. Author manuscript; available in PMC: 2021 Jun 11.

Published before final editing as: IEEE Trans Image Process. 2019 Jun 19:10.1109/TIP.2019.2919937. doi: 10.1109/TIP.2019.2919937

TABLE VI.

Comparison With the State-of-the-Art Methods on the Brain Tumor Dataset. The Dice Ratio and ASD of the Whole Tumor (WT), Tumor Core (TC) and Enhancing Tumor (ET) Are Reported. The High-Resolution Pathway Was Placed on the Second (Proposed-L2) and the Third Stage (Proposed-L3) to Find a Good Balance Between Semantic Resolution and Network Complexity. The Boldface Results Indicate no Significant Difference From the Best Result (p-Value <0.05 in Student’s t-Test). The Parameter Number (Param) of Each Network Is Also Reported

Networks	WT	TC	ET	Param (M)
Networks	DSC(%)			Param (M)
3D U-Net	84.6±10.4	74.0±20.5	67.7±18.6	6.53
Deepmedic	87.4±6.4	78.8±l5.4	75.4±12.1	2.86
E-UNet	88.5±5.6	80.1±l8.8	77.5±11.3	8.27
3D DenseSeg	88.0±6.7	80.1±l6.6	74.7±15.1	1.26
Proposed-L2	89.7±5.2	83.9±14.4	79.8±10.7	4.39
Proposed-L3	89.0±5.5	82.2±15.0	77.7±13.8	9.64
Networks	WT	TC	ET	Param (M)
Networks	ASD(mm)			Param (M)
3D U-Net	4.261±4.408	7.030±6.775	5.920±6.691	6.53
Deepmedic	1.643±0.624	1.999±l.387	1.069±0.597	2.86
E-UNet	1.467±0.604	1.737±l.621	1.004±0.712	8.27
3D DenseSeg	1.826±1.290	1.799±1.431	1.258±1.168	1.26
Proposed-L2	1.288±0.565	1.481±1.244	0.895±0.582	4.39
Proposed-L3	1.455±0.577	1.676±1.324	0.923±0.563	9.64