TABLE 2. Quantitative Evaluation of Different Refinement Methods in Stage 2 of 2D Placenta Segmentation.

Method	Dice(%)	ASSD(pixels)
Before refinement	85.86 $\pm$ 11.67	1.85 $\pm$ 1.30
Min-cut user-editing	87.04 $\pm$ 9.79	1.63 $\pm$ 1.15
2D R-Net(Euc)	88.26 $\pm$ 10.61	1.54 $\pm$ 1.18
2D R-Net	88.76 $\pm$ 5.56	1.31 $\pm$ 0.60
2D R-Net(Euc) + CRF-Net(fu)	88.71 $\pm$ 8.42	1.26 $\pm$ 0.59
2D R-Net + CRF-Net(fu)	89.31 $\pm$ 5.33	1.22 $\pm$ 0.55

The initial segmentation is obtained by 2D P-Net + CRF-Net(f). 2D R-Net(Euc) uses euclidean distance instead of geodesic distance. Significant improvement from 2D R-Net ($p$-value $<$ 0.05) is shown in bold font.