Table 5.

Accuracies of Vess-Net and existing methods for DRIVE dataset (unit: %).

Type	Method	Se	Sp	AUC	Acc
Handcrafted local feature-based methods	Akram et al. [23]	-	-	96.32	94.69
	Fraz et al. [24]	74.0	98.0	-	94.8
	Kar et al. [25]	75.48	97.92	-	96.16
	Zhao et al. (without Retinex) [26]	76.0	96.8	86.4	94.6
	Zhao et al. (with Retinex) [26]	78.2	97.9	88.6	95.7
	Pandey et al. [27]	81.06	97.61	96.50	96.23
	Neto et al. [28]	78.06	96.29	-	87.18
	Sundaram et al. [29]	69.0	94.0	-	93.0
	Zhao et al. [30]	74.2	98.2	86.2	95.4
	Jiang et al. [31]	83.75	96.94	-	95.97
	Rodrigues et al. [32]	71.65	98.01	-	94.65
	Sazak et al. [33]	71.8	98.1	-	95.9
	Chalakkal et al. [34]	76.53	97.35	-	95.42
	Akyas et al. [36]	74.21	98.03	-	95.92
Learned/deep feature-based methods	Zhang et al. (without post-processing) [37]	78.95	97.01	-	94.63
	Zhang et al. (with post-processing) [37]	78.61	97.12	-	94.66
	Tan et al. [38]	75.37	96.94	-	-
	Zhu et al. [39]	71.40	98.68	-	96.07
	Wang et al. [40]	76.48	98.17	-	95.41
	Tuba et al. [41]	67.49	97.73	-	95.38
	Girard et al. [43]	78.4	98.1	97.2	95.7
	Hu et al. [44]	77.72	97.93	97.59	95.33
	Fu et al. [45]	76.03	-	-	95.23
	Soomro et al. [46]	74.6	91.7	83.1	94.6
	Guo et al. [47]	78.90	98.03	98.02	95.60
	Chudzik et al. [48]	78.81	97.41	96.46	-
	Yan et al. [50]	76.31	98.20	97.50	95.38
	Soomro et al. [51]	73.9	95.6	84.4	94.8
	Jin et al. [52]	79.63	98.00	98.02	95.66
	Leopold et al. [53]	69.63	95.73	82.68	91.06
	Wang et al. [54]	79.86	97.36	97.40	95.11
	Feng et al. [55]	76.25	98.09	96.78	95.28
	Vess-Net (this work)	80.22	98.1	98.2	96.55