Artificial intelligence (AI) refers to the use of computer algorithms to simulate human intelligence. The use of such intelligence has become common in the field of medicine. AI allows the processing of patient information at greater levels of speed and accuracy than that of a single physician.
“Artificial intelligence” is an umbrella term for all computer software–based algorithms used to complete a given task. The two major focuses of AI in this study include machine learning (ML) and deep learning (DL). Machine learning is a subset of AI that encodes and modifies its parameters in response to data. The goal of ML is to use a computer-generated algorithm to make predictions and responses to data. Arthur Samuel, a pioneer of ML, used it to train a computer to play checkers. That program was eventually able to beat the checkers world champion. Deep learning is a form of machine learning that focuses on using multiple algorithms to process input data and automatically identify patterns and structures within a given dataset. These multiple layers of algorithms form a neural network that aims to replicate the neural connectivity of the human brain. For example, a DL program can identify an image of a banana from other fruits after being trained via a set of neural networks that include images of bananas.
Ophthalmology lends itself well to AI integration as the field utilizes extensive digital imaging modalities and objective metrics such as foveal thickness or visual acuity. Electronic medical datasets combined with multi-modal digital images provide a large pool of ophthalmic patient information for analysis by AI. Moreover, the precision required when performing ophthalmic surgery may promote the application of robotic technologies to improve surgical outcomes and patient safety. Finally, the recent COVID-19 pandemic has created many hurdles for individuals seeking ophthalmic care. Teleophthalmology solutions integrated with AI algorithms are able to reduce clinician review times by screening large numbers of batched image files for pathology.
Herein, we provide an update on the current applications of AI in the field of ophthalmology and its various subspecialties.
Screening for diabetic retinopathy (DR) is essential as it facilitates early detection and treatment, thereby preventing vision loss. This is relevant in Canada as 3.7 million people have diabetic retinopathy; the incidence of DR is reported to be as high as 40% in at-risk populations, and a significant proportion of patients are not screened. DR is an optimally suited area for AI, which can help overcome screening barriers, improving access and preventing vision loss.
Early studies of AI and DR focused on lesion detection and have evolved classifying DR with a predominant focus on standard color fundus photography. In 2016, both Abràmoff et al. and Gulshan et al. reported algorithms using convolutional neural networks (CNN) that were able to detect referrable diabetic retinopathy (area under the curve [AUC] of 0.980 and 0.991, respectively). Subsequent studies used larger data sets demonstrated good detection of referrable diabetic retinopathy with AUCs of 0.97 and 0.94, respectively. Further studies have prospectively evaluated the performance of AI in detecting referrable DR. Heydon et al. reported that EyeArt v2.1 had a 95.7% sensitivity for referrable DR.
In addition to standard fundus photography, AI detection of DR has been studied using optical coherence tomography (OCT) images, ultra-widefield (UWF) imaging, and even smartphone-captured retinal images. Intraretinal fluid identified by OCT can be identified accurately by CNN; for instance Lee et al. used manually segmented macular OCT images to develop a CNN capable of detecting macular edema (with a cross-validation Dice coefficient of 0.911). UWF imaging allows visualization of up to 200° of the fundus, potentially catching additional diabetic-related peripheral disease. Nagasawa et al. found high sensitivity (94.7%) and specificity (97.2%) of a CNN in detecting proliferative DR on UWF images. Similarly, Wang et al. found high sensitivity (91.7%), though limited specificity (50.0%), for referrable DR using UWF images. Access and availability of imaging tools are challenges in effective DR screening. Natarajan et al. reported a smartphone-based, offline AI system that had a high sensitivity for detecting referrable diabetic retinopathy.
There are a number of commercially available AI-developed DR screening platforms including IDx-DR (Iowa), which holds FDA approval, and EyeArt (California), which is designated as a European Union Class IIa medical device.
Age-related macular degeneration
Age-related macular degeneration (AMD) is a common cause of vision loss, with an estimated 196 million patients impacted globally. Early detection and treatment of wet AMD can minimize vision loss. Given the burden of disease, AI could assist in mass screening of OCT and retinal photographs without in-person evaluations.
The research in this field started from ML with databases of under 1000 images to now over 490,000 images with high sensitivity and specificity rates. Burlina et al. used a database of over 130,000 images from 4613 patients to develop a DL algorithm for automated detection of AMD. Their DL system reported a 92% accuracy in identifying individuals with moderate and advanced AMD. Similarly, a study by Vaghefi et al. demonstrated that combining DL modalities in AMD—specifically fundus photographs, OCT, and OCT angiography scans—increased accuracy from 91% to 96% in detecting AMD compared to OCT alone.
