Table 3 Dice similarity coefficient and Hausdorff distance for each stage with 3D U-Net only and cascaded 3D U-Net (DSC, Dice similarity coefficient; HD, Hausdorff distance).

From: Fully automated condyle segmentation using 3D convolutional neural networks

Metric

Model

Stage I

 

Stage II

 

Stage III

 

Stage IV

 

Stage V

DSC

Basic 3D U-Net

0.886 ± 0.034

 

0.903 ± 0.039

 

0.910 ± 0.029

 

0.909 ± 0.026

 

0.922 ± 0.021

Cascaded 3D U-Net

0.912 ± 0.043

 

0.921 ± 0.035

 

0.928 ± 0.038

 

0.916 ± 0.032

 

0.932 ± 0.023

P

0.245

 

 < 0.05

 

 < 0.001

 

0.120

 

 < 0.01

HD

Basic 3D U-Net

3.099 ± 1.218

 

2.882 ± 1.061

 

2.884 ± 0.992

 

2.491 ± 0.319

 

2.557 ± 0.250

Cascaded 3D U-Net

2.494 ± 0.652

 

2.518 ± 0.460

 

2.600 ± 0.726

 

2.538 ± 0.395

 

2.452 ± 0.332

P

 < 0.05

 

0.081

 

0.141

 

0.985

 

 < 0.001

Back to article page