TABLE 5.

Top 10 citing articles on the application of the use of AI in the diagnosis of ophthalmic diseases from 2012 to 2021.

Rank	Title of citing documents	DOI	Times cited	Interpretation of the findings	Research limitations
1	Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs (Gulshan et al., 2016)	10.1001/jama.2016.17216	2,591	Deep machine learning-based algorithms have high sensitivity and specificity for detecting actionable diabetic retinopathy.	1. Image subtly was difficult for ophthalmologists to interpret.
					2. The algorithm only displayed the lesion grade and did not count the actual diabetic retinopathy lesions.
					3. Ophthalmic examination image data sets were limited in number.
					4. The algorithm identified only diabetic retinopathy and diabetic macular edema.
					5. The clinical utility of user interface settings is unknown.
2	Development and Validation of a Deep Learning System for Diabetic Retinopathy and Related Eye Diseases Using Retinal Images From Multiethnic Populations With Diabetes (Ting et al., 2017)	10.1001/jama.2017.18152	769	Deep learning systems for the evaluation of retinal images in multiethnic diabetic patients are highly sensitive and specific for identifying diabetic retinopathy and associated eye diseases.	1. Inconsistencies in diagnostic criteria among ophthalmologists.
					2. The algorithm only displayed the lesion grade and did not count the actual diabetic retinopathy lesions.
					3. Diagnosis of all diabetic macular edema still requires the use of optical coherence tomography
3	Segmenting Retinal Blood Vessels with Deep Neural Networks (Liskowski and Krawiec, 2016)	10.1109/TMI.2016.2546227	481	Deep neural networks are a viable methodology for medical imaging.	Only a limited set of image data including drive database, start database, and chase database, were used. These data sets contained limited examination populations.
4	Automated Identification of Diabetic Retinopathy using Deep Learning (Gargeya and Leng, 2017)	10.1016/j.ophtha.2017.02.008	478	This study presented a novel deep learning-based automatic feature learning method for Diabetic Retinopathy detection that offered an efficient, low-cost, and objective diagnostic method, which has high efficiency without relying on clinicians to manually review and grade images.	1. It was difficult for the algorithm to automatically distinguish between partial and early-stage cases of diabetic retinopathy.
					2. Limitations in the number of image datasets analyzed.
5	Improved Automated Detection of Diabetic Retinopathy on a Publicly Available Dataset Through Integration of Deep Learning (Abramoff et al., 2016)	10.1167/iovs.16-19964	403	Deep learning enhanced algorithms have the potential to improve the efficiency of diabetic retinopathy screening	1. The ophthalmic disease examination images in the disclosed data set represented only part of the clinical examination images.
					2. Different reference standards may cause differences in the performance of device measurement algorithms.
					3. The approach lacked the same flexibility as an actual clinical diagnosis.
6	Pivotal Trial of an Autonomous AI-Based Diagnostic System for Detection of Diabetic Retinopathy in Primary Care Offices (Abramoff et al., 2018)	10.1038/s41746-018-0040-6	355	The algorithm developed in this study is the first autonomous artificial intelligence diagnosis system for the detection of diabetic retinopathy in any medical field authorized by the United States Food and Drug Administration.	1. Limitations of the spectrum of disease tested in the system.
					2. The sensitivity of the AI system was lower than that of a similar AI system that was tested using a laboratory dataset.
7	A Cross-Modality Learning Approach for Vessel Segmentation in Retinal Images (Li et al., 2016)	10.1109/TMI.2015.2457891	330	A novel supervised vascular segmentation method for retinal images was presented, which has potential applications in retinal image diagnostic systems	1. There are specific requirements for the quality of the images to be diagnosed.
					2. Special algorithms that simultaneously predict all pixel labels in one retinal image block remain unknown.
8	Automatic Segmentation of Nine Retinal Layer Boundaries in OCT Images of Non-Exudative AMD Patients using Deep Learning and Graph Search (Fang et al., 2017)	10.1364/BOE.8.002732	274	A new framework combining convolutional neural network and pattern search method was proposed for automatic segmentation of nine-layer boundaries of retinal optical coherence tomography image	The framework was validated in only subjects with non-exclusive age-related macular degeneration.
9	Joint Optic Disc and Cup Segmentation Based on Multi-Label Deep Network and Polar Transformation (Fu et al., 2018)	10.1109/TMI.2018.2791488	277	This study proposed a deep learning architecture called M-net, which jointly solved the problem of the optic disc and cup segmentation in fundus images in a single-stage multi-label system, and developed a function for glaucoma screening	The image data sets selected for verification were limited and included only ORIGA and SCES datasets.
10	Efficacy of a Deep Learning System for Detecting Glaucomatous Optic Neuropathy Based on Color Fundus Photographs (Li et al., 2018)	10.1016/j.ophtha.2018.01.023	272	This study proposed a deep learning system for detecting referable glaucomatous optic neuropathy with high sensitivity and specificity.	The ophthalmic images used in the study were only collected from Chinese hospitals, resulting in limitations associated with the image data