Keenan et al. recently published a paper on an AI algorithm that could accurately quantify volume of fluid in neovascular AMD patients. This has potential in monitoring response to treatment. Deep learning has also been used to quantify other key features associated with AMD including intraretinal fluid (IRF), subretinal fluid (SRF), pigment epithelial detachment (PED), ellipsoid zone loss, drusen, fibrosis, and subretinal hyperreflective material. Similarly, Moraes et al. published a paper on automated quantification of key features in AMD while Fu et al. demonstrated that automatically captured quantitative parameters could predict visual change following treatment.
Additional applications in the retina
Moving beyond diagnosis of individual disease entities, De Fauw et al. reported a deep learning architecture that identified referrable retinal disease via OCT images, achieving a performance comparable to retina subspecialists (AUC = 99.21). This system was able to identify neovascular AMD, geographic atrophy, drusen, macular edema, macular holes, central serous retinopathy, vitreomacular traction, and epiretinal membrane.
Deep learning is able to predict retinal function on microperimetry based on structural assessment of OCT in patients with Stargardt disease. This may assist in assessing patients with inherited retinal disease while monitoring progression or treatment effect in clinical trials. Other AI systems are able to identify central serous retinopathy, pachychoroid vasculopathy, sickle cell disease, and macular telangiectasia. Aside from ocular diagnosis, DL can also predict demographics including age, gender, and cardiovascular risk factors such as systolic blood pressure, smoking status, and major adverse cardiac events.
Cornea and Anterior Segment
While AI has been heavily researched in the posterior segment, the application of AI in anterior segment disease and diagnostic research is now coming to the forefront of ophthalmology literature.
Using the Japan Ocular Allergy Society (JOAS) classification, Hiroki Masumoto trained a neural network to grade conjunctival hyperemia. The system graded the severity of the hyperemia with a high degree of accuracy.
Trachoma is a blinding disease secondary to infection by ocular strains of Chlamydia trachomatis. Using eyelid images from a database of two clinical trials—the Niger arm of the Partnership for Rapid Elimination of Trachoma trial (PRET) and the Trachoma Amelioration in Northern Amhara (TANA) trial—machine learning was used to accurately classify trachomatis changes.
Lacrimal scintigraphy (LS) is an objective and reliable method of studying the lacrimal drainage system and tear flow. Park et al. developed machine and deep learning algorithms using LS images to classify lacrimal duct pathology in patients with epiphora. The system showed accuracy comparable to a trained oculoplastic specialist.
Meibomian glands (MGs) are believed to play a critical role in ocular surface health. Dysfunction of MGs is the most frequent cause of dry eyes. Meibography, or photo documentation of MGs of the eyelids with transillumination or infrared light, is a common test for the diagnosis, treatment, and management of MG dysfunction (MGD). Wang et al. developed a DL approach to digitally segment MG atrophy and computing percent atrophy in meibography images, providing quantitative information of gland atrophy. The algorithm achieved a 95.6% meiboscore grading accuracy, outperforming the lead clinical investigator by 16.0% and the clinical team by 40.6%. The algorithm also achieved a 97.6% and 95.4% accuracy for eyelid and atrophy segmentations, respectively.
Stegman et al. developed a ML segmentation algorithm to measure tear meniscus thickness via OCT to measure tear film quantity. The system showed reproducible results although the sample size was small.
Keratoconus is a non-inflammatory corneal disorder characterized by stromal thinning and astigmatism. Kuo et al. retrospectively collected corneal topographic results over time to develop a DL algorithm to detect keratoconus. The model had fair accuracy for keratoconus screening, and furthermore, it predicted subclinical keratoconus. The sensitivity and specificity of all CNN models were over 0.90, and the AUC reached 0.995 in one of the three tested models.
Dos Santos et al. designed and trained a neural network (CorneaNet) to segment cornea OCT images. The algorithm measured the thickness of the three main layers, namely the epithelium, Bowman’s layer, and the middle stroma, in patients with keratoconus and those with healthy eyes. All models revealed very similar performances when identifying keratoconus and had a validation accuracy ranging from 99.45% to 99.57%. Lavric et al. devised the KeratoDetect, a neural network that achieved a high level of performance in detecting keratoconus from cornea topographies. With an accuracy of 99.33%, the author claimed that it could assist ophthalmologists in rapid screening of patients. Similarly, Kamiya et al. evaluated the diagnostic accuracy of six colored anterior segment OCT maps: anterior elevation, anterior curvature, posterior elevation, posterior curvature, total refractive power, and pachymetry map. DL was able to identify and classify keratoconus eyes and stage of disease. Shi et al. developed an automated classification system using ML and combining Scheimpflug and ultra-high resolution OCT. The system showed excellent performance (AUC = 0.93) in discriminating subclinical keratoconus from normal corneas. The author found that epithelial features assessed by OCT were the most important features when identifying keratoconus. Abdelmotaal et al. was able to identify keratoconus and subclinical keratoconus via ML using color-coded corneal maps obtained by a Scheimpflug camera.
