Skip to main content

Table 2 Quantitative assessment of fat and water predictions compared with the original data, using four-fold cross-validation on 800 scans

From: Artifact-free fat-water separation in Dixon MRI using deep learning

Model

Run

Water

Fat

SSIM

PSNR (dB)

SSIM

PSNR (dB)

\(IP\rightarrow \hat{F},\hat{W}\)

1

0.919 \({\pm }\) 0.011

24.28 \({\pm }\) 0.78

0.945 \({\pm }\) 0.009

24.70 \({\pm }\) 0.84

 

2

0.913 \({\pm }\) 0.012

24.07 \({\pm }\) 0.70

0.942 \({\pm }\) 0.008

24.45 \({\pm }\) 0.75

 

3

0.926 \({\pm }\) 0.009

24.74 \({\pm }\) 0.77

0.942 \({\pm }\) 0.008

25.07 \({\pm }\) 0.83

 

4

0.919 \({\pm }\) 0.010

24.35 \({\pm }\) 0.74

0.945 \({\pm }\) 0.010

24.55 \({\pm }\) 0.81

\(IP,OP\rightarrow \hat{F},\hat{W}\)

1

0.961 \({\pm }\) 0.006

28.99 \({\pm }\) 0.91

0.975 \({\pm }\) 0.004

29.67 \({\pm }\) 0.96

 

2

0.962 \({\pm }\) 0.005

28.94 \({\pm }\) 0.82

0.972 \({\pm }\) 0.003

29.00 \({\pm }\) 0.80

 

3

0.966 \({\pm }\) 0.005

29.41 \({\pm }\) 0.83

0.976 \({\pm }\) 0.004

29.58 \({\pm }\) 0.85

 

4

0.963 \({\pm }\) 0.005

29.10 \({\pm }\) 0.84

0.974 \({\pm }\) 0.004

29.41 \({\pm }\) 0.83

\(IP,OP\rightarrow \hat{F},\hat{W}\)

1

0.930 \({\pm }\) 0.010

25.11 \({\pm }\) 0.83

0.953 \({\pm }\) 0.007

25.37 \({\pm }\) 0.91

(Dixon generator loss)

2

0.928 \({\pm }\) 0.008

25.51 \({\pm }\) 0.85

0.949 \({\pm }\) 0.007

25.51 \({\pm }\) 0.85

 

3

0.935 \({\pm }\) 0.009

25.94 \({\pm }\) 0.91

0.952 \({\pm }\) 0.008

26.14 \({\pm }\) 0.96

 

4

0.924 \({\pm }\) 0.009

25.36 \({\pm }\) 0.89

0.951 \({\pm }\) 0.008

25.35 \({\pm }\) 0.88

  1. Values reported are the average and standard deviation. The three models are: single-input \(IP\rightarrow \hat{F},\hat{W}\), dual-input \(IP,OP\rightarrow \hat{F},\hat{W}\) and dual-input (Dixon generator loss) \(IP,OP\rightarrow \hat{F},\hat{W}\)