Table 2 Quantitative comparison on the LIDC-IDRI and LNDb datasets

From: CoreFormer high fidelity pulmonary nodule segmentation with structural core priors and geodesic implicit fields

Method

LIDC-IDRI

LNDb

 

Dice↑

HD↓

ASSD↓

BF1↑

Dice↑

HD↓

ASSD↓

BF1↑

3D U-Net

84.2

3.64

1.29

81.5

82.7

3.89

1.35

79.3

V-Net

85.0

3.45

1.22

82.3

83.1

3.67

1.28

80.1

Attention U-Net

85.8

3.31

1.17

83.2

83.9

3.52

1.24

81.2

nnU-Net

86.5

3.21

1.14

83.7

84.6

3.35

1.17

82.8

STUNet

87.1

2.98

1.09

84.2

85.2

3.08

1.03

83.9

TransUNet

85.9

3.12

1.12

83.1

84.1

3.40

1.20

82.0

UNETR

86.3

3.05

1.11

83.5

84.7

3.20

1.15

82.6

CoreFormer (Ours)

88.7

2.54

0.93

87.6

86.4

2.89

0.91

86.1

  1. We report Dice (%), Hausdorff Distance (HD, mm), Average Symmetric Surface Distance (ASSD, mm), and Boundary F1 Score (BF1, %). Best results are in bold, second best are underlined.
Back to article page