Glaucoma is the second most common cause of irreversible blindness worldwide. Early detection of glaucoma has been shown to reduce vision loss. Digital photography of the optic nerve is a common method to screen for glaucoma and is used effectively as part of many teleglaucoma programs. Comprehensive evaluation of a glaucoma suspect might include spectral domain optical coherence tomography (SD-OCT), perimetry, tonometry, pachymetry, and gonioscopy. AI algorithms have been developed to identify optic nerve changes via optic disc photographs and SD-OCT and thereby predict glaucomatous field changes. Very little work has been done to date with DL for tonometry, pachymetry, and gonioscopy.
Glaucoma diagnosis and screening
Color fundus photography of the optic nerve is an inexpensive and available method to screen for glaucoma. ML has been utilized to improve identification of early glaucomatous changes to the optic nerve captured by photography. Computer segmentation of the optic nerve into disc, cup, and vessels to create a glaucoma score showed an AUC of 98.2% when compared to standard cup-to-disc ratio score (AUC 91.4%) on three glaucoma-related public datasets. Once incorporated into teleglaucoma screening programs, such algorithms will automate detection of early glaucoma.
Simultaneous or near-simultaneous capture of optic nerve OCT scans at the time of color fundus photography has enabled the rapid development of highly accurate DL algorithms for identification of glaucomatous nerve damage. This method of training DL systems—first published by Medeiros et al. in 2018 using high resolution digital images captured in the research setting—has removed the errors and biases associated with human grading of nerve damage. They showed that once trained with OCT data as truth, the DL system could discriminate glaucomatous optic nerves from healthy eyes with the area under the ROC curves of 0.944 (95% confidence interval [CI], 0.912–0.966) and 0.940 (95% CI, 0.902–0.966), respectfully (P = 0.724). This seminal work has been replicated and confirmed through publications by various groups around the world utilizing other fundus image repositories.
Machine-to-machine (M2M) deep learning algorithms trained with SD-OCT to assess monoscopic optic nerve photographs are able to identify glaucomatous optic nerve damage more accurately than glaucoma specialists. This increase in accuracy suggests that DL systems will replace human review of disc photographs for glaucoma screening programs of the future.
It is possible to detect progression of glaucomatous nerve damage by fundus photography utilizing DL algorithms confirmed by OCT. Medeiros et al. assessed temporally disparate disc photographs of 5529 patients over time. They utilized a DL CNN trained with OCT data from the same patients. The ROC curve area was 0.86 (95% CI, 0.83–0.88) when differentiating between progressors and non-progressors. Agitha et al. showed a similar benefit of a DL model used on 1113 fundus images to achieve an accuracy of 94%, sensitivity of 85%, and specificity of 100% in the automatic diagnosis of glaucoma.
In the setting of glaucoma detection, OCT has been utilized primarily to provide an objective truth reference for glaucoma-related DL CNN training. Recent work has shown that OCT DL algorithms are able to identify glaucomatous damage reliably by utilizing various datasets and various artificial intelligence algorithms. DL algorithms are also able to predict glaucomatous visual field with OCT nerve topography. The sensitivity and specificity of ML classifiers to diagnose glaucoma can be improved by combining standard automated perimetry and OCT data when compared to OCT alone.
Standard automated perimetry (SAP) is perhaps the most exciting area to assess with ML. The availability of long-term visual field test results, often over decades, in patients with and without glaucoma has provided extensive datasets for artificial intelligence researchers. Artificial intelligence is able to identify glaucoma four years in advance of diagnosis using original visual field data with good reliability. Asaoka et al. retrospectively assessed visual field data over 15 years in 51 patients with open-angle glaucoma and 87 healthy participants. Their deep feedforward neural network (FNN) showed an AUC of 92.6% (95% CI, 89.8%–95.4%) when identifying pre-perimetric glaucoma. Unsupervised ML classifiers showed a sensitivity of 82.8% and specificity of 93.1% in the identification of glaucomatous patterns by frequency doubling technology (FDT). This suggests that machine learning could become an important adjunct where visual field testing is performed as part of any glaucoma screening program.
Machine classifier algorithms are able to identify glaucoma progression by visual fields. Progression of patterns (POP) is a variational machine learning classifier that was able to identify more eyes with progression of glaucomatous optic neuropathy in glaucoma suspects and glaucoma than were identified by guided progression analysis (GPA). Deep learning is also able to forecast future Humphrey visual fields in patients with glaucoma. Wen et al. was able to predict the development of future visual field changes up to five years in the future using deep learning networks.
