Skip to main content
. Author manuscript; available in PMC: 2020 Jul 1.
Published in final edited form as: Med Phys. 2019 Jun 5;46(7):3142–3155. doi: 10.1002/mp.13583

TABLE IV.

Comparison of image quality of the lung phantom images between the tuned-CNN correction and the fASKS and non-tuned-CNN corrections (mean + SD). Mean absolute error (MAE), mean squared error (MSE), peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) over central 70 slices were used.

MAE [HU] MSE [HU2] PSNR [dB] SSIM
fASKS 32.5 ± 3.2 2423 ± 559 30.6 ± 0.9 0.9990 ± 0.0003
CNN w.o. tuning 31.8 ± 3.8 2699 ± 974 30.2 ± 1.2 0.9984 ± 0.0005
CNN w. tuning 29.0 ± 2.5 1882 ± 376 31.7 ± 0.8 0.9993 ± 0.0002
p-value (vs fASKS) 2.41×10−11 4.72×10−10 1.71×10−12 6.98×10−17
p-value
(vs CNN w.o. tuning)		 5.91×10−7 1.11×10−9 7.36×10−15 2.80×10−32