Current status and practical considerations of artificial intelligence use in screening and diagnosing retinal diseases: Vision Academy retinal expert consensus : Current Opinion in Ophthalmology

Secondary Logo

Journal Logo

ARTIFICIAL INTELLIGENCE/BIG DATA: Edited by Daniel Ting and Ehsan Rahimy

Current status and practical considerations of artificial intelligence use in screening and diagnosing retinal diseases: Vision Academy retinal expert consensus

Chou, Yu-Baia,b,∗; Kale, Aditya U.c,∗; Lanzetta, Paolod,e,∗; Aslam, Tariqf,∗; Barratt, Janeg; Danese, Carlad,h; Eldem, Borai; Eter, Nicolej; Gale, Richardk; Korobelnik, Jean-Françoisl,m; Kozak, Igorn; Li, Xiaorongo; Li, Xiaoxinp; Loewenstein, Anatq; Ruamviboonsuk, Paisanr; Sakamoto, Taijis; Ting, Daniel S.W.t; van Wijngaarden, Peteru,v; Waldstein, Sebastian M.w; Wong, Davidx; Wu, Lihtehy; Zapata, Miguel A.z; Zarranz-Ventura, Javieraa

Author Information
Current Opinion in Ophthalmology 34(5):p 403-413, September 2023. | DOI: 10.1097/ICU.0000000000000979


    Purpose of review 

    The application of artificial intelligence (AI) technologies in screening and diagnosing retinal diseases may play an important role in telemedicine and has potential to shape modern healthcare ecosystems, including within ophthalmology.

    Recent findings 

    In this article, we examine the latest publications relevant to AI in retinal disease and discuss the currently available algorithms. We summarize four key requirements underlining the successful application of AI algorithms in real-world practice: processing massive data; practicability of an AI model in ophthalmology; policy compliance and the regulatory environment; and balancing profit and cost when developing and maintaining AI models.


    The Vision Academy recognizes the advantages and disadvantages of AI-based technologies and gives insightful recommendations for future directions.


    The initial concept of artificial intelligence (AI) was first coined as far back as 1956 [1]. The concept of “machine learning” and “deep learning” (DL) were proposed subsequently and demonstrated great potential in computer learning and decision-making via various data training techniques [2–4]. AlexNet, which won the ImageNet Large Scale Visual Recognition Challenge in 2012, set a milestone for DL algorithms to handle large imaging data sets [5]. Since then, the application of DL algorithms to color fundus photography has been adopted for the diagnosis and monitoring of many retinal diseases, including diabetic retinopathy (DR) [6–10], age-related macular degeneration (AMD) [10,11], and retinopathy of prematurity [12]. Moreover, wide-field fundus photography and autofluorescence imaging have been used to differentiate not only referable DR and AMD but retinal vein occlusion, pathologic myopia, retinal detachment, vitreomacular interface disease, pathologic myopia, sickle cell retinopathy, and inherited retinal diseases by DL-based models [13–19].

    In recent years, many studies have applied AI algorithms to the most modern retinal imaging modalities, including optical coherence tomography, to detect or quantify an array of retinal features of interest (e.g. retinal fluid) [20–23]. AI-enabled technologies, which use not only different types of images but also different data modalities (e.g. structured medical data [24] and genomics data sets [25,26]), have also proven to demonstrate robust outcomes.

    Such AI-enabled technologies have potential to be implemented into clinical practice in several ways. The use of AI technologies to screen or classify retinal diseases may play a role in telemedicine. They may also assist healthcare providers with greater speed, repeatability, reproducibility, and consistency than human graders. Uniting clinicians with AI systems has been proven to be synergistic, achieving better performance than either alone [27]. Therefore, AI-enabled technology can help clinicians achieve rapid and accurate decision-making.

    Academic institutions and technology companies (e.g. Google) increasingly engage in AI research and boost their investment and involvement in this field [7]. Furthermore, the U.S. National Science and Technology Council's Committee on Technology noted that investments from the U.S. government in AI-enabled technologies were nearly $1.1 billion in 2015 and continue to increase [28].

    The key issues for deploying AI technologies in telemedicine or healthcare systems may have a profound and lasting influence on near-future practice in ophthalmology.

    In this article, we summarize four key requirements surrounding the application and execution of AI-enabled technology for diagnosis and screening in retinal diseases in real-world practice. Informing and operationalizing an AI healthcare system includes processing large data sets, practicability in ophthalmology, policy compliance and regulatory environment, and balancing profit and cost in adopting AI-enabled technologies. 

    Box 1:
    no caption available


    Data processing is crucial before developing an AI model. It includes data standardization, data sharing system, and the maintenance of data privacy in the infrastructure of AI systems.

    Data standardization in ophthalmology

    The high dependency of modern ophthalmology on imaging makes it an attractive field for the development of AI models. However, the diversity of proprietary devices, image acquisition, and data storage processes poses a barrier to research teams. The need for data standardization has become pivotal, not only for expanding the scale of AI models but for providing more effective ways to achieve clinical benefits. In 2021, the American Academy of Ophthalmology suggested that manufacturers of ophthalmic devices should standardize the format of digital images, integration of medical data, and picture archiving to comply with the 12 Digital Imaging and Communications in Medicine standards, developed by the American Academy of Ophthalmology in collaboration with manufacturers [29]. Such standardization of file formats in proprietary databases may streamline further analysis and increase the interoperability of different cameras or devices [30].

    Data sharing and privacy: are the protections adequate?

    As of 2021, nearly 94 ophthalmic data sets containing more than 500 000 images were openly accessible [31]. The transparency, accessibility, and limitations of each data set should be carefully shared and reported because these factors can affect the ground truth of image processing, not to mention the extensibility of AI algorithms.

    The shared data should be deidentified and anonymized to comply with privacy and cybersecurity frameworks. Data privacy poses challenges in technological, legal, and ethical fields. It can be difficult to precisely define data privacy because traditional deidentification is vulnerable to linkage attacks from intended third parties [32]. Moreover, fundus images are now considered uniquely recognizable information due to AI-enabled technology [33]. Awareness of such biometric identification, which may breach privacy rights, is crucial in data processing. Synthetic data generation (e.g. generative adversarial network) is a feasible way of creating plausible images for AI training while also maintaining confidentiality and provides an anonymization technique for data privacy [34,35].