Retinopathy of prematurity
Retinopathy of prematurity (ROP) is a vasoproliferative retinal disease that is a leading cause of childhood blindness. The Early Treatment for Retinopathy of Prematurity (ETROP) study has shown that screening and early intervention is critical for improving visual outcomes. Improved survival of extremely premature infants has increased the prevalence of ROP, particularly in developing nations. AI can play a vital role in assisting with ROP diagnosis, thereby improving treatment outcomes. In a study by Brown et al., researchers showed that a DL algorithm trained with wide-field retinal photographs outperformed 6 out of 8 ROP experts on an independent data set of 100 images in diagnosing ROP. The algorithm was trained with a database of 5511 fundus images and demonstrated a 93% sensitivity and 94% specificity in determining ROP severity. Tong et al. developed a neural network that was trained for ROP identification with 36,000 fundus images. This system achieved an accuracy of 0.903 for ROP severity classification and demonstrated comparable to or better diagnostic ability when compared to retina subspecialist. Other research have shown similar success with ROP severity classification and deep learning. Integration of AI into an ROP screening program will likely occur in the near future.
Pediatric cataract is one of the leading causes of juvenile blindness, with an estimated prevalence of 4.24 per 10,000 live births.
Congenital cataract guardian (CC-Guardian) is an AI agent that incorporates individualized prediction and scheduling, and intelligent telehealth follow-up computing for congenital cataracts. The system exhibits high sensitivity and specificity and has been integrated to a web-based smartphone app. The intelligent agent consists of three functional modules: (i) a prediction module that identifies potential high-risk congenital cataract patients who are likely to suffer complications, (ii) a dispatching module that schedules individual follow-up based on the prediction results, and (iii) a telehealth module that makes intervention decisions in each follow-up examination. All the records were derived from routine examinations at the Childhood Cataract Program of the Chinese Ministry of Health.
In Korea, Chun et al. assessed a DL system to predict the range of refractive error in children using a smartphone photorefraction image to screen for amblyopia and compared it to a cycloplegic refraction. The DL tool showed an accuracy of 81.6%.
The development and modern usage of AI in research has become a breakthrough for optimization and efficiency. With the growth of electronic medical records, healthcare providers and hospitals are able to accumulate a wealth of patient information. A common barrier to sifting through this information is the time required to appropriately review each individual item. With the advent of AI, after developing a computer-generated algorithm or suitably training an automated system to batch patient information, data collection can be completed in a fraction of the time that it would take to be done manually. Ophthalmology is a medical specialty that is conducive to retrieving these large amounts of data due to its rapid access of ophthalmic imaging and objective markers (e.g., visual acuity, intraocular pressure [IOP], retinal thickness, etc.). The Intelligent Research in Sight (IRIS) Registry is one of the largest clinical datasets that includes data about demographics, disease conditions, and visit rates in ophthalmology. The Smart Eye Database stores electronic medical records of ophthalmology patients which are stratified by eye conditions. Datasets such as IRIS and the Smart Eye Database allow us to appreciate subtle correlations, conduct multicenter studies, incorporate multimodal analyses, identify novel imaging patterns, and increase the power in studies, all of which may not be possible with smaller sets of data. As described by Joshi et al., this large collection of medical information, or “big data,” serves as a perfect substrate for AI, ML, and DL to develop and run algorithms at a scale that would never have been possible before.
Ophthalmology is a specialty well-suited for AI integration. The extensive use of multi-modal digital imaging and diagnostic tests captured over time in all ophthalmology subspecialties provide a treasure trove of opportunities for machine learning that are now being realized. Artificial intelligence and machine learning solutions have begun the evolution from research setting to a clinical tool that will be invaluable for ophthalmologists in all clinical settings.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
1. Briganti G, Le Moine O Artificial intelligence in medicine:Today and tomorrow. Front Med 2020; 7:27 doi:10.3389/fmed.2020.00027.
2. Wiederhold G, McCarthy J Arthur Samuel:Pioneer in machine learning. IBM J Res Dev 1992; 36:329–31.
3. Carneiro VLA, Andrade H, Matias L, de Sousa RARC Post-COVID-19 and the Portuguese national eye care system challenge. J Optom 2020; 13:257–61.
4. Chandra A, Romano MR, Ting DS, Chao DL Implementing the new normal in ophthalmology care beyond COVID-19. Eur J Ophthalmol 2021; 31:321–7.
5. Nilforushan N, Abolfathzadeh N The impact of the COVID-19 pandemic on ophthalmology residency training. J Ophthalmic Vis Res 2021; 16:312.
6. Nikolaidou A, Tsaousis KT Teleophthalmology and artificial intelligence as game changers in ophthalmic care after the COVID-19 pandemic. Cureus 2021; 13:e16392 doi:10.7759/cureus. 16392.
