. Author manuscript; available in PMC: 2021 Apr 1.

Published in final edited form as: Magn Reson Med. 2019 Oct 21;83(4):1471–1483. doi: 10.1002/mrm.28022

TABLE 1.

Mean (and standard deviation) Dice scores (cross-validation) of the FCNN models for abdominal adipose tissue segmentation. We show FDR corrected significance indicators of Wilcoxon signed-rank test [29] comparing the proposed CDFNet vs. benchmark FCNNs

	Subcutaneous (SAT)				Visceral (VAT)
Models (PRM)^†	Axial	Coronal	Sagittal	V. Aggregation	Axial	Coronal	Sagittal	V. Aggregation
UNet (~ 20 M)	0.965 (0.029)^*	0.960 (0.034)^*	0.960 (0.035)^*	0.972 (0.019)^*	0.810 (0.111)^*	0.804 (0.113)^*	0.820 (0.101)	0.837 (0.095)^*
SD-Net (~ 1,5M)	0.969 (0.027)^*	0.954 (0.040)^*	0.956 (0.034)^*	0.972 (0.020)^*	0.820 (0.097)^*	0.812 (0.099)^*	0.822 (0.091)^*	0.843 (0.081)^*
Dense-UNet (~ 3,3M)	0.972 (0.025)^*	0.959 (0.037)^*	0.963 (0.029)^*	0.975 (0.019)^*	0.824 (0.091)^*	0.814 (0.097)^*	0.827 (0.090)^*	0.847(0.080)^*

Proposed (~2,5M)	0.970 (0.025)	0.966 (0.029)	0.966 (0.027)	0.975(0.018)	0.826 (0.095)	0.826 (0.085)	0.824 (0.092)	0.850(0.076)

Inter-rater variability	0.982 (0.018)				0.788 (0.060)

^†

The approximately number of learn parameters reported is for the models without the View-Aggregation Network

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