Table 4 The effect of different modules in our method on the Synapse dataset.

From: MedFuseNet: fusing local and global deep feature representations with hybrid attention mechanisms for medical image segmentation

Methods

Average DSC

HD

Aorta

Gallbladder

Kidney(L)

Kidney(R)

Liver

Pancreas

Spleen

Stomach

w/o CNN

28.60

42.55

31.32

0.49

20.77

18.30

70.68

31.21

35.38

27.97

w/o Swin-Transformer

72.73

25.26

84.99

48.07

81.35

73.75

92.84

50.24

84.63

66.00

w/o Cross attention

73.67

30.67

85.57

53.61

77.63

78.52

92.59

51.49

82.76

67.18

w/o Adaptive cross attention

74.93

32.47

84.55

54.62

77.55

71.63

93.28

56.01

87.75

74.06

w/o ASPP

77.91

24.27

85.61

57.98

82.27

77.44

94.12

56.82

88.40

78.50

w/o SE-Attention

75.46

23.54

84.48

60.99

81.15

72.74

93.36

55.29

84.66

71.02

w/o Skip connection

65.66

19.31

73.68

27.19

78.55

72.30

89.12

42.63

82.72

59.11

Ours

78.40

18.44

85.71

62.58

85.10

77.76

93.98

58.16

75.33

78.74