7. Hooper P, Boucher MC, Cruess A, Dawson KG, Delpero W, Greve M, et al. Excerpt from the Canadian Ophthalmological Society evidence-based clinical practice guidelines for the management of diabetic retinopathy. Can J Ophthalmol 2017; 52:S45–74.
8. Ross SA, McKenna A, Mozejko S, Fick GH Diabetic retinopathy in native and nonnative Canadians. Exp Diabetes Res 2008; 2007:76271 doi:10.1155/2007/76271.
9. Grewal PS, Oloumi F, Rubin U, Tennant MT Deep learning in ophthalmology:A review. Can J Ophthalmol 2018; 53:309–13.
10. Kanjee R, Dookeran RI Tele-ophthalmology for diabetic retinopathy in Canada—meeting the needs of a growing epidemic. Can J Ophthalmol 2017; 52:S13–4.
11. Abràmoff MD, Lou Y, Erginay A, Clarida W, Amelon R, Folk JC, et al. Improved automated detection of diabetic retinopathy on a publicly available dataset through integration of deep learning. Investig Ophthalmol Vis Sci 2016; 57:5200–6.
12. Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 2016; 316:2402–10.
13. Archer D, Amoaku W, Gardiner T Radiation retinopathy—clinical, histopathological, ultrastructural and experimental correlations. Eye 1991; 5:239–51.
14. Ting DSW, Cheung CY-L, Lim G, Tan GSW, Quang ND, Gan A, et al. Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes. JAMA 2017; 318:2211–23.
15. Heydon P, Egan C, Bolter L, Chambers R, Anderson J, Aldington S, et al. Prospective evaluation of an artificial intelligence-enabled algorithm for automated diabetic retinopathy screening of 30 000 patients. Br J Ophthalmol 2021; 105:723–8.
16. Lee CS, Baughman DM, Lee AY Deep learning is effective for classifying normal versus age-related macular degeneration OCT images. Ophthalmol Retina 2017; 1:322–7.
17. Nagasawa T, Tabuchi H, Masumoto H, Enno H, Niki M, Ohara Z, et al. Accuracy of ultrawide-field fundus ophthalmoscopy-assisted deep learning for detecting treatment-naïve proliferative diabetic retinopathy. Int Ophthalmol 2019; 39:2153–9.
18. Wang X, Ji Z, Ma X, Zhang Z, Yi Z, Zheng H, et al. Automated grading of diabetic retinopathy with ultra-widefield fluorescein angiography and deep learning. J Diabetes Res 2021; 2021:2611250 doi:10.1155/2021/2611250.
19. Natarajan S, Jain A, Krishnan R, Rogye A, Sivaprasad S Diagnostic accuracy of community-based diabetic retinopathy screening with an offline artificial intelligence system on a smartphone. JAMA Ophthalmol 2019; 137:1182–8.
20. Grzybowski A, Brona P Analysis and comparison of two artificial intelligence diabetic retinopathy screening algorithms in a pilot study:IDx-DR and retinalyze. J Clin Med 2021; 10:2352.
21. Grzybowski A, Brona P, Lim G, Ruamviboonsuk P, Tan GS, Abramoff M, et al. Artificial intelligence for diabetic retinopathy screening:A review. Eye 2020; 34:451–60.
22. Wong WL, Su X, Li X, Cheung CMG, Klein R, Cheng C-Y, et al. Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040 :A systematic review and meta-analysis. Lancet Glob Health 2014; 2:e106–16.
23. Grassmann F, Mengelkamp J, Brandl C, Harsch S, Zimmermann ME, Linkohr B, et al. A deep learning algorithm for prediction of age-related eye disease study severity scale for age-related macular degeneration from color fundus photography. Ophthalmology 2018; 125:1410–20.
24. Burlina PM, Joshi N, Pekala M, Pacheco KD, Freund DE, Bressler NM Automated grading of age-related macular degeneration from color fundus images using deep convolutional neural networks. JAMA Ophthalmol 2017; 135:1170–6.
25. Akkara JD, Kuriakose A Role of artificial intelligence and machine learning in ophthalmology. Kerala J Ophthalmol 2019; 31:150.
26. Vaghefi E, Hill S, Kersten HM, Squirrell D Multimodal retinal image analysis via deep learning for the diagnosis of intermediate dry age-related macular degeneration:A feasibility study. J Ophthalmol 2020; 2020:7493419 doi:10.1155/2020/7493419.
27. Keenan TD, Chakravarthy U, Loewenstein A, Chew EY, Schmidt-Erfurth U Automated quantitative assessment of retinal fluid volumes as important biomarkers in neovascular age-related macular degeneration. Am J Ophthalmol 2021; 224:267–81.
