Table 1.

Study characteristics of included studies.

Study	Study location	AI classifier	Imaging technique (Acquisition Machine)	# Study participants	# Images used	AMD type or stage	Training and testing process	Database and datasets	Reference standard
Burlina et al. (2017)	United States	CNN (AlexNet, OverFeat)	Fundus	4613	67,401	All Types of AMD; Early, Intermediate, and Advanced Stage	2-class classification; Class 0 – 37,418 images; Class 1 – 29,983 images	National Institutes of Health AREDS	Trained clinical graders
Burlina et al. (2019)	United States	CNN	Fundus	4613	133 821	All Types of AMD; Early, Intermediate, and Advanced Stage	2-class classification AMD referral challenge derived from the original 4-step AREDS enrolment scale	National Institutes of Health AREDS	2 Retinal specialists
Fraccaro et al. (2015)	Italy	SVM, AdaBoost, Random Forest	Topcon 3D OCT	487	974	Dry and wet AMD	Linear kernel and the nu-classification, 50 bootstrap tests using out-of-bag predictions	Medical Retina Center of the University Eye Clinic of Genoa	2 Ophthalmologists
Hassan et al. (2018)	Pakistan	CNN	Topcon 3D OCT Heidelberg Spectralis	499	46,913	Dry and wet AMD	4992 images used for training, 41,921 images for validation	Duke Dataset - I, Duke Dataset - II, Duke Dataset - III, AFIO Dataset, Amanat Dataset	Ophthalmologists
Hwang et al. (2019)	Taiwan	CNN (VGG16 InceptionV3, ResNet50)	Zeiss Cirrus HD OCT, Optovue RTVue-XR Avanti	747	35,900	Dry, active wet, inactive wet AMD	28,720 images used for training, 7180 images used for validation	Department of Ophthalmology of Taipei Veterans General Hospital	2 Retinal specialists
Lee et al. (2016)	United States	CNN	SD-OCT Heidelberg Spectralis	9285	101,002	Dry, wet, and unspecified AMD	80,839 images used for training, 20,163 images used for validation	Heidelberg Spectralis imaging database	Retinal specialist
Li et al. (2019)	China	CNN	SD-OCT Heidelberg Spectralis	5319	109,312	Early AMD with drusen indicators	108,312 images used for training, 1000 images for validation	Shiley Eye Institute of the University of California San Diego, the California Retinal Research Foundation, Medical Center Ophthalmology Associates, the Shanghai First People’s Hospital, and the Beijing Tongren Eye Center	2 Ophthalmologists and a retinal specialist to resolve disagreements
Liu et al. (2011)	United States	SVM	Zeiss Cirrus SD-OCT	136	326	All types of AMD	tenfold cross validation	University of Pittsburgh Medical Center Eye Center, and New England Eye Center	3 Ophthalmologists
Srinivasan et al. (2014)	United States	SVM	SD-OCT Heidelberg Spectralis	45	90	Dry AMD	Leave-three-out cross-validation	Duke University, Harvard University, and the University of Michigan	Ophthalmologists
Tan et al. (2018)	Singapore	CNN	Fundus	–	1110	Dry and wet AMD	tenfold cross validation	Ophthalmology Department of Kasturba Medical College (KMC), Manipal, India	Retinal specialists
Ting et al. (2017)	Singapore	DLS	Topcon OCT	38,189	108,558	Intermediate AMD according to AREDS grading system	72,610 images used for training, 35,948 images used for validation	Singapore National Diabetic Retinopathy Screening Program 2010–2013, Singapore Malay Eye Study, Singapore Indian Eye Study, Singapore Chinese Eye Study, Singapore National Eye Centre AMD	Retinal specialists
Treder et al. (2018)	Germany	CNN	SD-OCT Heidelberg Spectralis	701	1112	Wet AMD	500 training step procedure with training and validation accuracy	University of Muenster Medical, Germany	–
Wang et al. (2016)	China	LCP	3D SD-OCT	45	3000	Dry AMD	tenfold cross validation	Duke University, Harvard University, and the University of Michigan	Ophthalmologists
Yoo et al. (2019)	South Korea	CNN (VGG-19)	SD-OCT Heidelberg Spectralis, Fundus	119	3000	Dry and wet AMD	fivefold cross validation	Project Macula Database	2 Ophthalmologists

“–” indicates no data reported”.

AdaBoost adaptive boosting, CNN convolutional neural network, DLS deep learning system, OCT optical coherence tomography, SVM support vector machine, LCP linear configuration patterns.