Skip to main content
. Author manuscript; available in PMC: 2021 Oct 28.
Published in final edited form as: IEEE Access. 2021;9:140824–140834. doi: 10.1109/access.2021.3118290

FIGURE 4.

FIGURE 4.

Inter-subject performance analysis of the 4-stage U-Net with 64 convolutional filters at the initial layer. Adam optimizer worked better than SGD in terms of SSIM for all scenarios. (Top Row) For 3T Body SAR prediction, an average SSIM of 85.1∓6.2% and an average MSE of 0.4∓0.4% was observed (Adam), as opposed to an SSIM of 69.3∓4.5% and an MSE of 0.5∓0.4% (SGD). (Bottom Row) In the 7T brain SAR prediction, we observed an average SSIM of 90.5∓3.6% and an average MSE of 0.7∓0.6% (Adam), compared to an SSIM of 81.4∓2.6% and an MSE of 0.5∓0.5% (SGD). The higher spatial resolution of the 7T data (i.e., isotropic pixel size of ~1.33 mm compared to 2.67 mm at 3T) resulted in improved SSIM and MSE.