Artificial intelligence as a medical device for ophthalmic image analysis: a scoping review of regulated devices

Artificial intelligence as a medical device for ophthalmic image analysis: a scoping review of regulated devices | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Artificial intelligence as a medical device for ophthalmic image analysis: a scoping review of regulated devices Ariel Yuhan Ong, Priyal Taribagil, Mertcan Sevgi, Aditya U Kale, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6026482/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 29 May, 2025 Read the published version in npj Digital Medicine → Version 1 posted 8 You are reading this latest preprint version Abstract This scoping review aims to identify regulator-approved ophthalmic image analysis AIaMDs in three jurisdictions, examine their characteristics and regulatory approvals, and evaluate the available evidence underpinning them, as a step towards identifying best practice and areas for improvement. 36 AIaMDs from 28 manufacturers were identified − 97% (35/36) approved in the EU, 22% (8/36) in Australia, and 8% (3/36) in the USA. Most targeted diabetic retinopathy detection. 19% (7/36) did not have published evidence describing performance. For the remainder, 131 clinical evaluation studies (range 1–22/AIaMD) describing 192 datasets/cohorts were identified. Demographics were poorly reported (age recorded in 52%, sex 51%, ethnicity 21%). On a study-level, few included head-to-head comparisons against other AIaMDs (8%,10/131) or humans (22%, 29/131), and 37% (49/131) were conducted independently of the manufacturer. Only 11 studies (8%) were interventional. There is scope for expanding AIaMD applications to other ophthalmic imaging modalities, conditions, and use cases. Facilitating greater transparency from manufacturers, better dataset reporting, validation across diverse populations, and high-quality interventional studies with implementation-focused outcomes are key steps towards building user confidence and supporting clinical integration. Health sciences/Health care Health sciences/Health care/Diagnosis Figures Figure 1 Figure 2 Figure 3 INTRODUCTION Artificial intelligence (AI) represents a rapidly evolving frontier in healthcare which offers transformative potential across multiple specialties. In ophthalmology, AI can help enhance diagnostic accuracy, provide insights into systemic diseases, streamline clinical and research workflows, optimise treatment, with the ultimate goal of improving patient outcomes. 1 It can potentially help in addressing challenges such as variability in subjective human interpretation, the increasing volume of complex imaging data, and the global shortage of ophthalmic specialists. This shortage has resulted in a capacity-demand imbalance that risks irreversible sight loss from treatment delays, 2 – 4 affecting quality of life for patients and carers, 5 – 7 and posing a significant economic burden to individuals, healthcare services, and society. 8 , 9 However, while AI promises significant advancements, it may also introduce new complexities in the evaluation of safety, effectiveness, and equity/bias, which are essential to ensure they achieve their intended purpose for their target population. 10 At present, the evidence requirements to support the regulatory approval of artificial intelligence as a medical device (AIaMD) are less transparent compared to more established interventions such as drugs, which follow well-established and rigorous pathways for evaluation prior to reaching the market. Evidence standards for AIaMDs are informed by regulations which are designed to be broadly applicable across all medical devices and clinical contexts. This therefore requires some level of abstraction, which allows for varied interpretations when applied to specific AIaMDs and use cases. This raises questions about what constitutes a ‘sufficient’ level of evidence for regulatory approval or real-world deployment, particularly for AIaMDs intended to support or replace clinicians in their decision-making processes. 11 , 12 Given this context, understanding the level and variation of evidence underpinning AIaMDs which have received regulatory approval for clinical use may help to identify best practices and opportunities for improvement in AIaMD evidence generation and appraisal. This would support the use of AIaMDs that are safe, effective, and beneficial for the populations they aim to serve. As such, this scoping review focuses specifically on ophthalmic imaging AIaMDs that help inform clinical management and which have received regulatory approval in three jurisdictions with established regulatory pathways – Europe, Australia, and the United States of America (USA). The study objectives were: 1. To identify ophthalmic imaging AIaMDs with regulatory approval for clinical use in Europe, Australia, and the USA (covering all forms of market approval within that jurisdiction); 2. To describe the characteristics of these AIaMDs and the regulatory approvals granted to them; 3. To report and characterise the available evidence on model performance and clinical outcomes for these AIaMDs. RESULTS Characteristics of eligible ophthalmic imaging AIaMDs Forty-four potentially eligible AIaMDs for ophthalmic imaging were identified. Eight AIaMDs were excluded for the following reasons: they focused on image quality or denoising alone without impacting clinical care, were AI in a medical device (AIiMD) rather than AIaMDs, or were regulator-approved image management systems or platforms which may support AI models that are not themselves approved for commercial use (Supplementary Table 1). In total, there were 36 eligible AIaMDs from 28 manufacturers. The 28 manufacturers were headquartered across a range of regions: Europe (12/28, 43%), Asia (6/28, 21%), the USA (5/28, 18%), Australasia (3/28, 11%), and the Middle East (2/28, 7%). In terms of task or intended purpose, 36% (13/36) were designed for diabetic retinopathy (DR) screening or detection alone. 