Table 7 The effect of different positions of ASPP and SE-Attention modules on the ACDC dataset.

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

Position

Average DSC

RV

Myo

LV

A[\(A_2\)-\(A_3\)] +S[\(B_1\)-\(B_2\)]

89.73

88.36

86.29

94.54

A[\(A_2\)-\(A_3\)] +S[\(B_0\)-\(B_1\)]

87.73

84.23

84.79

94.17

A[\(A_2\)-\(A_3\)] +S[\(B_2\)-\(B_3\)]

86.37

83.40

82.82

92.91

A[\(A_3\)-\(A_4\)] +S[\(B_1\)-\(B_2\)]

87.76

85.06

84.40

93.83

A[\(A_1\)-\(A_2\)] +S[\(B_1\)-\(B_2\)]

87.29

84.49

83.97

93.41

  1. ’A’ represents the ASPP module, ’S’ denotes the SE-Attention module, \(A_1\), \(A_2\), \(A_3\), and \(A_4\) denote the position in the encoder, whereas \(B_1\), \(B_2\), and \(B_3\) is the position in the decoder
Back to article page