Andreas Petrovic1,2, Yiqiu Dong3,
Stephen Keeling3, Rudolf Stollberger1
1Institute of Medical Engineering,
University of Technology Graz, Graz, Austria; 2Ludwig Boltzmann
Institute for Clinical Forensic Imaging, Graz, Austria, Austria; 3University
of Graz
MR
images at high field strengths (≥1.5T) suffer from artifacts caused by
the inhomogeneity of the RF excitation field B1+ in the human
body. Measurements of B1+ can be used for the correction of those
artifacts. However, these B1+-images suffer from perturbations
themselves and have to be smoothed and interpolated. In this work a new
variational approach for smoothing is compared to a standard median filter
for test images, as well as real in-vivo data. Simulations show that the
variational approach combined with an outlier suppression algorithm
outperforms the median filter in terms of accuracy and precision. In contrast
to the median filter the variational approach produces very smooth results
that are physically likely.