28% (10/36) could detect multiple fundus pathologies - of these, eight focused on three common conditions (DR, age-related macular degeneration (AMD) and glaucoma, depending on the jurisdiction), one detected these conditions and 9 other diseases, and one highlighted pathological findings and diseases on fundus images. One (3%) (1/36) performed glaucoma screening. 19% (7/36) performed optical coherence tomography (OCT) segmentation for detecting or monitoring diseases and/or biomarkers. The remainder were designed for oculomics tasks (inferences about systemic health from via ophthalmic biomarkers, most commonly obtained through retinal imaging 13 ) alone (2/36, 6%), oculomics tasks plus detection of DR, AMD, and glaucoma (2/36, 6%), or assessing microaneurysm turnover in DR to aid prediction and monitoring (1/36, 3%) (Fig. 1 ). Input ophthalmic imaging modalities were either colour fundus photographs (CFP) (29/36, 81%) or retinal OCT scans (7/36, 19%). One AIaMD (ARDA, Verily Health) which was trained for DR screening using colour fundus images as inputs has also been used in ultrawidefield pseudocolour images as Optos AI (unable to confirm status of CE mark). For AIaMDs using CFPs, three were approved for clinical use or tested on images captured on handheld cameras - two on both tabletop and handheld devices (AEYE-DS, AEYE Health; SELENA+, EyRIS), and one on a handheld device only (Medios AI, Remidio). The remainder utilised a range of standard tabletop imaging devices. 58% (21/36) were paired with an image quality assessment system; the status was unclear in the remainder. Deep learning models constituted the majority (29/36, 81%), of which 18 utilised convolutional neural networks and 11 did not specify the model architecture. Support vector machines represented a smaller proportion (2/36, 6%). For the remaining five AIaMDs, the model type could not be ascertained as this information was not provided by the manufacturer nor available publicly. Characteristics of regulatory approvals Almost all (35/36, 97%) were approved for use in the EU, and only 22% (8/36) and 8% (3/36) in Australia and the USA respectively (Fig. 2 and Table 1). 72% (26/36) were approved in a single jurisdiction – 67% (24/36) in the EU alone and 3% (1/36) in the USA alone; the remainder were approved across two jurisdictions – the EU and USA (6%, 2/36), or the EU and Australia (22%, 8/36). None were approved across all three jurisdictions. While there is some variation in regulatory classification across the three jurisdictions, broadly speaking, the EU, ARTG (Australia), and FDA (USA) have three classes of medical devices, and the class assigned increases with the perceived risk level of the device. Class I AIaMDs pose the lowest risk to patient safety, and class III represents the highest risk. For AIaMDs approved for use in the EU, the majority were qualified as CE class IIa (23/35, 66%), followed by class I (10/35, 29%), and class IIb (1/35, 3%). For Australia, the AIaMDs were class IIa (6/8, 75%) or class I (2/8, 25%) only (Table 1). All regulatory approvals in the USA were class II. The UK is a separate jurisdiction within Europe that accepts the CE mark. All 13 ophthalmic imaging AIaMDs registered on PARD (UK) were also approved for commercial use in the EU. These products had the same regulatory classes in both jurisdictions and have therefore not been considered separately. Details on pivotal trials supporting regulatory approval were only available from summary documents on the FDA (USA) website; clinical evidence supporting regulatory approval was not available in the public domain for all other regulatory bodies. Study characteristics The PubMed search identified 1164 studies, and manual searches (reference lists, correspondence with manufacturers, information on manufacturer websites) identified an additional 37 unique studies. Following de-duplication and abstract screening, 152 papers underwent full text review, resulting in 131 studies eligible for inclusion in the scoping review. The search strategy for each AIaMD is presented in Supplementary Table 1, and the PRISMA flow diagram for study selection in Supplementary Fig. 1. Overall, the 36 AIaMDs were supported by 131 clinical evaluation studies (range 0–22, median 2, interquartile range (IQR) 1–6). Overall, 19% (7/36) of commercially available AIaMDs did not have published peer-reviewed evidence supporting their efficacy. 