28. Liefers B, Colijn JM, González-Gonzalo C, Verzijden T, Wang JJ, Joachim N, et al. A deep learning model for segmentation of geographic atrophy to study its long-term natural history. Ophthalmology 2020; 127:1086–96.
29. Moraes G, Fu DJ, Wilson M, Khalid H, Wagner SK, Korot E, et al. Quantitative analysis of OCT for neovascular age-related macular degeneration using deep learning. Ophthalmology 2021; 128:693–705.
30. Fu DJ, Faes L, Wagner SK, Moraes G, Chopra R, Patel PJ, et al. Predicting incremental and future visual change in neovascular age-related macular degeneration using deep learning. Ophthalmol Retina 2021; 5:1074–84.
31. De Fauw J, Ledsam JR, Romera-Paredes B, Nikolov S, Tomasev N, Blackwell S, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med 2018; 24:1342–50.
32. Müller PL, Odainic A, Treis T, Herrmann P, Tufail A, Holz FG, et al. Inferred retinal sensitivity in recessive Stargardt disease using machine learning. Sci Rep 2021; 11:1–11.
33. Ko J, Han J, Yoon J, Park JI, Hwang JS, Han JM, et al. Assessing central serous chorioretinopathy with deep learning and multiple optical coherence tomography images. Sci Rep 2022; 12:1–8 doi:10.1038/s41598-022-05051-y.
34. Chou Y-B, Hsu C-H, Chen W-S, Chen S-J, Hwang D-K, Huang Y-M, et al. Deep learning and ensemble stacking technique for differentiating polypoidal choroidal vasculopathy from neovascular age-related macular degeneration. Sci Rep 2021; 11:1–9 doi:10.1038/s41598-021-86526-2.
35. Cai S, Han IC, Scott AW Artificial intelligence for improving sickle cell retinopathy diagnosis and management. Eye 2021; 35:2675–84.
36. Loo J, Cai CX, Choong J, Chew EY, Friedlander M, Jaffe GJ, et al. Deep learning-based classification and segmentation of retinal cavitations on optical coherence tomography images of macular telangiectasia type 2. Br J Ophthalmol 2022; 106:396–402.
37. Poplin R, Varadarajan AV, Blumer K, Liu Y, McConnell MV, Corrado GS, et al. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat Biomed Eng 2018; 2:158–64.
38. Masumoto H, Tabuchi H, Yoneda T, Nakakura S, Ohsugi H, Sumi T, et al. Severity classification of conjunctival hyperaemia by deep neural network ensembles. J Ophthalmol 2019; 2019:7820971 doi:10.1155/2019/7820971.
39. Kim MC, Okada K, Ryner AM, Amza A, Tadesse Z, Cotter SY, et al. Sensitivity and specificity of computer vision classification of eyelid photographs for programmatic trachoma assessment. PloS One 2019; 14:e0210463 doi:10.1371/journal.pone.0210463.
40. Park Y-J, Bae JH, Shin MH, Hyun SH, Cho YS, Choe YS, et al. Development of predictive models in patients with epiphora using lacrimal scintigraphy and machine learning. Nucl Med Mol Imaging 2019; 53:125–35.
41. Chhadva P, Goldhardt R, Galor A Meibomian gland disease:The role of gland dysfunction in dry eye disease. Ophthalmology 2017; 124:S20–6.
42. Wang J, Yeh TN, Chakraborty R, Stella XY, Lin MC A deep learning approach for meibomian gland atrophy evaluation in meibography images. Transl Vis Sci Technol 2019; 8:37 doi:10.1167/tvst.8.6.37.
43. Stegmann H, Werkmeister RM, Pfister M, Garhöfer G, Schmetterer L, Dos Santos VA Deep learning segmentation for optical coherence tomography measurements of the lower tear meniscus. Biomed Optics Express 2020; 11:1539–54.
44. Kuo B-I, Chang W-Y, Liao T-S, Liu F-Y, Liu H-Y, Chu H-S, et al. Keratoconus screening based on deep learning approach of corneal topography. Transl Vis Sci Technol 2020; 9:53.
45. Dos Santos VA, Schmetterer L, Stegmann H, Pfister M, Messner A, Schmidinger G, et al. CorneaNet:Fast segmentation of cornea OCT scans of healthy and keratoconic eyes using deep learning. Biomed Optics Express 2019; 10:622–41.
46. Lavric A, Valentin P KeratoDetect:Keratoconus detection algorithm using convolutional neural networks. Comput Intell Neurosci 2019; 2019:8162567 doi:10.1155/2019/8162567.
47. Kamiya K, Ayatsuka Y, Kato Y, Fujimura F, Takahashi M, Shoji N, et al. Keratoconus detection using deep learning of colour-coded maps with anterior segment optical coherence tomography:A diagnostic accuracy study. BMJ Open 2019; 9:e031313 doi:10.1136/bmjopen-2019-031313.