    Data privacy may also be addressed by collating all relevant data into trusted research environments or data decentralization. In federated analysis, the algorithmic code is sent to each data site for individual analysis; then the results are brought back to the central site for aggregation and further analysis [36]. Recently, swarm learning was proposed to provide blockchain-based peer-to-peer data security. In contrast to federated learning, which requires a central analytic server, swarm learning produces complete data decentralization [37]. Compared with traditional centralized data, these new AI-enabled technologies with data decentralization can preserve privacy by retaining the data in each institution while still achieving similar outcomes [38].

    Ethical considerations and legal liability

    In 2021, the American Academy of Ophthalmology Committee on Artificial Intelligence raised three ethical concerns: transparency, meaning the adequate explanation or interpretation of the AI model; responsibility, which addresses moral or legal concerns; and scalability of implementation of AI models, which depends on equality of data distribution and potential systemic bias in AI models [39▪]. AI technologies may change relationships between physicians, healthcare organizations, and patients. However, there is still no universal guidance for legal liability. The American Medical Association recommends that developers should take legal liability and maintain insurance for systemic failure or misdiagnosis from an autonomous AI system [40]. The more autonomous the design of an AI model is, the more reinforcement is needed in terms of legal liability [41]. Different entities may share liability: physician errors belong to negligence liability; healthcare organization errors belong to vicarious liability; and manufacturer errors are attributed to product liability and incomplete disclosure of the actual functions and limitations of an AI model [42]. These entities should take responsibility to compensate for financial and physical loss to the injured party [43].


    As the field of ophthalmic AI evolves, the quality of reporting results from different AI systems may be discrepant and incomprehensive. It is necessary to have consensus in determining adequate description, translation, and appraisal of ophthalmic AI research to ensure robust algorithms and generalizability into real-world settings.

    Role of physicians and researchers in clinical trials

    In 2020, CONSORT-AI (CONsolidated Standards Of Reporting Trials–Artificial Intelligence) and SPIRIT-AI (Standard Protocol Items: Recommendations for Interventional Trials–Artificial Intelligence) were announced to provide reporting guidance in AI-related clinical trials [44▪▪]. Subsequently, other reporting guidelines (i.e. STARD-AI, DECIDE-AI, and TRIPOD-AI) further emphasized the transparency of AI technologies in healthcare. These have presented new standards for evaluating the results from AI-related clinical trials [44▪▪,45▪▪,46] and allowed for clear evidence generation and decision making.

    Role of healthcare providers in telemedicine

    Healthcare equity has been a major concern due to the imbalanced distribution of resources between urban and rural areas. The deployment of AI screening algorithms accompanied by well-established infrastructure of cybersecurity, including cloud-based systems or even home-based devices, can facilitate the adoption of AI technologies and reduce the medical resources gap between urban and rural areas. On the contrary, the quality of data-driven technologies may be affected by the inequality of data distribution (e.g. differences between racial and ethnic groups), which reduces the generalizability of the AI models to specific populations due to the scarcity of related data. Overemphasizing an AI system without carefully considering the condition of health data poverty has potential to cause harm [47]. Even if AI-enabled technologies represent an opportunity to overcome some challenges in rural areas, the general application and clinical interpretation of AI models should be treated with caution.


    Regulatory considerations for AI medical devices or software should include data security and sourcing, the design and development of algorithms, and evidence generation from AI-enabled technologies. Governmental regulatory bodies should provide clear guidance regarding the evidence requirements for AI medical devices and, to streamline the process of training, continuous education, and relicensing [43].

    For data privacy and confidentiality, in 2016 and 2017 the EU introduced the General Data Protection Regulation 2016/679 [48], the EU Network and Information Security Directive 2016/1148 [49], and the Medical Devices Regulation 2017/745 [50,51]. In the United States, the Health Insurance Portability and Accountability Act of 1996 covers confidentiality issues in medical data [43].

    For adopting and marketing AI-based instruments and algorithms, different regulatory groups are responsible for ensuring the security and safety of the products. The U.S. Food and Drug Administration (FDA) announced the Digital Health Innovation Action Plan to streamline the premarket review and to outline its approach to AI-based frameworks, known as “software as medical device” [52], which is a term defined by the International Medical Device Regulators Forum [42,53]. With regards to the risk of AI products, three risk classes represent safety and effectiveness for patients: class I (low risk); class II (moderate risk); and class III (high risk). AI medical devices must be evaluated rigorously through appropriate regulatory pathways. FDA regulatory pathways include the 510(k), De Novo Classification, and Premarket Approval. The appropriate pathway is determined by the risk class of the AI medical device and whether there is a predicate device available on the market already [54].

    In contrast to the United States, which adopts more market-oriented regulations, the EU takes a more customer-oriented approach to build the framework for AI-enabled technologies, with the Conformitè Européenne playing a critical role in the licensing of AI products [54]. Many other regulatory parties from other countries are involved in licensing AI-enabled technologies, such as the UK Conformity Assessed mark and the Chinese National Medical Products Administration.


    For adoption of AI technologies, achieving a practical balance between profit and cost is another important issue. Premarketing costs include significant effort and workforce in data collection, research development, and validation. Postmarketing costs include upgrading software, sustaining hardware over the long term, training operators, and incorporating new patient data [55,56]. Funding support can ensure the adoption and ongoing maintenance of AI services, but only when a balance between profit and cost is reached can the development of AI steadily progress.

    The cost of adopting AI-enabled technologies should be balanced between manufacturing price and reimbursements as it is considered by and for healthcare providers. In the example of Singapore's national DR screening programs, cost savings of approximately U.S. $21.9 million were achieved for a group of 170 000 patients with diabetes who underwent AI-assisted screening [57]. In the UK, the National Institute for Health and Care Excellence set up the Evidence Standards Framework to enable assessment of digital health technologies and guide government, developers, and healthcare providers on the level of evidence for economic and clinical evaluation [58]. Such guidance can motivate the application and prevent the overuse of AI-enabled technologies in the healthcare ecosystem. Government, industry, and academia will be the iron triangle for the future implementation of AI-enabled technologies in screening and diagnosing retinal diseases.


    Current AI-enabled systems with regulatory compliance are outlined in Table 1, with further details on their performance provided in Table 2.

