Robust Detection Model of Vascular Landmarks for Retinal Image Registration: A Two-Stage Convolutional Neural Network

Table 2

Comparison of the segmentation performance on DRIVE database.

Method	Sensitivity	Specificity	Accuracy	AUC
Unsupervised method
Aguirre-Ramos et al. [34]	0.785	0.966	0.953	—
Shah et al. [35]	0.742	0.977	0.947	—
Memari et al. [36]	0.761	0.981	0.961	0.871
Solís-Pérez et al. [37]	0.827	0.965	0.956	—
Zhou et al. [38]	0.726	0.980	0.948	—
Supervised method
Zhuang [39]	0.786	0.981	0.956	0.979
Alom et al. [40]	0.779	0.981	0.956	0.978
Guo et al. [41]	0.789	0.980	0.956	0.981
Feng et al. [42]	0.762	0.981	0.953	0.968
Kushol et al. [43]	0.759	0.975	0.946	—
Adapa et al. [44]	0.629	0.984	0.945	0.951
Jin et al. [45]	0.739	0.983	—	0.976
Khan et al. [46]	0.825	0.979	0.965	0.978
Proposed method	0.805	0.982	0.966	0.982