48. Shi C, Wang M, Zhu T, Zhang Y, Ye Y, Jiang J, et al. Machine learning helps improve diagnostic ability of subclinical keratoconus using Scheimpflug and OCT imaging modalities. Eye Vis 2020; 7:1–12 doi:10.1186/s40662-020-00213-3.
49. Abdelmotaal H, Mostafa MM, Mostafa AN, Mohamed AA, Abdelazeem K Classification of color-coded Scheimpflug camera corneal tomography images using deep learning. Transl Vis Sci Technol 2020; 9:30.
50. Flaxman SR, Bourne RRA, Resnikoff S, Ackland P, Braithwaite T, Cicinelli MV, et al. Global causes of blindness and distance vision impairment 1990-2020:A systematic review and meta-analysis. Lancet Glob Health 2017; 5:e1221–34.
51. Leske MC, Heijl A, Hussein M, Bengtsson B, Hyman L, Komaroff E, et al. Factors for glaucoma progression and the effect of treatment:The early manifest glaucoma trial. Arch Ophthalmol 2003; 121:48–56.
52. Thomas SM, Jeyaraman MM, Hodge WG, Hutnik C, Costella J, Malvankar-Mehta MS The effectiveness of teleglaucoma versus in-patient examination for glaucoma screening:A systematic review and meta-analysis. PLoS One 2014; 9:e113779 doi:10.1371/journal.pone.0113779.
53. Fu H, Baskaran M, Xu Y, Lin S, Wong DWK, Liu J, et al. A deep learning system for automated angle-closure detection in anterior segment optical coherence tomography images. Am J Ophthalmol 2019; 203:37–45.
54. Spaide T, Wu Y, Yanagihara RT, Feng S, Ghabra O, Yi JS, et al. Using deep learning to automate goldmann applanation tonometry readings. Ophthalmology 2020; 127:1498–506.
55. Girard F, Hurtut T, Kavalec C, Cheriet F Atlas-based score for automatic glaucoma risk stratification. Comput Med Imaging Graph 2020; 87:101797 doi:10.1016/j.compmedimag.2020.101797.
56. Medeiros FA, Jammal AA, Thompson AC From machine to machine:An OCT-trained deep learning algorithm for objective quantification of glaucomatous damage in fundus photographs. Ophthalmology 2019; 126:513–21.
57. Yang HK, Kim YJ, Sung JY, Kim DH, Kim KG, Hwang JM Efficacy for differentiating nonglaucomatous versus glaucomatous optic neuropathy using deep learning systems. Am J Ophthalmol 2020; 216:140–6.
58. Orlando JI, Fu H, Barbosa Breda J, van Keer K, Bathula DR, Diaz-Pinto A, et al. REFUGE challenge:A unified framework for evaluating automated methods for glaucoma assessment from fundus photographs. Med Image Anal 2020; 59:101570 doi:10.1016/j.media.2019.101570.
59. Thompson AC, Jammal AA, Medeiros FA A deep learning algorithm to quantify neuroretinal rim loss from optic disc photographs. Am J Ophthalmol 2019; 201:9–18.
60. Jammal AA, Thompson AC, Mariottoni EB, Berchuck SI, Urata CN, Estrela T, et al. Human versus machine:Comparing a deep learning algorithm to human gradings for detecting glaucoma on fundus photographs. Am J Ophthalmol 2020; 211:123–31.
61. Medeiros FA, Jammal AA, Mariottoni EB Detection of progressive glaucomatous optic nerve damage on fundus photographs with deep learning. Ophthalmology 2021; 128:383–92.
62. Ajitha S, Akkara JD, Judy M Identification of glaucoma from fundus images using deep learning techniques. Indian J Ophthalmol 2021; 69:2702–9.
63. Ran AR, Cheung CY, Wang X, Chen H, Luo LY, Chan PP, et al. Detection of glaucomatous optic neuropathy with spectral-domain optical coherence tomography:A retrospective training and validation deep-learning analysis. Lancet Digit Health 2019; 1 e172-e82.
64. Garcia G, Colomer A, Naranjo V Glaucoma detection from raw SD-OCT volumes:A novel approach focused on spatial dependencies. Comput Methods Programs Biomed 2021 Mar 200:105855 doi:10.1016/j.cmpb.2020.105855.
65. Mariottoni EB, Datta S, Dov D, Jammal AA, Berchuck SI, Tavares IM, et al. Artificial intelligence mapping of structure to function in glaucoma. Transl Vis Sci Technol 2020; 9:19.