22% (8/36) AIaMDs were supported by one validation study only. In total, only 37% (49/131) of studies were conducted independently of the manufacturer. The remaining studies were directly funded by the manufacturer (14/131, 11%), were co-authored by researchers affiliated with the manufacturer (80/131, 61%), or both (79/131, 60%). Model version was generally poorly reported across all studies (27%, 35/131). On a study-level, 22% (29/131) included comparisons of the AIaMD against human performance with no additional reference standard. Only 8% (10/131) of studies performed head-to-head comparisons of two or more AIaMDs. Sample size calculations were performed in 22% (29/131), of which 5 did not meet the required sample size. Only 11 studies (8%) were interventional, meaning that the AIaMD impacted clinical care. Of these, 3 were post-deployment studies where data from routine clinical care was analysed retrospectively, and 8 were experimental (7 non-randomised prospective studies, 1 RCT). These studies encompassed 7 different AIaMDs with a DR screening use case; of these, 2 (iGradingM 14 and Retmarker/ DAIRET 15 ) have been deployed in the Scottish and Portuguese national DR screening services respectively for over a decade. The remaining studies were non-interventional, and were predominantly retrospective in nature (71/120, 59%). Distinguishing ‘silent’ trials (also known as translational trials) with certainty in this cohort was not always possible due to the ambiguous descriptions of study methodology in many cases. These data are summarised in Table 2 . Dataset characteristics The 131 studies described 192 datasets or patient cohorts across 31 countries, most commonly the USA (39), China (27), the UK (15), India (15), France (13), and Singapore (13) (Fig. 3 ). 25% (48/192) of the datasets were from low- and middle-income countries (LMICs) (based on the World Bank’s Classification). 16 The datasets were mostly from multiple sites (107/192, 56%). Dataset size ranged from 19 to 30,000 patients for datasets where the numbers of patients were reported; this could not be summarised due to the heterogeneity of the unit of reporting (patient, visit, eye, or image). Demographic subgroups were poorly reported across the 192 datasets – age was reported in 52% (101/192), sex in 51% (97/192), and ethnicity in 21% (40/192). Study duration (or duration of data collection) was reported in 54% (103/192) only. 45% (87/192) of the datasets used for validation were from a range of publicly available datasets with different levels of data accessibility, 17 such as Messidor/ Messidor-2 (8 instances); datasets from pre-existing epidemiological studies such as the Singapore Epidemiology of Eye Diseases study (8 instances), AREDS study (3 instances), or the UK Biobank (3 instances); or landmark RCTs such as the HARBOR trial (2 instances), or the HAWK, HARRIER, and FILLY trials (1 instance of each). Reference standard setting Reference standard setting was variable. For the 167 datasets used to evaluate AIaMD diagnostic accuracy, reference standards were typically determined by experienced human graders grading the same image used as inputs for the AIaMD, although a small subset used the findings from routine clinical care (e.g. dilated fundus examination), or different imaging protocols (e.g. 7-field ETDRS or 4-wide field photography protocol for DR screening), or both, as the reference standard. Datasets were labelled by 1 grader (29/167, 17%), 2 graders (43/167, 26%), 3 or more graders (41/167, 25%), or not specified in the remainder. The approach to adjudication varied considerably as well. Single grader studies did not require adjudication, although some elected to adjudicate those cases where the AIaMD and the human grader disagreed. For disagreements between 2 or more graders, many did not require additional adjudicators, instead opting for consensus discussion, a majority voting rule, or re-review in a round robin fashion until consensus was achieved. Others sought the input of an additional senior clinician to arbitrate. The majority (84%, 141/167) provided some description of the graders involved in setting the reference standard, predominantly by stating the profession (e.g. ophthalmologist, retinal specialist, non-ophthalmologist grader). The graders’ level of experience was not well characterised overall, with many citing “trained graders”, “experienced graders”, or “experts”, without elucidating the number of years of experience or familiarity with the specific task. DISCUSSION This scoping review has identified and described the characteristics of ophthalmic imaging AIaMDs with regulatory approval for clinical use. The available evidence for the effectiveness of these AIaMDs has also been curated and characterised. Thirty-six ophthalmic imaging AIaMDs with regulatory approvals in Europe, Australia, and the USA were identified. They serve four main intended uses: detection or screening of 1) DR screening or 2) DR and other fundus pathologies, 3) OCT segmentation for biomarker and/or disease detection or progression, and 4) oculomics tasks. The heavy emphasis on DR aligns with a significant public health need, given that DR is a leading cause of preventable blindness in the working-age population, and early detection and intervention can reduce the risk of vision loss. 18 As diabetes becomes more common globally, there is an opportunity for AIaMDs to help improve the scalability and efficiency of screening processes, alleviate some of the burden on healthcare systems, and improve access to care. Existing national or regional DR screening programmes lend themselves well to AI integration due to their standardised nature and pre-existing quality assurance frameworks, particularly as many mandate double-reader screening for a subset of cases. 19 However, there is significant scope for expanding AIaMD applications to other imaging modalities, ocular conditions, and use cases as well. In particular, there is rising interest in further oculomics applications to detect or predict the risk of chronic systemic diseases with a significant morbidity and mortality burden, including neurodegenerative diseases such as Alzheimer’s dementia or Parkinson's disease. 