Table 2.

Quantitative results for the effectiveness of different network components

	Aff				Aff+local
	Dice	Jac	SSIM	ASSD	Dice	Jac	SSIM	ASSD
Original	0.860	0.755	0.893	2.04	0.939	0.886	0.932	0.84
	(0.024)	(0.036)	(0.015)	(0.35)	(0.016)	(0.027)	(0.013)	(0.20)
2 aff encoders	0.855	0.748	0.891	2.13	0.940	0.888	0.933	0.82
	(0.023)	(0.035)	(0.016)	(0.34)	(0.016)	(0.027)	(0.013)	(0.20)
Wo skip	0.846	0.734	0.887	2.29	0.937	0.883	0.931	0.86
	(0.035)	(0.050)	(0.016)	(0.52)	(0.017)	(0.029)	(0.013)	(0.21)
Single 3D dec	0.853	0.744	0.890	2.16	0.930	0.870	0.925	0.97
	(0.021)	(0.032)	(0.017)	(0.33)	(0.018)	(0.030)	(0.014)	(0.24)
Wo inv-ProST	0.851	0.742	0.889	2.18	0.932	0.873	0.927	0.95
	(0.026)	(0.038)	(0.015)	(0.36)	(0.016)	(0.027)	(0.013)	(0.21)
2 aff encoders	${10}^{-7}$	${10}^{-7}$	${10}^{-7}$	${10}^{-9}$	${10}^{-2}$	${10}^{-2}$	${10}^{-1}$	${10}^{-2}$
Wo skip	${10}^{-8}$	${10}^{-8}$	${10}^{-9}$	${10}^{-7}$	${10}^{-3}$	${10}^{-3}$	${10}^{-3}$	${10}^{-3}$
Single 3D dec	${10}^{-10}$	${10}^{-10}$	${10}^{-8}$	${10}^{-11}$	${10}^{-21}$	${10}^{-21}$	${10}^{-23}$	${10}^{-22}$
Wo inv-ProST	${10}^{-12}$	${10}^{-13}$	${10}^{-13}$	${10}^{-12}$	${10}^{-17}$	${10}^{-17}$	${10}^{-16}$	${10}^{-17}$

The mean and standard deviation (between brackets) of the evaluation metrics are tabulated for the different network variations. The bottom table shows the p-values of a paired t-test between the original network and each variation on the network architecture