66. Silva FR, Vidotti VG, Cremasco F, Dias M, Gomi ES, Costa VP Sensitivity and specificity of machine learning classifiers for glaucoma diagnosis using Spectral Domain OCT and standard automated perimetry. Arq Bras Oftalmol 2013; 76:170–4.
67. Thakur A, Goldbaum M, Yousefi S Convex representations using deep archetypal analysis for predicting glaucoma. IEEE J Transl Eng Health Med 2020; 8:3800107 doi:10.1109/JTEHM.2020.2982150.
68. Asaoka R, Murata H, Iwase A, Araie M Detecting Preperimetric glaucoma with standard automated perimetry using a deep learning classifier. Ophthalmology 2016; 123:1974–80.
69. Bowd C, Weinreb RN, Balasubramanian M, Lee I, Jang G, Yousefi S, et al. Glaucomatous patterns in frequency doubling technology (FDT) perimetry data identified by unsupervised machine learning classifiers. PLoS One 2014; 9:e85941 doi:10.1371/journal.pone.0085941.
70. Goldbaum MH, Lee I, Jang G, Balasubramanian M, Sample PA, Weinreb RN, et al. Progression of patterns (POP):A machine classifier algorithm to identify glaucoma progression in visual fields. Invest Ophthalmol Vis Sci 2012; 53:6557–67.
71. Wen JC, Lee CS, Keane PA, Xiao S, Rokem AS, Chen PP, et al. Forecasting future humphrey Visual Fields using deep learning. PLoS One 2019; 14:e0214875.
72. Solebo AL, Teoh L, Rahi J Epidemiology of blindness in children. Arch Dis Childhood 2017; 102:853–7.
73. Good WV, Hardy RJ, Dobson V, Palmer EA, Phelps DL, et al.Early Treatment for Retinopathy of Prematurity Cooperative Group Final visual acuity results in the early treatment for retinopathy of prematurity study. Arch Ophthalmol 2010; 128:663–71.
74. Brown JM, Campbell JP, Beers A, Chang K, Ostmo S, Chan RP, et al. Automated diagnosis of plus disease in retinopathy of prematurity using deep convolutional neural networks. JAMA Ophthalmol 2018; 136:803–10.
75. Tong Y, Lu W, Deng Q-q, Chen C, Shen Y Automated identification of retinopathy of prematurity by image-based deep learning. Eye Vis 2020; 7:1–12 doi:10.1186/s40662-020-00206-2.
76. Mulay S, Ram K, Sivaprakasam M, Vinekar A Early detection of retinopathy of prematurity stage using deep learning approach Medical Imaging 2019:Computer-Aided Diagnosis. International Society for Optics and Photonics 2019.
77. Zhao J, Lei B, Wu Z, Zhang Y, Li Y, Wang L, et al. A deep learning framework for identifying zone I in RetCam images. IEEE Access 2019; 7:103530–7.
78. Taylor S, Brown JM, Gupta K, Campbell JP, Ostmo S, Chan RP, et al. Monitoring disease progression with a quantitative severity scale for retinopathy of prematurity using deep learning. JAMA Ophthalmol 2019; 137:1022–8.
79. Redd TK, Campbell JP, Brown JM, Kim SJ, Ostmo S, Chan RVP, et al. Evaluation of a deep learning image assessment system for detecting severe retinopathy of prematurity. Br J Ophthalmol 2019; 103:580–4.
80. Long E, Chen J, Wu X, Liu Z, Wang L, Jiang J, et al. Artificial intelligence manages congenital cataract with individualized prediction and telehealth computing. NPJ Digit Med 2020; 3:1–10 doi:10.1038/s41746-020-00319-x.
81. Chun J, Kim Y, Shin KY, Han SH, Oh SY, Chung T-Y, et al. Deep Learning–Based Prediction of Refractive Error Using Photorefraction Images Captured by a Smartphone:Model Development and Validation Study. JMIR medical informatics 2020; 8 5 e16225.
82. Chiang MF, Sommer A, Rich WL, Lum F, Parke DW II The 2016 American Academy of Ophthalmology IRIS®registry (Intelligent Research in Sight) database:Characteristics and methods. Ophthalmology 2018; 125:1143–8.
83. Kortüm KU, Müller M, Kern C, Babenko A, Mayer WJ, Kampik A, et al. Using electronic health records to build an ophthalmologic data warehouse and visualize patients'data. Am J Ophthalmol 2017; 178:84–93.
84. Lee CS, Brandt JD, Lee AY Big data and artificial intelligence in ophthalmology:Where are we now?Ophthalmol Sci 2021; 1 doi:10.1016/j.xops.2021.100036.
85. Joshi S, Vibhute G, Ayachit A, Ayachit G Big data and artificial intelligence-Tools to be future ready?. Indian J Ophthalmol 2021; 69:1652–3.