Table 2.

The effect of post-processing (PP) in 2D CNN segmentation results.

Network	DSC ↑	Precision ↑	Recall ↑	ASD [mm] ↓	MSD [mm] ↓
2D CNN* PP, F:64, L:4	0.920 ± 0.040	0.991 ± 0.010	0.861 ± 0.060	0.72 ± 0.38	11.70 ± 3.74
2D CNN* PP, F:64, L:3	0.935 ± 0.034	0.990 ± 0.010	0.889 ± 0.056	0.62 ± 0.36	10.50 ± 3.23
2D CNN PP, F:64, L:4	0.960 ± 0.022	0.978 ± 0.015	0.943 ± 0.036	0.39 ± 0.44	8.18 ± 5.87
2D CNN PP, F:64, L:3	0.953 ± 0.027	0.979 ± 0.013	0.930 ± 0.046	0.47 ± 0.37	9.61 ± 4.60
2D CNN†, F:64, L:4	0.937 ± 0.026	0.932 ± 0.037	0.943 ± 0.036	2.13 ± 1.23	42.22 ± 5.52

Segmentation results of different network architectures are presented here. F is the number of initial feature maps, L is the number of layers. * indicates the 2D CNN with unpadded convolutions. ^† indicates the best performing 2D CNN model prior to post-processing from Table 1.