    Table 1 - Summary of current AI systems with regulatory approval for different retinal diseases
    AI system Type of application Year approved Target disease Imaging modality Target user Regulatory status in the United States, Europe, and elsewhere
    Intelligent Retinal Imaging Systems (IRIS; Intelligent Retinal Imaging Systems, Pensacola, FL, USA) [59] Cloud-based algorithm 2015 Screen for DR Fundus photography Healthcare provider FDA clearance (class II)
    Automated Retinal Disease Assessment (ARDA; Google LLC, Mountain View, CA, USA) [60] Cloud-based algorithm 2016 Screen for referable DR Fundus photography Healthcare provider CE mark
    SELENA+ (EyRIS Pte Ltd, Singapore) [61] Cloud-based algorithm 2019 and 2020 Screen for vision-threatening DR, suspected glaucoma, AMD Fundus photography Healthcare provider CE mark, HAS approval (Singapore)
    IDx-DR (Digital Diagnostics Inc., Coralville, IA, USA) [62] Device and cloud-based algorithm 2018 Screen for referable DR, including DME Fundus photography Healthcare provider FDA approval
    Medios AI (Remidio Innovative Solutions Pvt Ltd., Karnataka, India) [63] Offline AI algorithm 2023 Screening for referable DR and referable glaucoma Fundus photography from smartphone-based camera Healthcare provider CE mark
    RetCAD (Thirona Retina BV, Nijmegen, Netherlands) [64] Cloud-based algorithm 2022 Screen for referable DR, AMD Fundus photography Healthcare provider CE mark
    EyeArt (Eyenuk, Inc., Woodland Hills, CA, USA) [65] Cloud-based algorithm 2015 and 2020 Screen for referable DR Fundus photography Healthcare provider FDA clearance (class II), CE mark
    VUNO Med-Fundus AI (VUNO Inc., Seoul, Korea) [66] Cloud-based algorithm 2020 DR, AMD, glaucoma Fundus photography Healthcare provider CE mark (class IIa), MFDS approval (Korea, class III), HAS approval (Singapore)
    THEIA (Toku Eyes, Auckland, New Zealand) [67] Cloud-based algorithm 2020 Smoking status, AMD, DR, cataract Fundus photography, OCT-A Healthcare provider In progress
    iPredict (iHealthScreen Inc., Richmond Hill, NY, USA) [68] Cloud-based algorithm 2021 and 2022 DR, AMD, glaucoma Fundus photography, OCT Healthcare provider CE mark, TGA approval (Australia)
    Notal Home OCT (Notal Vision, Inc., Manassas, VA, USA) [69] In-home device and cloud-based algorithm 2018 Neovascular AMD OCT Patient-driven healthcare FDA Breakthrough Device Designation
    OphtAI (Evolucare/ADCIS, Villers-Bretonneux, France) [70] Cloud-based algorithm 2019 DR, DME, AMD, glaucoma Fundus photography Healthcare provider CE mark, HC approval (Canada), FDA in progress
    Retmarker (Retmarker, SA, Taveiro, Portugal) [71] Cloud-based algorithm 2010 DR, AMD Fundus photography Healthcare provider CE mark (class IIa), TGA approval (Australia)
    RetinaLyze (RetinaLyze System A/S, Hellerup, Denmark) [72] Cloud-based algorithm 2021 DR, dry AMD, glaucoma Fundus photography, OCT Healthcare provider CE mark (class I, self-certified)
    RetinAI Discovery (RetinAI Medical AG, Bern, Switzerland) [73] Cloud-based algorithm 2022 AMD, DR, DME, RVO Fundus photography, OCT Healthcare provider FDA clearance (class II), CE mark
    Note: All medical devices approved by the FDA or accredited by the CE mark from January 2015 to November 2022 were collected. These devices were searched for in the European Database on Medical Devices (EUDAMED) database [74], the FDA website (on the webpage of Artificial Intelligence and Machine Learning-Enabled Medical Devices) [75], and the FDA 510(k) Premarket Notification [76].ARDA, Automatic Retinal Disease Assessment; DME, diabetic macular edema; HAS, Health Sciences Authority (Singapore); HC, Health Canada; IRIS, Intelligent Retinal Imaging Systems; MFDS, Ministry of Food and Drug Safety (Korea); OCT, optical coherence tomography; OCT-A, optical coherence tomography angiography; RVO, retinal vein occlusion; SELENA, Singapore Eye LEsioN Analyzer; TGA, Therapeutic Goods Administration.

