An Effective Method for Detecting and Classifying Diabetic Retinopathy Lesions Based on Deep Learning

Table 1

Studies using MESSIDOR dataset and results.

Authors	Training type	Method	Process type	ACC	AUC	SEN
Zhang et al. [49]	ZFNet	TL	Optic disc localization	99.9		—
Alghamdi et al. [50]	CNN	EE	Optic disc localization	99.2	—	—
Xu et al. [51]	CNN	TL	Optic disc localization	99.4	—	—
Abràmoff et al. [52]	CNN	EE	Lesion detection	—	—	100
Grinsven et al. [53]	CNN	EE	Lesion detection	—	97.9	93.1
Gulshan et al. [54]	CNN	TL	Fundus classification	—	99.0	87.0
Costa and Campilho [55]	SURF + CNN	EE	Fundus classification	—	90.0	—
Gargeya and Leng [56]	CNN	EE	Fundus classification	—	94.0	—
Wang et al. [57]	Zoom	EE	Fundus classification	91.1	95.7	—
Chen et al. [58]	SI2DRNet	EE	Fundus classification	91.2	96.5	—