Table 4 Comparison of glioma segmentation on T1Gd images synthesized by different methods on the MR-1 testing set and the MR-2 dataset

From: Contrast-free identification of glioma blood-brain barrier status via generative diffusion AI and non-contrast MRI

Dataset

Input images

Measures

Pix2pix

CycleGAN

SwinUNETR

Hi-Net

DDPM

CBSI (ours)

MR-1 Testing Set (n = 96 samples)

Synthetic T1Gd

DSC

0.71 ± 0.20

0.80 ± 0.15

0.68 ± 0.23

0.81 ± 0.07

0.77 ± 0.13

0.82 ± 0.09

p-value

1.9e-6

0.8066

1.8e-8

0.7148

1.8e-4

-

z

−4.77

−0.24

−5.63

−3.75

−0.37

-

r

−0.49

−0.02

−0.58

−0.38

−0.04

-

95%CI

(0.03, 0.10)

(−0.01, 0.02)

(0.06, 0.11)

(0.02, 0.05)

(−0.01, 0.03)

-

HD95

20.49 ± 34.95

8.99 ± 22.37

20.66 ± 37.03

5.85 ± 8.27

11.56 ± 25.05

4.94 ± 8.02

ASD

7.30 ± 17.31

2.92 ± 7.50

8.28 ± 20.83

1.19 ± 1.24

2.73 ± 6.58

1.04 ± 1.07

ASSD

5.75 ± 14.93

2.07 ± 4.64

6.03 ± 12.86

1.27 ± 1.07

2.24 ± 4.11

1.13 ± 1.42

Jaccard

0.58 ± 0.19

0.68 ± 0.15

0.55 ± 0.21

0.69 ± 0.09

0.64 ± 0.14

0.70 ± 0.11

Synthetic T1Gd + non-contrast MR

DSC

0.93 ± 0.10

0.93 ± 0.10

0.92 ± 0.10

0.94 ± 0.10

0.93 ± 0.10

0.94 ± 0.10

p-value

2.6e-5

3.2e-3

2.8e-6

0.5083

5.8e-4

-

z

−4.20

−2.95

−4.69

−3.44

−0.66

-

r

−0.43

−0.30

−0.48

−0.35

−0.07

-

95%CI

(0.01, 0.02)

(0.00, 0.01)

(0.01, 0.02)

(0.00, 0.01)

(−0.00, 0.01)

-

HD95

1.98 ± 6.89

2.48 ± 11.68

2.06 ± 7.42

2.3 ± 11.73

2.42 ± 12.07

2.12 ± 10.20

ASD

0.57 ± 3.24

0.68 ± 4.44

0.60 ± 3.04

0.62 ± 4.24

0.79 ± 5.46

0.50 ± 3.20

ASSD

0.54 ± 2.40

0.58 ± 2.91

0.53 ± 2.16

0.50 ± 2.72

0.64 ± 3.57

0.43 ± 2.08

Jaccard

0.87 ± 0.10

0.88 ± 0.10

0.86 ± 0.10

0.89 ± 0.10

0.88 ± 0.10

0.90 ± 0.10

MR-2 Dataset (n = 94 samples)

Synthetic T1Gd

DSC

0.77 ± 0.20

0.76 ± 0.25

0.56 ± 0.30

0.82 ± 0.15

0.78 ± 0.18

0.82 ± 0.12

p-value

0.3167

0.7586

2.7e-11

0.2386

0.1716

-

z

−1.00

−0.31

−6.66

−1.37

−1.18

-

r

−0.10

−0.03

−0.69

−0.14

−0.12

-

95%CI

(−0.02, 0.03)

(−0.04, 0.01)

(0.10, 0.23)

(−0.01, 0.03)

(−0.03, 0.01)

-

HD95

13.03 ± 30.96

14.27 ± 27.31

35.96 ± 46.76

9.85 ± 27.28

7.50 ± 12.67

7.02 ± 13.53

ASD

6.01 ± 22.33

4.67 ± 11.38

13.35 ± 26.01

3.27 ± 14.97

2.03 ± 6.81

1.81 ± 8.85

ASSD

4.85 ± 17.73

3.75 ± 9.52

12.39 ± 22.16

2.71 ± 10.38

2.38 ± 6.40

2.12 ± 8.70

Jaccard

0.66 ± 0.19

0.66 ± 0.24

0.44 ± 0.26

0.71 ± 0.15

0.67 ± 0.17

0.70 ± 0.12

Synthetic T1Gd + non-contrast MR

DSC

0.94 ± 0.02

0.94 ± 0.02

0.95 ± 0.02

0.95 ± 0.02

0.94 ± 0.02

0.95 ± 0.02

p-value

2.6e-6

7.2e-8

0.0185

0.0779

4.4e-5

-

z

−4.70

−5.39

−2.35

−4.09

−1.76

-

r

−0.48

−0.56

−0.24

−0.42

−0.18

-

95%CI

(0.01, 0.01)

(0.01, 0.02)

(−0.00, 0.01)

(0.00, 0.02)

(−0.00, 0.01)

-

HD95

1.22 ± 0.58

1.42 ± 0.75

1.11 ± 0.57

1.14 ± 0.46

1.28 ± 0.90

1.04 ± 0.36

ASD

0.18 ± 0.12

0.19 ± 0.12

0.18 ± 0.09

0.16 ± 0.08

0.18 ± 0.09

0.16 ± 0.07

ASSD

0.28 ± 0.17

0.32 ± 0.20

0.24 ± 0.12

0.24 ± 0.13

0.27 ± 0.16

0.21 ± 0.09

Jaccard

0.89 ± 0.04

0.88 ± 0.04

0.90 ± 0.03

0.90 ± 0.03

0.89 ± 0.03

0.91 ± 0.03

  1. Bold values indicate the best performance within each row. Statistical significance of DSC was assessed using two-sided Wilcoxon signed-rank tests; z represents the test statistic, r denotes the effect size, 95% confidence intervals (CI) are reported, and exact p-values are provided.