    Table 2 - Description and notification of current AI systems
    AI system Description and notification
    IRIS (Intelligent Retinal Imaging Systems, Pensacola, FL, USA) IRIS is an FDA class II cleared medical system that has a moderate risk to consumers and must demonstrate that it is “substantially equivalent” to similar products. The IRIS program is a cloud-based platform to screen for vision-threatening DR, with sensitivity and specificity of 66.4% and 72.8%, respectively [77]
    ARDA (Google LLC, Mountain View, CA, USA) ARDA is a DL algorithm developed by Google Health from >128 000 retinal photographs of patients from the United States and India and validated in >10 000 photographs from the UK to detect referrable and sight-threatening DR [7]. The validation study of ARDA was the first showing robust performance of DL to detect referrable DR with >95% of both sensitivity and specificity [7]. Later, ARDA was prospectively validated in India [78] and a nationwide screening program in Thailand [79]
    SELENA+ (EyRIS Pte Ltd, Singapore) The Singapore Eye Research Institute and Singapore National Eye Center has developed a DL-based algorithm, SELENA+, to screen for referable DR, vision-threatening DR, DR-related vascular risk factors, suspected glaucoma, and late-stage AMD. It is a multicenter collaborative research effort with half a million retinal images from people of different ethnicities such as Caucasians from Australia and the United States, and Singapore Chinese, Malayans, Indians, Chinese, individuals from Hong Kong, Mexicans, Hispanics, and African Americans. Real-world application and clinical translation of SELENA+ has been integrated into the Singapore Integrated Diabetic Retinopathy Programme in recent years. SELENA+ has significant diagnostic performance in DR, with sensitivity of 91%, specificity of 90%, and area under the curve of 0.93 [80]
    IDx-DR (Digital Diagnostics Inc., Coralville, IA, USA) IDx-DR was the first FDA-approved ophthalmic device to autonomously detect DR, including DME. It can analyze retinal images, detect vision-threatening DR, and provide referral recommendations [6,81]. The external validation to detect referable DR showed sensitivity and specificity of 91% and 84%, respectively [79]
    Medios AI (Remidio Innovative Solutions Pvt Ltd., Karnataka, India) Medios AI is an integrated offline system with a Remidio smartphone-based, nonmydriatic retinal camera to detect referable DR. The fundus images can be captured by minimally trained healthcare providers. The sensitivity and specificity of diagnosing referable DR were 100% and 88.4%, respectively [82]
    RetCAD (Thirona Retina BV, Nijmegen, Netherlands) This commercially available DL algorithm can determine referable DR and AMD based on a dataset of CFPs to reduce the workload of screening programs by up to 96%, with sensitivity of 90.53% and specificity of 97.13%. Patients’ CFPs can be captured by camera and then transferred to the Thirona server for analysis. The examination report will provide referable suggestions and visualization of heatmaps [83]
    EyeArt (Eyenuk, Inc., Woodland Hills, CA, USA) This cloud-based autonomous AI system can detect more-than-mild DR and vision-threatening DR by submitting fundus photography to the platform. It is designed to work with various types of retinal cameras. It assesses the quality of uploaded images and explains the reasons behind grading. This algorithm can provide the grading of DR and report the results for each eye based on the UK National Health Service diabetic eye screening program scale. The sensitivity and specificity showed 96% and 98%, respectively [84]
    VUNO Med-Fundus AI (VUNO Inc., Seoul, Korea) The AI-based VUNO Med-Fundus AI analyzes CFP to detect multiple retinal lesions (areas under receiver operating characteristic curves for all findings were at 96.2%) [85]. The area under the receiver operating characteristic curves for DR-related findings was 95%. It was approved as a class III medical device by the Ministry of Food and Drug Safety in Korea
    THEIA (Toku Eyes, Auckland, New Zealand) The New Zealand company Toku Eyes developed THEIA, an AI platform for cloud-based multimodal image analysis of referable DR and AMD. The THEIA system was developed from two of the largest screening data sets in Auckland, New Zealand: the Auckland District Health Board and the Counties Manukau District Health Board. It can analyze color fundus images, OCT, and OCT-A to provide results about referable DR (sensitivity of 93% and specificity of 63%) and intermediate dry AMD (accuracy of 96%) [86,87]. This AI system is considered to be useful in reducing the workload in the New Zealand National Diabetic Retinopathy Screening Program [88]
    iPredict (iHealthScreen Inc., Richmond Hill, NY, USA) The iPredict AI Eye Screening System offers fully automated diagnosis of referable DR (sensitivity of 97.0% and specificity of 96.3%) and AMD (sensitivity of 86.6% and specificity of 92.1%) by analyzing CFPs [89]
    Notal Home OCT (Notal Vision, Inc., Manassas, VA, USA) Notal Home OCT, the first FDA-cleared in-home OCT device, which includes an AI algorithm and monitoring center, is designed to detect AMD. The imaging quality showed great correlation with in-office OCT for detecting the presence of fluid in 95% agreement with human graders [90]. A patient's ability to use an in-home setting for self-imaging without training demonstrated good capacity with a 95% success rate [91]
    OphtAI (Evolucare/ADCIS, Villers-Bretonneux, France) OphtAI DR is a semiautomatic AI algorithm that assesses the pathologic lesions and grading of DR and detects AMD and glaucoma. In a multicenter, head-to-head, real-world validation study to compare different algorithms in detecting DR, the OphtAI DR algorithm provided better results (sensitivity of 80.47% and specificity of 81.28%) than an ophthalmologist [92]. It is also deemed clinically safe and economically efficient in reducing the costs by more than U.S. $15 per patient [92]
    Retmarker (Retmarker, SA, Taveiro, Portugal) This AI technology can provide screening for DR and AMD by annotating pathologic lesions, such as microaneurysms, drusen, hypopigmentation, hyperpigmentation, and geographic atrophy [93]. The sensitivity in classifying DR is 73.0% for any DR, 85.0% for referable DR, and 97.9% for proliferative DR [93]. The screening performance of Retmarker appeared to vary with patients’ age, ethnicity, and camera type. In economic analysis, the Retmarker was more cost effective than manual grading [93]
    RetinaLyze (RetinaLyze System A/S, Hellerup, Denmark) The RetinaLyze system is a screening software that can detect DR on nonmydriatic CFPs (sensitivity of 89.9% and specificity of 85.7%) [94]. It can detect DR lesions, including microaneurysms and minor hemorrhages (specificity of 71.4%) [95]. It can also evaluate biological aging [96] and hemoglobin on optic disc photographs [97]
    RetinAI Discovery (RetinAI Medical AG, Bern, Switzerland) The Discovery platform can analyze medical data and ophthalmic images such as OCT scans and CFP from a variety of devices. It can help automatically detect the location of the fovea (mean total location error of 0.101 mm), the quantification of pathologic fluid, and the segmentation of atrophic retina on OCT in patients with geographic atrophy [98–100]. It can detect and quantify fluid from DR, DME, AMD, and RVO. The performance of the AI system showed that the accuracy, specificity, and sensitivity for intraretinal fluid was 0.87, 0.88, 0.84 and 0.93, 0.95, 0.93 and for subretinal fluid was 0.93, 0.93, 0.93 and 0.95, 0.95, 0.95 in the AMD and DME cohorts, respectively [101▪,102]
    Note: All medical devices approved by the FDA or accredited by the CE mark from January 2015 to January 2023 were collected. These devices were searched for in the European Database on Medical Devices (EUDAMED) database [74], the FDA website (on the webpage of Artificial Intelligence and Machine Learning-Enabled Medical Devices) [75], and the FDA 510(k) Premarket Notification [76]. The approved devices were summarized and their performance in related trials was searched for in PubMed, with the data source cited as the reference.ARDA, Automatic Retinal Disease Assessment; CFP, color fundus photograph; DME, diabetic macular edema; IRIS, Intelligent Retinal Imaging Systems; OCT, optical coherence tomography; OCT-A, optical coherence tomography angiography; RVO, retinal vein occlusion; SELENA, Singapore Eye LEsioN Analyzer.

    The Vision Academy recognizes the advantages of AI technology and recommends the use of them to be of additive and synergistic value to current standards of care. In terms of applying such technologies in diagnosing and screening retinal diseases, we summarize the following directions and emphasize several viewpoints important for the future.

    Recommendation 1: integration of meta-data and data sets

    The integration of meta-data, including multimodal images and structured clinical information from multiple data sets with different ethnic groups, and establishment of a data processing and sharing system will empower data-driven AI technologies in ophthalmic practice. Ongoing research will be needed to build up data storage and sharing systems in a cybersecurity framework for broader use.

