Table 2 Statistical comparison of the performance of generator networks on HC data.

From: Robust resolution improvement of 3D UTE-MR angiogram of normal vasculatures using super-resolution convolutional neural network

PSNR-oriented training

 

SR-ResNet

MRDG64

LSRDG

SR-ResNet vs. LSRDG

MRDG64 vs. LSRDG

SSIM

0.964 (± 0.01)

0.978 (± 0.004)

0.983 (± 0.005)

 < 10–5***

0.0002*

PSNR

34.38 (± 0.68)

35.47 (± 0.81)

36.80 (± 0.85)

 < 10–5***

 < 10–4***

MSE

0.00037 (± 6e-5)

0.00029 (± 5.7e-5)

0.00021 (± 4.8e-5)

 < 10–5***

 < 10–4***

PSNR + GAN–driven training

 

SRGAN

ESRGAN

LSRDGAN

SRGAN vs. LSRDGAN

ESRGAN vs. LSRDGAN

SSIM

0.953 (± 0.01)

0.976 (± 0.006)

0.981 (± 0.004)

 < 10–6***

0.002*

PSNR

32.88 (± 0.79)

35.35 (± 1.06)

36.30 (± 0.94)

 < 10–6***

0.007*

MSE

0.00053 (± 9.6e-5)

0.0003 (± 7.4e-5)

0.00024 (± 5.6e-5)

 < 10–6***

0.006*

  1. Data are shown as mean ± standard deviation. *: P < 0.05, and ***: P < 10−4 using the Mann–Whitney U test.
Back to article page