TABLE IV:

Semantic segmentation results measured by IoU (mean±s.d. %) for U-Net, wide U-Net, UNet+ (our intermediate proposal), and UNet++ (our final proposal). Both UNet+ and UNet++ are evaluated with and without deep supervision (DS). We have performed independent two sample t-test between U-Net [5] vs. others for 20 independent trials and highlighted boxes in red when the differences are statistically significant (p < 0.05).

Architecture	DS	Params	2D Application					Architecture	DS	Params	3D Application
			EM	Cell	Nuclei	Brain Tumor†	Liver				Lung Nodule
U-Net [5]	$\mathcal{X}$	7.8M	88.30±0.24	88.73±1.64	90.57±1.26	89.21±1.55	79.90±1.38	V-Net [28]	$\mathcal{X}$	22.6M	71.17±4.53
wide U-Net	$\mathcal{X}$	9.1M	88.37±0.13	88.91±1.43	90.47±1.15	89.35±1.49	80.25±1.31	wide V-Net	$\mathcal{X}$	27.0M	73.12±3.99
UNet+	$\mathcal{X}$	8.7M	88.39±0.15	90.71±1.25	91.73±1.09	90.70±0.91	79.62±1.20	VNet+	$\mathcal{X}$	25.3M	75.93±2.93
UNet+	✓	8.7M	88.89±0.12	91.18±1.13	92.04±0.89	91.15±0.65	82.83±0.92	VNet+	✓	25.3M	76.72±2.48
UNet++	$\mathcal{X}$	9.0M	88.92±0.14	91.03±1.34	92.44±1.20	90.86±0.81	82.51±1.29	VNet++	$\mathcal{X}$	26.2M	76.24±3.11
UNet++	✓	9.0M	89.33±0.10	91.21±0.98	92.37±0.98	91.21±0.68	82.60±1.11	VNet++	✓	26.2M	77.05±2.42

^†The winner in BraTS-2013 holds a “complete” Dice of 92% vs. 90.83%±2.46% (our UNet++ with deep supervision).