    Recommendation 2: data privacy versus transparency – a balance or conflict?

    While retinal images possess biometric information that could be reidentified by AI technologies, care should be taken when collecting and processing these images. Some novel learning tasks (e.g. generative adversarial networks) can obscure bioidentical information or even provide unsupervised models for small-scale data sets. The question of how to universalize data formats will be one of the key factors for extending the scalability and generalizability of AI-enabled technologies.

    The complexity and inexplicability of AI are encompassed in the term “black box phenomenon.” Black box algorithms have potential to cause misuse of AI in healthcare ecosystems [74,101▪]. Transparency of the algorithms is therefore another critical point to overcome users’ hesitation. Maintaining adequate balance between data privacy and transparency should be a concern in the application of AI-enabled technologies.

    Recommendation 3: implementation of artificial intelligence in clinical practice – replacement or rectification?

    The role of AI-enabled technologies in the real world is not to replace ophthalmologists but to assist them and to hybridize both AI models and human experience for making more efficient and accurate decisions. Such time-saving abilities could streamline medical procedures and give clinicians more time to communicate with their patients. Improper implementation of AI could be harmful to doctor–patient relationships and could affect patients’ trust if AI algorithms were used only for improving workflow but not patient care.

    Recommendation 4: ethical concerns and regulatory issues

    A key hurdle in deploying AI-enabled technologies in clinical practice is the fear of making an incorrect decision and harming patients. Legal liability should be well defined as the implementation of AI becomes more popular. Such liability should only be at the precise claim of screening targeted diseases. Unlike retinal specialists, the developers of an AI model should only be liable for the designed algorithm for screening specific diseases. Healthcare providers should still take full responsibility for being aware of the capacity of AI models.

    The legal boundaries between developers and healthcare providers are still unresolved, and legislative and governance systems need to be more established to refine liability rules and the regulatory environment. Policy and specific authorities should be set up not only for verification of AI models but for data security and legal liability. Cross-sector and cross-disciplinary collaborations will be important to ensure the integrity of AI healthcare ecosystems.

    Recommendation 5: long-term basis

    Continuing education, promotion of practical application, and user-friendly, understandable interfaces for healthcare providers are equally important to streamline the workflow and broaden the applicability of AI systems. Communication and collaboration between cross-functional teams, including ophthalmologists, optometrists, computer scientists, statisticians, data scientists, patient organizations, and engineers, can have a positive impact on vision health and preservation through AI-enabled technologies.


    The establishment of AI-enabled technologies may have potential to improve the efficiency of existing healthcare pathways, provide better patient-centered services, minimize the impact of labor shortage, and bridge the gap between urban and rural areas. However, no advancement in clinical practice is flawless, so it is necessary for healthcare providers and legislators to be aware of the limitations of AI-enabled devices.


    Editorial assistance was provided by Elle Lindsay, PhD, Macha Aldighieri, PhD, and Rachel Fairbanks, BA (Hons), of Complete HealthVizion, Ltd, an IPG Health Company, funded by Bayer Consumer Care AG, Pharmaceuticals Division, Basel, Switzerland.

    Financial support and sponsorship

    The Vision Academy is a group of over 100 international ophthalmology experts who provide guidance for best clinical practice through their collective expertise in areas of controversy or with insufficient conclusive evidence. The Vision Academy is funded and facilitated by Bayer. The opinions and guidance of the Vision Academy outputs are those of its members and do not necessarily reflect the opinions of Bayer.

    Financial arrangements of the authors with companies whose products may be related to the present report are listed in the “Conflicts of interest” section, as declared by the authors.

    Conflicts of interest

    Yu-Bai Chou is a consultant for Alcon and Bayer. Paolo Lanzetta is a consultant for Aerie, AbbVie, Apellis, Bausch & Lomb, Bayer, Biogen, Boehringer Ingelheim, Genentech, Novartis, Ocular Therapeutix, Outlook Therapeutics, and Roche. Tariq Aslam is a consultant for, and has received grants, speaker fees, and honoraria from, Allergan, Bayer Pharmaceuticals, Canon, NIHR, Roche, and Topcon. He is also a board member for the Vision Academy, Macular Society, and Fight for Sight charity. Jane Barratt has received honoraria from Bayer. Carla Danese is a consultant for Bayer. Bora Eldem is a consultant for Allergan, Bayer, Novartis, and Roche. Nicole Eter is an advisor for AbbVie, Alcon, Apellis, Bayer, Biogen, Janssen, Novartis, and Roche and has received speaker fees from AbbVie, Apellis, Bayer, Novartis, and Roche and research grants from Bayer and Novartis. Richard Gale is a consultant for AbbVie, Allergan, Apellis, Bayer, Biogen, Boehringer Ingelheim, Notal, Novartis, Roche, and Santen and has received research grants from Bayer, Novartis, and Roche. Jean-François Korobelnik is a consultant for Allergan/AbbVie, Apellis, Bayer, Carl Zeiss Meditec, Janssen, Nano Retina, Roche, and Théa and is a member of the Data and Safety Monitoring Boards for Alexion and Novo Nordisk. Igor Kozak is a consultant for Alcon, Bayer, and Novartis. Anat Loewenstein is a consultant for Allergan, Annexon, Bayer Healthcare, Beyeonics, Biogen, ForSight Labs, IQVIA, Iveric Bio, Johnson & Johnson, MJH Events, Nano Retina, Notal Vision, Novartis, Ocuphire Pharma, OcuTerra, OphtiMedRx, Roche, Ripple Therapeutics, Syneos, WebMD, and Xbrane. Paisan Ruamviboonsuk is a consultant for, and has received research funds from, Bayer and Roche. Taiji Sakamoto is a consultant for Bayer Yakuhin, Boehringer Ingelheim, Chugai, Nidek, Nikon, Novartis, Santen, and Senju. Daniel S.W. Ting has received research grants from the National Medical Research Council Singapore, Duke-NUS Medical School Singapore, and Agency for Science, Technology and Research Singapore. Peter van Wijngaarden is the co-founder of Enlighten Imaging, an early-stage medical technology start-up company devoted to hyperspectral retinal imaging and image analysis, including the development of AI systems, and has received research grant support from Bayer and Roche and honoraria from Bayer, Mylan, Novartis, and Roche. Sebastian M. Waldstein is a consultant for Apellis, Bayer, Boehringer Ingelheim, Novartis, Roche, and Santen. David Wong is a consultant for AbbVie, Alcon, Apellis, Bayer, Bausch Health, Biogen, Boehringer Ingelheim, Novartis, Ripple Therapeutics, Roche, Topcon, and Zeiss, has received financial support (to institution) from Bayer, Novartis, and Roche, and is an equity owner at ArcticDx. Lihteh Wu is a consultant for Bayer, Lumibird Medical, Novartis, and Roche. Miguel A. Zapata is a consultant for Novartis and Roche, has received grants and speaker fees from DORC, Novartis, and Roche, honoraria from Alcon, Bayer, DORC, Novartis, and Roche, has served on advisory boards for Novartis and Roche, has received equipment from Allergan, and has stock or stock options in UpRetina. Javier Zarranz-Ventura has received grants from AbbVie, Allergan, Bayer, Novartis, and Roche, has served on scientific advisory boards for AbbVie, Allergan, Bayer, Novartis, and Roche, and has been a speaker for AbbVie, Alcon, Alimera Sciences, Allergan, Bausch & Lomb, Bayer, Brill Pharma, DORC, Esteve, Novartis, Roche, Topcon Healthcare, and Zeiss. Aditya U. Kale, Xiaorong Li, and Xiaoxin Li have no conflicts of interest to report.


