIR's adiabatic inversion pulse on a slice-by-slice basis using the B1 information without degradation of image quality.