Table 2.

Summary of most relevant works where fundus images are taken as input.

Year	Reference	Topic	Model	Dataset (patients)	Output: classes	AUC	Sens.	Spec.	Better than experts? (N)
2020	Varadarajan et al. [68]	DME	CNN (InceptionV3)	Thailand (4732) / EyePACS-DME (554)	Center-involved DME	0.890	0.850	0.800	Yes (3)
					Intraretinal fluid	0.810
					Subretinal fluid	0.880
2020	Singh and Gorantla [71]	DME	CNNs (Hierarchical Ensemble of 5)	Messidor (1200) / IDRiD (516)	DME		0.947/ 0.979	0.972/ 0.945
					DME: (three-class grading)		0.964	0.958
2019	Arsalan et al. [23]	DR, Vessel seg.	CNN (Custom, Vess-Net)	DRIVE (40) / CHASE-DB1 (128) / STARE (20)	Vessel segmentation map	0.980–0.988	0.802–0.853	0.979–0.984
2019	Sayres et al. [69]	DR	CNN (InceptionV4)	Patients from Krause2018 (1612)	DR: None, mild, moderate, severe, prolif.		0.925	0.946	Yes (10)
2019	Cao et al. [22]	DR	Naive Bayes on grey coocurrence matrix	Private (1000)	DR	0.938	0.949	0.928
2019	Yang et al. [67]	DR	CNN (IDx-DR)	Private (500)	DR: None, mild, moderate, severe, prolif.		0.988	0.880
2019	Peng et al. [70]	AMD (Bilateral images)	CNN (Inceptionv3: DeepSeeNet)	AREDS (4549)	AMD: AREDS early AMD progression risk scale (five classes)		0.590	0.930	Yes (88)
					Drusen: Small / none, medium, large	0.940	0.718	0.871	Yes
					Pigment. abnorm.	0.930	0.732	0.957	Yes
					Late AMD	0.970	0.627/ 0.538	0.987/ 0.898	No
2019	Coyner et al. [50]	ROP	CNN (InceptionV3)	i-ROP (898)	ROP: Image quality assesment (2 classes)	0.9650	0.939	0.836	Same (6)
2019	Tan et al. [72]	ROP	CNN (ROP.AI)	Australasian ART-ROP ($\sim$500)	ROP: Normal, plus-disease	0.9770	0.939	0.807
					ROP: Normal, pre-plus disease		0.814	0.807
2018	Grassmann et al. [45]	AMD	CNN (Ensemble of six)	AREDS (3654) / KORA (5555)	AMD: AREDS scale (13 classes)		0.538/0.328	0.969/0.957	Yes
2018	Rajalakshmi et al. [73]	DR (Phone images)	CNN (EyeArt)	Private (296)	DR grade: Any, DME, prolif., referrable		0.781–0.993	0.688–0.898
2018	Kanagasingam et al. [74]	DR	CNN (IDx-DR)	Private (193)	DR: None, mild, moderate, severe, prolif.			0.920
2018	Krause et al. [36]	DR	CNN (InceptionV4)	EyePACS (242252)	DR: None, mild, moderate, severe, prolif.	0.986	0.970	0.917	Same (6)
2018	Brown et al. [52]	ROP	CNN (Vessel seg.: U-Net) + CNN (Classif.: InceptionV1)	i-ROP (898)	ROP	0.940	0.930	0.940	Yes (8)
					ROP: Plus disease	0.980	1,000	0.940
2018	Wang et al. [51]	ROP	CNN (custom, Inception-like)	Chengdu Women Children's Central Hospital (1273)	ROP		0.849	0.969	Same (3)
					ROP: Minor, severe		0,736	0,933
2017	Li et al. [16]	Glaucoma	CNN (InceptionV3)	Guangdong (3970)	Glaucoma: Referable, nonreferable	0.986	0.956	0.920
2017	Gargeya and Leng [34]	DR	CNN (Custom) + Decission tree (Grad. Boosting)	EyePACS ($\sim$15000) / Messidor-2 (874) / E-Optha ($\sim$100)	DR: Referable, Non Referable	0.940–0.970	0.900–0.940	0.870–0.980
2017	Ting et al. [39]	DR	CNN (VGG-19)	SIDRP 10-15 (27979) / Guangdong (3970) + 9 others	DR: Referable, non referable	0.889-0.983	0.905-1.000	0.733-0.916
		Glaucoma		SIDRP 10-15 (27979) + 5 others	Glaucoma: Referable, non referable	0.942	0.964	0.872
		AMD		Idem	AMD: Referable, non referable	0.931	0.932	0.887
2017	Burlina et al. [2]	AMD	CNN (AlexNet)	AREDS (4613)	AMD	0.940	0.846	0.920	Same
2016	Abràmoff et al. [75]	DR	CNN (IDx-DR vX2.1)	Messidor-2 (874)	DR: None or mild, present, vision-threatening	0.980	0.968	0.870
2016	Gulshan et al. [3]	DR, DME	CNN (InceptionV3)	EyePACS (4997) / Messidor-2 (874)	DR: Referable, non referable	0.991/ 0.990	0.903/ 0.870	0.981/ 0.985
					DME: Referable, non referable	0.974	0.907	0.938
2012	Zheng et al. [44]	AMD	Hierarchical trees + SVM	ARIA (161) / STARE (97)	AMD		0.994	1,000
2011	Agurto et al. [8]	AMD, DR (3 FOV fundus)	Feature extraction + Partial Least Squares	RIST (378) / UTHSCSA (444)	DR: Normal, sight-threatening	0.890/ 0.920	0.950	0.500
					DR-related: Microaneurysms, hemorrhages, exudates, neovascularization, etc.	0.770-0.980	0.830–1.00	0.500
					AMD-related: Drusen, Pigment. abnorm., GA	0.770-0.920	0.88–1.000	0.500