    Papers of particular interest, published within the annual period of review, have been highlighted as:

    ▪ of special interest

    ▪▪ of outstanding interest


    1. McCarthy J, Minsky ML, Rochester N, Shannon CE. A proposal for the Dartmouth summer research project on artificial intelligence, August 31, 1955. AI Mag 2006; 27:1214.
    2. Samuel AL. Some studies in machine learning using the game of checkers. IBM J 1959; 3:535554.
    3. Bengio Y, Courville A, Vincent P. Representation learning: a review and new perspectives. IEEE Trans Pattern Anal Mach Intell 2013; 35:17981828.
    4. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature 2015; 521:436444.
    5. Krizhevsky A, Sutskever I, Hinton GE. ImageNet classification with deep convolutional neural networks. Commun ACM 2017; 60:8490.
    6. Abràmoff MD, Lou Y, Erginay A, et al. Improved automated detection of diabetic retinopathy on a publicly available dataset through integration of deep learning. Invest Ophthalmol Vis Sci 2016; 57:52005206.
    7. Gulshan V, Peng L, Coram M, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 2016; 316:24022410.
    8. Ting DSW, Cheung CY-L, Lim G, 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; 318:22112223.
    9. Gargeya R, Leng T. Automated identification of diabetic retinopathy using deep learning. Ophthalmology 2017; 124:962969.
    10. Burlina PM, Joshi N, Pekala M, et al. Automated grading of age-related macular degeneration from color fundus images using deep convolutional neural networks. JAMA Ophthalmol 2017; 135:11701176.
    11. Grassmann F, Mengelkamp J, Brandl C, 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; 125:14101420.
    12. Brown JM, Campbell JP, Beers A, et al. Automated diagnosis of plus disease in retinopathy of prematurity using deep convolutional neural networks. JAMA Ophthalmol 2018; 136:803810.
    13. Sun G, Wang X, Xu L, et al. Deep learning for the detection of multiple fundus diseases using ultra-widefield images. Ophthalmol Ther 2023; 12:895907.
    14. Li B, Chen H, Zhang B, et al. Development and evaluation of a deep learning model for the detection of multiple fundus diseases based on colour fundus photography. Br J Ophthalmol 2022; 106:10791086.
    15. Abitbol E, Miere A, Excoffier JB, et al. Deep learning-based classification of retinal vascular diseases using ultra-widefield colour fundus photographs. BMJ Open Ophthalmol 2022; 7:e000924.
    16. Cai S, Parker F, Urias MG, et al. Deep learning detection of sea fan neovascularization from ultra-widefield color fundus photographs of patients with sickle cell hemoglobinopathy. JAMA Ophthalmol 2021; 139:206213.
    17. Wang Z, Hu ZJ. Artificial intelligence for assessment of Stargardt macular atrophy. Neural Regen Res 2022; 17:26322636.
    18. Miere A, Le Meur T, Bitton K, et al. Deep learning-based classification of inherited retinal diseases using fundus autofluorescence. J Clin Med 2020; 9:3303.
    19. Fujinami-Yokokawa Y, Ninomiya H, Liu X, et al. Prediction of causative genes in inherited retinal disorder from fundus photography and autofluorescence imaging using deep learning techniques. Br J Ophthalmol 2021; 105:12721279.
    20. Lee CS, Baughman DM, Lee AY. Deep learning is effective for the classification of OCT images of normal versus age-related macular degeneration. Ophthalmol Retina 2017; 1:322327.
    21. Schmidt-Erfurth U, Vogl WD, Jampol LM, Bogunović H. Application of automated quantification of fluid volumes to anti-VEGF therapy of neovascular age-related macular degeneration. Ophthalmology 2020; 127:12111219.
    22. Schlegl T, Waldstein SM, Bogunovic H, et al. Fully automated detection and quantification of macular fluid in OCT using deep learning. Ophthalmology 2018; 125:549558.
    23. De Fauw J, Ledsam JR, Romera-Paredes B, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med 2018; 24:13421350.
    24. Oh E, Yoo TK, Park EC. Diabetic retinopathy risk prediction for fundus examination using sparse learning: a cross-sectional study. BMC Med Inform Decis Mak 2013; 13:106.
    25. Mordi IR, Trucco E, Syed MG, et al. Prediction of major adverse cardiovascular events from retinal, clinical, and genomic data in individuals with type 2 diabetes: a population cohort study. Diabetes Care 2022; 45:710716.
    26. Veturi YA, Woof W, Lazebnik T, et al. SynthEye: investigating the impact of synthetic data on artificial intelligence-assisted gene diagnosis of inherited retinal disease. Ophthalmol Sci 2023; 3:100258.
    27. Lakhani P, Sundaram B. Deep learning at chest radiography: automated classification of pulmonary tuberculosis by using convolutional neural networks. Radiology 2017; 284:574582.
    28. Executive Office of the President, National Science and Technology Council Committee on Technology. Preparing for the future of artificial intelligence. 2016. Available at: [Accessed January 2023].
    29. Lee AY, Campbell JP, Hwang TS, et al. Recommendations for standardization of images in ophthalmology. Ophthalmology 2021; 128:969970.
    30. Chiang MF, Boland MV, Brewer A, et al. Special requirements for electronic health record systems in ophthalmology. Ophthalmology 2011; 118:16811687.
    31. Khan SM, Liu X, Nath S, et al. A global review of publicly available datasets for ophthalmological imaging: barriers to access, usability, and generalisability. Lancet Digit Health 2021; 3:e51e66.
    32. Narayanan A, Shmatikov V. Robust de-anonymization of large sparse datasets. 2008; Oakland, CA, USA: IEEE Symposium on Security and Privacy, 111–125.
    33. Akram MU, Abdul Salam A, Khawaja SG, et al. RIDB: a dataset of fundus images for retina based person identification. Data Brief 2020; 33:106433.
