Wyger Brink1, Sahar Yousefi1,2, Prernna Bhatnagar1, Marius Staring2, Rob Remis3, and Andrew Webb1
1C.J. Gorter Center, dept. of Radiology, Leiden University Medical Center, Leiden, Netherlands, 2Division of Image Processing, dept. of Radiology, Leiden University Medical Center, Leiden, Netherlands, 3Circuits and Systems, dept. of Microelectronics, Delft University of Technology, Delft, Netherlands
1C.J. Gorter Center, dept. of Radiology, Leiden University Medical Center, Leiden, Netherlands, 2Division of Image Processing, dept. of Radiology, Leiden University Medical Center, Leiden, Netherlands, 3Circuits and Systems, dept. of Microelectronics, Delft University of Technology, Delft, Netherlands
In this work we have developed a semantic segmentation method based on deep learning, which is able to generate a subject-specific body model for personalized RF exposure prediction at 7T.
Fig.
1. Schematic illustration of the segmentation pipeline to obtain a
subject-specific body model for RF exposure analysis. The semi-automatic
segmentation process involves many steps with elaborate user interaction, while
the deep learning approach is able to generate an accurate body model directly
from 7T T1-weighted images.
Fig.
4. Comparison of simulated SAR10g distributions in ground truth
(top) and network-generated body
models (middle), and corresponding difference maps (bottom). Figure headings denote
peak SAR10g values (top, middle) and relative difference of the peak
SAR10g (bottom).
