TABLE 2.

Mean (and standard deviation) Dice scores (cross-validation) of hard-coded balanced weights, hard-coded axial focus weights and the proposed view-aggregation for abdominal adipose tissue segmentation. We show FDR corrected significance indicators of Wilcoxon signed-rank test [29] comparing the proposed data-driven aggregation scheme vs. each hard-coded method.

	Subcutaneous (SAT)			Visceral (VAT)
Single-View Model	Balanced	Axial Focus	Proposed	Balanced	Axial Focus	Proposed
UNet	0.970 (0.026)	0.970 (0.026)	0.972 (0.019)	0.830 (0.098)*	0.829 (0.099)*	0.837 (0.095)
SD-Net	0.970 (0.026)*	0.972 (0.025)*	0.972 (0.020)	0.839 (0.084)*	0.838 (0.085)*	0.843 (0.082)
Dense-UNet	0.973 (0.025)	0.974 (0.024)*	0.975 (0.019)	0.841 (0.081)*	0.840 (0.082)*	0.847 (0.080)
CDFNet	0.972 (0.025)*	0.973 (0.024)	0.975 (0.018)	0.844 (0.077)*	0.841 (0.080)*	0.850 (0.076)

^*Statistical difference using a one-sided adaptive FDR multiple comparison correction [30] at a level of 0.05