20 Tools such as target product profiles, which are well-established in other fields and are in development for AIaMDs, 21 can guide product development and evaluation by laying out the requirements necessary for successful implementation. This may help accelerate the development of AIaMDs that align with stakeholders’ needs. It is also important to note how the interplay between regulatory approval, development costs, and reimbursement structures may shape the commercialisation strategies for AIaMDs, affecting both their availability and the scope of applications pursued by manufacturers. 22 Notably, nearly all 36 AIaMDs were commercially available in the EU, but only three were approved for use in the USA. This discrepancy is likely to be multifactorial. We speculate that key contributors may include the varied value propositions and reimbursement structures for tools across different healthcare systems, as well as differing regulatory frameworks across jurisdictions. 23 , 24 For example, the clinical evidence requirements appear to differ substantially - all FDA-authorised ophthalmic imaging AIaMDs to date have been supported by pivotal trials, whereas several EU MDR approvals have been based on retrospective observational data, which has obvious time and financial implications. In addition, the EU market comprises multiple different healthcare systems with diverse reimbursement models, whereas AIaMDs in the USA must secure reimbursement through Medicare, a process that can be particularly challenging in a fee-for-service paradigm. Notably, the FDA has designed a ‘Breakthrough Device Designation’ pathway to expedite regulatory review and facilitate increased regulator interaction and support with commercialisation for eligible devices, potentially leading to faster market access and hence patient benefit, over conventional pathways. 25 One AIaMD we have identified (IDx-DR/ LumineticsCore, Digital Diagnostics) has previously benefited from this, and another (CLaiR, Toku Eyes) has latterly received this designation. This pathway was established in 2015 but does not appear to have contributed significantly to addressing the discrepancy, suggesting that market factors may play a more significant role. Future work should consider qualitative research to elucidate the true underlying reasons for these differences, and to consider how the regulation of AI health technologies can balance safety and maximise patient benefit. This study found that many clinical validation studies were predominantly or solely conducted on existing datasets. These included retrospective open access datasets, epidemiological studies, and data repurposed from previous RCTs, which tend to have strict eligibility criteria and may not reflect real-world practice settings. It has previously been reported that publicly available ophthalmic imaging datasets tend towards inadequate reporting of basic demographic characteristics (age, sex, ethnicity), disparities in representation of different population and disease groups, and uneven geographical distributions, highlighting issues of health data poverty that may encode biases into AI models. 17 , 26 In addition, their differing disease prevalence and relatively high image quality may not reflect real world clinical practice, potentially affecting their suitability for robust clinical evaluation of AIaMDs. To mitigate this, future validation studies should consider the STANDING Together (STANdards for data Diversity, INclusivity, & Generalisability) recommendations for documenting and using health datasets in developing and testing AI health technologies, 27 as well as model cards or similar initiatives that encourage transparency of model reporting, including details on training datasets where feasible, to enhance accountability and mitigate biases while respecting proprietary constraints. 28 In addition, we demonstrate that the evidence base for ophthalmic imaging AIaMDs with regulatory approvals remains heavily weighted towards retrospective and observational studies. This mirrors findings from a 2021 review of FDA-authorised AIaMDs, which found that few regulatory submissions reported prospective data. 29 While leveraging large retrospective datasets is resource- and cost-effective, it has become increasingly recognised that this is merely an initial step, and that AI deployment requires a sociotechnical approach to inform safe integration into current clinical workflows. 30 Testing the fragility of AIaMD performance in prospective implementation-focused trials (either silent or interventional) is essential to identify challenges that may not be apparent in silico , and may help drive improvements in model design, training, and deployment strategies. 31 , 32 This is particularly important for AIaMDs that are intended for use as clinical decision support tools, in which incorporation and evaluation of human-computer interaction is essential. Our review found that few studies of commercially available AIaMDs examined their performance in a real-world clinical workflow. There was significant variation in the number and depth of validation studies across the AIaMDs under study. IDx-DR/ LumineticsCore (Digital Diagnostics Inc.) exemplifies high quality evidence, with external validation across a wide range of countries, population groups, and study types demonstrating real-world clinical effectiveness. Beyond diagnostic performance, this AIaMD has been tested in a RCT demonstrating improved adherence to follow-up compared to traditional referral routes. 