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

Table 4

Performance comparison of different methods for the DRIVE fundus database.

Method	Algorithm	ACC	TPR	TNR
Unsupervised method	Vlachos and Dermatas [12]	0.9290	0.7470	—
	Zana and Klein [6]	0.9377	0.6971	—
	Mendonça and Campilho [8]	0.9452	0.7344	0.9764
	Jiang and Mojon [39]	0.9212	0.6399	—
	You et al. [44]	0.9434	0.7410	—
	Chaudhuri et al. [45]	0.8773	0.2716	0.9785
	Zhang et al. [46]	0.9476	0.7743	0.9725
	Zhao et al. [47]	0.9540	0.7420	0.9820
Supervised method	Fraz et al. [5]	0.9480	0.7406	0.9807
	Lin et al. [27]	0.9536	0.7632	—
	Fu et al. [26]	0.9470	0.7294	—
	Niemeijer et al. [7]	0.9416	—	—
	Maji et al. [21]	0.9470	—	—
	Soares et al. [48]	0.9466	0.7332	0.9782
	Staal et al. [49]	0.9442	0.7322	0.9646
	Zhu et al. [50]	0.9607	0.7528	0.9820
	Marin et al. [18]	0.9452	0.7067	0.9801
	GAN [51]	0.9562	0.7746	0.9753
	Wang et al. [20]	0.9767	0.8173	0.9733
	Khowaja et al. [52]	0.9753	0.8176	0.9709
	Our algorithm	0.9701	0.8011	0.9849