Table 7.

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

Type	Method	Se	Sp	AUC	Acc
Handcrafted local feature-based methods	Akram et al. [23]	-	-	97.06	95.02
	Fraz et al. [24]	75.54	97.6	-	95.3
	Kar et al. (normal cases) [25]	75.77	97.88	-	97.30
	Kar et al. (abnormal cases) [25]	75.49	96.99	-	97.41
	Zhao et al. (without Retinex) [26]	76.6	97.72	86.9	94.9
	Zhao et al. (with Retinex) [26]	78.9	97.8	88.5	95.6
	Pandey et al. [27]	83.19	96.23	95.47	94.44
	Neto et al. [28]	83.44	94.43	-	88.94
	Zhao et al. [30]	78.0	97.8	87.4	95.6
	Jiang et al. [31]	77.67	97.05	-	95.79
	Sazak et al. [33]	73.0	97.9	-	96.2
Learned/deep feature-based methods	Zhang et al. (without post-processing) [37]	77.24	97.04	-	95.13
	Zhang et al. (with post-processing) [37]	78.82	97.29	-	95.47
	Wang et al. [40]	75.23	98.85	-	96.40
	Hu et al. [44]	75.43	98.14	97.51	96.32
	Fu et al. [45]	74.12	-	-	95.85
	Soomro et al. [46]	74.8	92.2	83.5	94.8
	Chudzik et al. [48]	82.69	98.04	98.37	-
	Hajabdollahi et at. (CNN) [49]	78.23	97.70	-	96.17
	Hajabdollahi et at. (Quantized CNN) [49]	77.92	97.40	-	95.87
	Hajabdollahi et at. (Pruned-quantized CNN) [49]	75.99	97.57	-	95.81
	Yan et al. [50]	77.35	98.57	98.33	96.38
	Soomro et al. [51]	74.8	96.2	85.5	94.7
	Jin et al. [52]	75.95	98.78	98.32	96.41
	Leopold et al. [53]	64.33	94.72	79.52	90.45
	Wang et al. [54]	79.14	97.22	97.04	95.38
	Feng et al. [55]	77.09	98.48	97.0	96.33
	Vess-Net (this work)	85.26	97.91	98.83	96.97