. Author manuscript; available in PMC: 2022 Jun 1.

Published in final edited form as: IEEE Access. 2022 Mar 4;10:29322–29332. doi: 10.1109/access.2022.3156894

TABLE 3.

Results of an experiment to investigate the performance of the proposed network with different numbers of training images. This experiment was performed on brain cortical plate segmentation labels from the dHCP dataset [14].

No. of training images	Method	DSC	HD95 (nun)	ASSD (mm)
n_train = 10	Proposed	0.910 ± 0.033 ^*	0.814 ± 0.042	0.231 ± 0.070 ^*
n_train = 10	UNet++	0.884 ± 0.040	0.816 ± 0.043	0.247 ± 0.065

n_train = 100	Proposed	0.926 ± 0.030 ^*	0.806 ± 0.040 ^*	0.226 ± 0.068 ^*
n_train = 100	UNet++	0.916 ± 0.032	0.813 ± 0.044	0.241 ± 0.062

n_train = 500	Proposed	0.928 ± 0.031 ^*	0.805 ± 0.041	0.221 ± 0.070 ^*
n_train = 500	UNet++	0.916 ± 0.028	0.805 ± 0.044	0.240 ± 0.057

Asterisks denote statistically significant differences at p = 0.01, computed using paired t-tests.