COVID-19 lung CT image segmentation using deep learning methods: U-Net versus SegNet

BMC Medical Imaging

Table 7 Five-fold experiment results for the best network of each architecture

			Sensitivity		Specificity		Dice		G-mean		F2
			\(\mu\)	\(\sigma\)	\(\mu\)	\(\sigma\)	\(\mu\)	\(\sigma\)	\(\mu\)	\(\sigma\)	\(\mu\)	\(\sigma\)
Binary	SegNet		0.947	0.048	0.945	0.015	0.703	0.055	0.945	0.019	0.829	0.029
Binary	U-NET		0.961	0.033	0.923	0.018	0.643	0.058	0.941	0.014	0.800	0.033
Multi	SegNet	\({\hbox {C}}_1\)	0.653	0.043	0.927	0.030	0.425	0.084	0.778	0.035	0.535	0.069
		\({\hbox {C}}_2\)	0.688	0.072	0.963	0.004	0.410	0.081	0.813	0.043	0.537	0.073
		\({\hbox {C}}_3\)	0.679	0.270	0.987	0.006	0.117	0.068	0.804	0.167	0.214	0.110
	U-NET	\({\hbox {C}}_1\)	0.685	0.130	0.910	0.045	0.402	0.144	0.786	0.083	0.527	0.135
		\({\hbox {C}}_2\)	0.666	0.120	0.973	0.013	0.458	0.093	0.801	0.066	0.550	0.052
		\({\hbox {C}}_3\)	0.632	0.26	0.991	0.005	0.152	0.092	0.777	0.165	0.250	0.118

Results are the mean and standard deviation of sensitivity, specificity, dice, g-mean, and F2

ISSN: 1471-2342