Table 1 Comparison of the metrics on the test set consisting of 54 patients for the three network configurations.

From: Deep learning methods allow fully automated segmentation of metacarpal bones to quantify volumetric bone mineral density

Network

2D U-Net

3D U-Net

2D U-Net (pre-trained)

IOU

0.933 (0.009)

0.937 (0.003)

0.943

Dice

0.965 (0.006)

0.969 (0.002)

0.972

AUC

0.996 (0.005)

0.998 (0.001)

0.999

  1. Mean values are followed by standard deviation in brackets. The 2D U-Net (pre-trained) is trained once, the two other configurations are trained three times with different random seeds. Best metrics are highlighted in bold font.
Back to article page