Skip to main content

Table 2 Mean Dice Scores and average symmetric surface distance (ASSD) in mm over all classes for multi-view AbdomenNet models trained on single datasets versus combinations of two and of all three datasets

From: AbdomenNet: deep neural network for abdominal organ segmentation in epidemiologic imaging studies

 

Dice

ASSD

 

KORA

GNC

UKB

KORA

GNC

UKB

KORA

0.697 ± 0.109

0.644 ± 0.075

0.385 ± 0.095

2.164 ± 0.758

3.287 ± 1.940

7.440 ± 3.519

GNC

0.637 ± 0.120

0.682 ± 0.089

0.311 ± 0.130

2.923 ± 1.018

2.899 ± 2.121

8.768 ± 7.267

UKB

0.433 ± 0.128

0.401 ± 0.143

0.610 ± 0.114

6.273 ± 3.074

7.831 ± 3.511

3.518 ± 3.634

KORA + GNC

0.728 ± 0.106

0.710 ± 0.073

0.498 ± 0.111

1.658 ± 0.709

2.382 ± 1.762

3.780 ± 1.468

KORA + UKB

0.712 ± 0.112

0.683 ± 0.080

0.625 ± 0.119

2.036 ± 0.932

2.674 ± 1.726

3.238 ± 3.830

UKB + GNC

0.689 ± 0.106

0.697 ± 0.088

0.622 ± 0.116

1.757 ± 0.469

2.245 ± 1.533

5.074 ± 3.630

Joint Model

0.731 ± 0.109

0.716 ± 0.077

0.637 ± 0.119

1.780 ± 0.818

2.190 ± 1.559

3.021 ± 3.666

  1. Bold numbers indicate highest Dice scores and lowest ASSD scores