    34. Imtiaz S, Arsalan M, Vlassov V, Sadre R. Synthetic and private smart health care data generation using GANs. IoTSPT-ML Workshop, International Conference on Computer Communications and Networks (ICCCN), Athens, Greece, 2021:1–7.
    35. Kaur D, Sobiesk M, Patil S, et al. Application of Bayesian networks to generate synthetic health data. J Am Med Inform Assoc 2021; 28:801811.
    36. Yang Q, Liu Y, Chen T, Tong Y. Federated machine learning: concept and applications. ACM Trans Intell Syst Technol 2019; 10:Article 12.
    37. Warnat-Herresthal S, Schultze H, Shastry KL, et al. Swarm learning for decentralized and confidential clinical machine learning. Nature 2021; 594:265270.
    38. Lo J, Yu TT, Ma D, et al. Federated learning for microvasculature segmentation and diabetic retinopathy classification of OCT data. Ophthalmol Sci 2021; 1:100069.
    39▪. Evans NG, Wenner DM, Cohen IG, et al. Emerging ethical considerations for the use of artificial intelligence. Ophthalmol Sci 2022; 2:100141.
    40. American Medical Association. Augmented intelligence in health care. 2019. Available at: [Accessed January 2023].
    41. Vladeck DC. Machines without principals: liability rules and artificial intelligence. Wash L Rev 2014; 89:117150.
    42. He J, Baxter SL, Xu J, et al. The practical implementation of artificial intelligence technologies in medicine. Nat Med 2019; 25:3036.
    43. Abdullah YI, Schuman JS, Shabsigh R, et al. Ethics of artificial intelligence in medicine and ophthalmology. Asia Pac J Ophthalmol (Phila) 2021; 10:289298.
    44▪▪. Liu X, Cruz Rivera S, Moher D, et al. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Lancet Digit Health 2020; 2:e537e548.
    45▪▪. Vasey B, Nagendran M, Campbell B, et al. Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nat Med 2022; 28:924933.
    46. Collins GS, Dhiman P, Andaur Navarro CL, et al. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open 2021; 11:e048008.
    47. Ibrahim H, Liu X, Zariffa N, et al. Health data poverty: an assailable barrier to equitable digital health care. Lancet Digit Health 2021; 3:e260e265.
    48. European Parliament, Council of the European Union. Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation). 2016. Available at: [Accessed March 2023].
    49. European Parliament, Council of the European Union. Directive (EU) 2016/1148 of the European Parliament and of the Council of 6 July 2016 concerning measures for a high common level of security of network and information systems across the Union. 2016. Available at: [Accessed March 2023].
    50. European Parliament, Council of the European Union. Regulation (EU) 2017/745 of the European Parliament and of the Council of 5 April 2017 on medical devices, amending Directive 2001/83/EC, Regulation (EC) No 178/2002 and Regulation (EC) No 1223/2009 and repealing Council Directives 90/385/EEC and 93/42/EEC. 2017. Available at: [Accessed March 2023].
    51. Goodman B, Flaxman S. European Union regulations on algorithmic decision-making and a “right to explanation”. AI Mag 2017; 38:5057.
    52. International Medical Device Regulators Forum. Software as a Medical Device (SaMD): key definitions. 2013. Available at: [Accessed January 2023].
    53. Ting DSW, Lee AY, Wong TY. An ophthalmologist's guide to deciphering studies in artificial intelligence. Ophthalmology 2019; 126:14751479.
    54. European Commission. Proposal for a regulation of the European Parliament and of the council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain union legislative acts, COM(2021) 206. 2021. Available at: [Accessed January 2023].
    55. Ruamviboonsuk P, Chantra S, Seresirikachorn K, et al. Economic evaluations of artificial intelligence in ophthalmology. Asia Pac J Ophthalmol 2021; 10:307316.
    56. Chen EM, Chen D, Chilakamarri P, et al. Economic challenges of artificial intelligence adoption for diabetic retinopathy. Ophthalmology 2021; 128:475477.
    57. Nguyen HV, Tan GSW, Tapp RJ, et al. Cost-effectiveness of a national telemedicine diabetic retinopathy screening program in Singapore. Ophthalmology 2016; 123:25712580.
    58. Greaves F, Joshi I, Campbell M, et al. What is an appropriate level of evidence for a digital health intervention? Lancet 2019; 392:26652667.
    59. Intelligent Retinal Imaging Systems. iris™. 2023. Available at: [Accessed March 2023].
    60. Google Health. Automated Retinal Disease Assessment. 2023. Available at: [Accessed March 2023].
    61. EyRIS Pte Ltd. Technology behind SELENA+. 2023. Available at: [Accessed March 2023].
    62. Digital Diagnostics. IDx-DR. 2023. Available at: [Accessed March 2023].
    63. Remidio Innovative Solutions. Medios DR. 2023. Available at: [Accessed March 2023].
    64. Thirona Retina. RetCAD™. Available at: [Accessed March 2023].
    65. Eyenuk. EyeArt® AI Eye Screening System. Available at: [Accessed March 2023].
    66. VUNO. VUNO Med-Fundus AI. 2023. Available at: [Accessed March 2023].
    67. Toku Eyes. Introducing THEIA™. 2023. Available at: [Accessed March 2023].
    68. iHealthScreen. iPredict. 2020. Available at: [Accessed March 2023].
    69. Notal Vision. Notal Home OCT. 2023. Available at: [Accessed March 2023].
    70. OphtAI. Artificial intelligence dedicated to ophthalmology | Home. 2023. Available at: [Accessed March 2023].
    71. Retmarker, a METEDA company. Retmarker. 2023. Available at: [Accessed March 2023].
    72. RetinaLyze System A/S. RetinaLyze®. 2020. Available at: [Accessed March 2023].
    73. RetinAI. RetinAI Discovery®. 2022. Available at: [Accessed March 2023].
    74. European Commission. EUDAMED database. 2023. Available at: [Accessed January 2023].
