Retinal Vessel Segmentation by Deep Residual Learning with Wide Activation

Table 2

Comparison of the performance among different methods on the DRIVE dataset.

Type	Method	Year	F1	Sens	Spec	Acc	AUC
Unsupervised	Roychowdhury et al. [37]	2015	——	0.7395	0.9782	0.9494	0.9672
	Azzopardi et al. [38]	2015	——	0.7655	0.9704	0.9442	0.9614
	Zhang et al. [39]	2016	——	0.7743	0.9725	0.9476	0.9636
Supervised	Li et al. [40]	2015	——	0.7569	0.9816	0.9527	0.9738
	Liskowski and Krawiec[17]	2016	——	0.7763	0.9768	0.9495	0.9720
	Chen [41]	2017	——	0.7426	0.9735	0.9453	0.9516
	Yan et al. [42]	2018	——	0.7631	0.9820	0.9538	0.9750
	R2U-Net [43]	2018	0.8171	0.7792	0.9813	0.9556	0.9784
	DCCMED-Net^a [24]	2019	——	0.7268	0.9912	0.9681	0.9819
	CE-Net [19]	2019	——	0.8309	——	0.9545	0.9779
	WA-Net (ours)	2020	0.8222	0.7875	0.9813	0.9566	0.9794

^aThe listed result is obtained by DCCMED-Net with batch size 32, which is the same as WA-Net.