. Author manuscript; available in PMC: 2022 Sep 23.

Published in final edited form as: Med Image Anal. 2021 Sep 8;74:102222. doi: 10.1016/j.media.2021.102222

Table 7.

A comparison of prediction accuracy on MMWHS MR and CT test datasets from different methods trained with images from MMWHS training set.

			Epi	LA	LV	RA	RV	Ao	PA	WH

	Dice (↑)	Ours	0.880	0.926	0.931	0.868	0.885 ^*	0.945	0.786^*	0.900 ^*
		2DUNet	0.877^*	0.916	0.926	0.855	0.876^*	0.916	0.805	0.892^*
		3DUNet	0.816^*	0.916	0.914	0.848	0.878	0.923	0.793	0.877
		Voxel2Mesh	0.501^*	0.748^*	0.669^*	0.717^*	0.698^*	0.555^*	0.491^*	0.656^*
	Jaccard (↑)	Ours	0.790	0.863	0.874	0.773	0.798 ^*	0.897	0.666^*	0.819 ^*
		2DUNet	0.784^*	0.847	0.864	0.753	0.787^*	0.850	0.692	0.807^*
		3DUNet	0.696^*	0.848	0.844	0.741	0.787	0.860	0.670	0.782
		Voxel2Mesh	0.337^*	0.600^*	0.510^*	0.570^*	0.543^*	0.397^*	0.337^*	0.491^*
CT	ASSD (mm) (↓)	Ours	1.357	1.137	0.966	1.750	1.320	0.729 ^*	2.020	1.333^*
		2DUNet	1.014 ^*	1.141	0.911	1.702	1.433^*	0.808	1.754	1.240 ^*
		3DUNet	1.809^*	1.389	1.134	2.176	1.585	0.832	2.276	1.668
		Voxel2Mesh	3.412^*	3.147^*	4.973^*	3.638^*	4.300^*	4.326^*	5.857^*	4.287^*
	HD (mm) (↓)	Ours	13.789	10.362	9.628	14.467	12.766	12.740^*	25.362	27.567
		2DUNet	13.582	10.221	6.700	14.788	16.608^*	11.410	28.128	32.514
		3DUNet	15.044	40.157^*	9.730	15.037	13.777	10.821	27.467	48.731
		Voxel2Mesh	15.526^*	13.683^*	22.146^*	16.834^*	18.390^*	19.419^*	35.322^*	37.065^*
	Dice (↑)	Ours	0.773	0.826^*	0.913	0.838 ^*	0.861	0.824^*	0.663^*	0.846 ^*
		2DUNet	0.751	0.831	0.880	0.815	0.852	0.838 ^*	0.747	0.834
		3DUNet	0.733	0.811	0.885	0.827^*	0.829	0.825	0.741	0.823
		Voxel2Mesh	0.282^*	0.498^*	0.515^*	0.599^*	0.539^*	0.241^*	0.300^*	0.483^*
	Jaccard (↑)	Ours	0.639	0.712^*	0.842	0.727 ^*	0.768	0.715^*	0.517^*	0.737 ^*
		2DUNet	0.611^*	0.720	0.793	0.702	0.753	0.726 ^*	0.608	0.719^*
		3DUNet	0.588	0.695^*	0.803	0.718^*	0.727	0.718	0.615	0.709
		Voxel2Mesh	0.170^*	0.339^*	0.367^*	0.442^*	0.388^*	0.144^*	0.187^*	0.327^*
MR	ASSD (mm) (↓)	Ours	2.385	2.166^*	1.300	2.358 ^*	1.812	3.243	3.138	2.235^*
		2DUNet	2.692^*	1.688	1.603	3.151^*	1.736	2.920	2.281^*	1.897
		3DUNet	2.713	3.866	1.551	2.475	1.931	4.049	2.259	2.120
		Voxel2Mesh	6.886^*	5.987^*	8.679^*	6.173^*	8.192^*	7.877^*	9.200^*	7.419^*
	HD (mm) (↓)	Ours	16.804	15.559^*	12.197	17.286 ^*	14.480	26.012	19.927	29.983
		2DUNet	23.798	14.887 ^*	14.651	22.028^*	22.810^*	24.237	22.883^*	39.724^*
		3DUNet	20.136	32.978	13.643	23.735	22.351	31.900	21.363	43.475
		Voxel2Mesh	27.272^*	22.748^*	31.327^*	24.456^*	28.987^*	29.381	33.637^*	40.072^*

An asterisk * indicates statistically significant accuracy differences, compared with Table 2, resulted from training on a smaller datset based on t-tests (p < 0.05 ).