    75. US Food and Drug Administration. Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices. 2022. Available at: [Accessed January 2023].
    76. US Food and Drug Administration. 510(k) Premarket Notification. 2023. Available at: [Accessed January 2023].
    77. Walton OB IV, Garoon RB, Weng CY, et al. Evaluation of automated teleretinal screening program for diabetic retinopathy. JAMA Ophthalmol 2016; 134:204209.
    78. Gulshan V, Rajan RP, Widner K, et al. Performance of a deep-learning algorithm vs manual grading for detecting diabetic retinopathy in India. JAMA Ophthalmol 2019; 137:987993.
    79. van der Heijden AA, Abramoff MD, Verbraak F, et al. Validation of automated screening for referable diabetic retinopathy with the IDx-DR device in the Hoorn Diabetes Care System. Acta Ophthalmol 2018; 96:6368.
    80. Wong DCS, Kiew G, Jeon S, Ting D. Singapore Eye Lesions Analyzer (SELENA): The Deep Learning System for Retinal Diseases. In: Grzybowski, A., editor. Artificial Intelligence in Ophthalmology. New York, NY, USA: Springer; 2021. pp. 177–185.
    81. Abràmoff MD, Lavin PT, Birch M, et al. Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices. NPJ Digit Med 2018; 1:39.
    82. Natarajan S, Jain A, Krishnan R, et al. Diagnostic accuracy of community-based diabetic retinopathy screening with an offline artificial intelligence system on a smartphone. JAMA Ophthalmol 2019; 137:11821188.
    83. González-Gonzalo C, Sánchez-Gutiérrez V, Hernández-Martínez P, et al. Evaluation of a deep learning system for the joint automated detection of diabetic retinopathy and age-related macular degeneration. Acta Ophthalmol 2020; 98:368377.
    84. Ipp E, Liljenquist D, Bode B, et al. Pivotal evaluation of an artificial intelligence system for autonomous detection of referrable and vision-threatening diabetic retinopathy. JAMA Netw Open 2021; 4:e2134254.
    85. Son J, Shin JY, Kim HD, et al. Development and validation of deep learning models for screening multiple abnormal findings in retinal fundus images. Ophthalmology 2020; 127:8594.
    86. 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.
    87. Xie L, Yang S, Squirrell D, Vaghefi E. Towards implementation of AI in New Zealand national screening program: cloud-based, robust, and bespoke. PLoS One 2020; 15:e0225015.
    88. Vaghefi E, Yang S, Xie L, et al. THEIA™ development, and testing of artificial intelligence-based primary triage of diabetic retinopathy screening images in New Zealand. Diabet Med 2021; 38:e14386.
    89. Bhuiyan A, Govindaiah A, Alauddin S, et al. Combined automated screening for age-related macular degeneration and diabetic retinopathy in primary care settings. Ann Eye Sci 2021; 6:12.
    90. Kim JE, Tomkins-Netzer O, Elman MJ, et al. Evaluation of a self-imaging SD-OCT system designed for remote home monitoring. BMC Ophthalmol 2022; 22:261.
    91. Liu Y, Holekamp NM, Heier JS. Prospective, longitudinal study: daily self-imaging with home OCT for neovascular age-related macular degeneration. Ophthalmol Retina 2022; 6:575585.
    92. Lee AY, Yanagihara RT, Lee CS, et al. Multicenter, head-to-head, real-world validation study of seven automated artificial intelligence diabetic retinopathy screening systems. Diabetes Care 2021; 44:11681175.
    93. Tufail A, Kapetanakis VV, Salas-Vega S, et al. An observational study to assess if automated diabetic retinopathy image assessment software can replace one or more steps of manual imaging grading and to determine their cost-effectiveness. Health Technol Assess 2016; 20:172.
    94. Hansen AB, Hartvig NV, Jensen MS, et al. Diabetic retinopathy screening using digital non-mydriatic fundus photography and automated image analysis. Acta Ophthalmol Scand 2004; 82:666672.
    95. Larsen M, Godt J, Larsen N, et al. Automated detection of fundus photographic red lesions in diabetic retinopathy. Invest Ophthalmol Vis Sci 2003; 44:761766.
    96. Rodríguez-Esteve P, Gonzalez-Hernandez M, Pareja-Rios A, Gonzalez de la Rosa M. Non invasive evaluation of the aging of the cristaline lens in normals and diabetics. Invest Ophthalmol Vis Sci; ARVO Annual Meeting, May 1–7, 2021. Abstract 72.
    97. Rocha JAG, Dias DT, Lemos MBC, et al. Optic nerve head hemoglobin levels in glaucoma: a structural and functional correlation study. J Ophthalmol 2021; 2021:9916102.
    98. Montesel A, Gigon A, Mosinska A, et al. Automated foveal location detection on spectral-domain optical coherence tomography in geographic atrophy patients. Graefes Arch Clin Exp Ophthalmol 2022; 260:22612270.
    99. Derradji Y, Mosinska A, Apostolopoulos S, et al. Fully-automated atrophy segmentation in dry age-related macular degeneration in optical coherence tomography. Sci Rep 2021; 11:21893.
    100. Mantel I, Mosinska A, Bergin C, et al. Automated quantification of pathological fluids in neovascular age-related macular degeneration, and its repeatability using deep learning. Transl Vis Sci Technol 2021; 10:17.
    101▪. Martin-Pinardel R, Izquierdo-Serra J, De Zanet S, et al. Artificial intelligence-based fluid quantification and associated visual outcomes in a real-world, multicentre neovascular age-related macular degeneration national database. Br J Ophthalmol 2023; doi: 10.1136/bjo-2022-322297.
    102. Habra O, Gallardo M, Meyer zu Westram T, et al. Evaluation of an artificial intelligence-based detector of sub- and intraretinal fluid on a large set of optical coherence tomography volumes in age-related macular degeneration and diabetic macular edema. Ophthalmologica 2022; 245:516527.

    Lead author, on behalf of the Vision Academy Retinal AI Workstream.


    artificial intelligence; diagnosis; retina; retinal imaging

    Copyright © 2023 The Author(s). Published by Wolters Kluwer Health, Inc.