Table 3.

Automated liver parenchyma, portal veins, and hepatic veins segmentation on single-modal and multi-modal T1 vibe Dixon acquisitions.

		In	In-W	In-Opp-W	In-Opp	In-Opp-F-W	W	Opp	p-value
Liver parenchyma	DSC	0.936 ± 0.02	0.935 ± 0.02	0.935 ± 0.02	0.934 ± 0.02	0.934 ± 0.02	0.931 ± 0.02	0.921 ± 0.03	0.090
	AP	0.981 ± 0.01	0.981 ± 0.01	0.981 ± 0.01	0.979 ± 0.01	0.980 ± 0.01	0.976 ± 0.02	0.970 ± 0.02
Portal veins	DSC	0.634 ± 0.09	0.634 ± 0.08	0.631 ± 0.08	0.626 ± 0.08	0.632 ± 0.08	0.571 ± 0.14	0.505 ± 0.12	< 0.001
	AP	0.680 ± 0.12	0.678 ± 0.12	0.672 ± 0.12	0.667 ± 0.12	0.672 ± 0.12	0.590 ± 0.15	0.526 ± 0.13
Hepatic veins	DSC	0.532 ± 0.12	0.523 ± 0.11	0.523 ± 0.11	0.525 ± 0.11	0.519 ± 0.11	0.475 ± 0.16	0.395 ± 0.12	< 0.001
	AP	0.553 ± 0.12	0.535 ± 0.12	0.537 ± 0.11	0.540 ± 0.12	0.531 ± 0.12	0.479 ± 0.16	0.399 ± 0.13

Results are measured by Dice similarity coefficient (DSC) and average precision (AP) and presented as mean ± SD.

P-values were calculated using the Kruskal–Wallis test with Dunn’s multiple comparison post-hoc test. The single-modal neural network inputs are In, in-phase; W, water; Opp, opposed phase. The multi-modal neural network inputs are In-W, in-phase, water; In-Opp-W, in-phase, opposed-phase, water; In-Opp, in-phase, opposed-phase; In-Opp-F-W, in-phase, opposed-phase, fat, water. Best results are shown in bold.