Table 2.

Model comparison for COVID-19 infection segmentation.

Methods	Pre-trained architecture	DSC$\uparrow$	Sens$\uparrow$	Spec$\uparrow$	${\mathit{S}}_{\mathit{\alpha }}\uparrow$	${\mathit{E}}_{\mathit{\varphi }}\uparrow$	MAE$\downarrow$
U-Net [9]	VGG16	0.459	0.568	0.881	0.639	0.651	0.196
H-DenseUNet [11]	DenseNet-101	0.537	0.611	0.870	0.663	0.683	0.189
U-Net++ [12]	VGG16	0.607	0.701	0.932	0.739	0.751	0.139
SegNet [13]	VGG16	0.657	0.728	0.941	0.744	0.750	0.129
Inf-Net [39]	Res2Net	0.705	0.746	0.966	0.798	0.851	0.086
SE-Net [18]	–	0.621	0.719	0.949	0.751	0.801	0.142
Semi-Inf-Net [39]	Res2Net	0.752	0.757	0.965	0.818	0.902	0.061
*FSS-2019-nCov	Res2Net	0.798	0.803	0.986	0.834	0.908	0.065

$\uparrow$denote ‘higher is better’, $\downarrow$denote ‘lower is better’.