Table 7 Ablation study on the number of the proposed blocks using the Brats 2023 dataset.

From: SwinCLNet: a robust framework for brain tumor segmentation via shifted window attention and cross-scale fusion

Segmentation Methods

Dice score (%, \(\uparrow\))

HD95 (mm, \(\downarrow\))

WT

TC

ET

Mean

WT

TC

ET

Mean

(Model-0) Baseline

88.38 ± 0.31

79.51 ± 0.55

74.49 ± 0.90

80.79 ± 0.59

7.28 ± 0.80

9.88 ± 1.20

25.81 ± 3.10

14.33 ± 1.70

(Model-1) Enhancement block  ×  1

89.56 ± 0.28

81.94 ± 0.48

76.47 ± 0.81

82.70 ± 0.52

6.66 ± 0.72

8.82 ± 1.05

23.65 ± 2.80

13.19 ± 1.52

(Model-2) Enhancement block  ×  2

90.64 ± 0.25

84.08 ± 0.41

78.36 ± 0.70

84.42 ± 0.45

6.09 ± 0.65

7.91 ± 0.90

21.84 ± 2.50

12.18 ± 1.35

(Model-3) Enhancement block  ×  3

91.90 ± 0.20

86.60 ± 0.35

80.59 ± 0.55

86.44 ± 0.37

5.39 ± 0.50

6.79 ± 0.70

19.66 ± 2.10

10.94 ± 1.10

(Model-4) Enhancement block  ×  4

92.49 ± 0.18

87.82 ± 0.30

81.58 ± 0.48

87.39 ± 0.32

5.03 ± 0.45

6.26 ± 0.60

18.49 ± 1.80

10.26 ± 0.95

(Model-5) Enhancement block  ×  5 (Proposed)

93.27 ± 0.15

89.78 ± 0.25

82.62 ± 0.40

88.56 ± 0.27

4.27 ± 0.30

5.49 ± 0.50

16.12 ± 1.50

8.63 ± 0.77