Automatic Retinal Vessel Segmentation Based on an Improved U-Net Approach

Table 5

Performance comparison of different methods for the STARE fundus database.

Method	Algorithm	ACC	TPR	TNR
Unsupervised method	Zhang et al. [46]	0.9554	0.7791	0.9758
	Hoover et al. [13]	0.9264	0.6747	0.9565
	Mendonça and Campilho[8]	0.9440	0.6996	0.9730
	Jiang and Mojon [39]	0.9009	—	—
	You et al. [44]	0.9497	0.7260	—
	Zhao et al. [47]	0.9560	0.7800	0.9780
Supervised method	Fraz et al. [5]	0.9534	0.7548	0.9763
	Lin et al. [27]	0.9603	0.7423	—
	Fu et al. [26]	0.9545	0.7140	—
	Niemeijer et al. [7]	0.9534	0.7548	0.9763
	Soares et al. [48]	0.9480	0.7207	0.9747
	Staal et al. [49]	0.9516	—	—
	Marin et al. [18]	0.9526	0.6944	0.9819
	GAN [51]	0.9647	0.7940	0.9869
	Wang et al. [20]	0.9813	0.8104	0.9791
	Khowaja et al. [52]	0.9751	0.8239	0.9749
	Our algorithm	0.9683	0.6032	0.9967