33 Post-deployment studies have also demonstrated the utility of AI-driven point-of-care screening in improving patient access to DR screening and closing the health equity gap, 34 while also improving ophthalmology follow-up rates for patients with referrable DR, potentially by reducing the time taken to receive their screening results. 35 While RCTs are the gold standard for generating evidence in many fields of medicine, whether they are necessarily the best method of validating AIaMDs’ safety and effectiveness remains to be determined, given that the problems AIaMDs address often lack a reference standard, and human-computer interactions and explainability issues may limit replicability and reproducibility. At the very minimum, for diagnostic AI, moving beyond diagnostic accuracy metrics to real-world evidence including patient-centered and implementation-related outcomes will be instrumental in making the case for real-world deployment and integration into the clinical workflow. One-fifth of ophthalmic imaging AIaMDs did not have publicly available peer-reviewed evidence supporting their effectiveness. This does not necessarily equate to an absence of evidence, as some manufacturers choose not to publish results of studies submitted to regulatory bodies or conferences. However, this raises important questions about the incentives for manufacturers to invest in, conduct, or publish rigorous studies on their AIaMDs beyond regulatory requirements, particularly given the significant financial, logistical and time costs, 36 especially for small and medium-sized enterprises with limited resources. Without strong incentives – whether regulatory, financial, or reputational – manufacturers may not necessarily prioritise evidence generation for real-world deployment. This pushes the due diligence on to cross-functional AI adopter teams, which may have varying levels of resources and different processes for obtaining and critically appraising this evidence. Alternatively, evidence of AIaMD performance can be generated independently of the manufacturer, either by facilitating participation in research led by academic institutions or conducting post-deployment studies. This can be helpful in providing objective evidence of performance, but was only the case for one-third of studies identified. For example, in three researcher-led head-to-head comparison studies of multiple AIaMDs, several manufacturers either did not respond to enquiries or ultimately withdrew from participation, citing commercial or unspecified reasons. 37 – 39 Facilitating greater transparency from vendors is a key step in building trust in AIaMDs among stakeholder groups. Possible strategies could include regulatory mandates for public disclosure of clinical evidence from development through to post-market surveillance (particularly for more mature AIaMDs), 40 supported by additional funding, which could help align AI development and deployment with the ethical imperatives of safety, inclusivity, and equity. Conducting this scoping review has surfaced several challenges in navigating regulatory databases due to limited access and/or search functionality, data fragmentation, and a dearth of useful information. This presents a real challenge to healthcare provider organisations considering AI implementation, who are unlikely to have the resources or expertise to identify all regulated AIaMDs that may meet their needs. To mitigate this, establishing public-facing databases could facilitate stakeholder access to information about available products, their performance, and safety risks. 41 This approach has been led by the field of radiology, with examples such as the Health AI Register listing regulator-approved AI products, 42 or the Royal College of Radiologists’ AI registry featuring AIaMDs being deployed or tested in the UK. 43 Other groups have developed an open-access database summarising information about FDA-approved AIaMDs. 44 National or international registries, for example through a federated registration approach, 45 could also help standardise the reporting and evaluation of AIaMDs, and ensure that information is accessible, consistent, and reliable to inform successful implementation. This study identified significant variability in reference standard setting, in terms of the number and experience level of graders as well as the arbitration process. Image-based reference standards are subjective by nature, and interpretation may sometimes differ even between experts, potentially leading to inconsistencies in the labelling and ground-truthing process. 46 , 47 Any variation in the reference standards against which AIaMDs are evaluated can influence performance metrics and affect the perceived effectiveness of these models. 48 To address this, researchers should consider increasing the number and experience of graders required, and ensure an unbiased arbitration process, all while carefully balancing the trade-off between the quality of labelling and the resources required. 49 In any case, transparency in this process is a valuable safety mechanism, but information on reference standard setting was not always clearly documented in the studies identified. Another key consideration is whether and how AIaMD performance may be influenced by the imaging device used. Differences in hardware may produce variations in image resolution, size, field of view, and quality. Several AIaMDs identified in our searches have reported differences in model performance across some types of cameras used to capture colour fundus images, 50 , 51 which was not necessarily the case across all AIaMDs. 52 For other modalities such as OCT scans, re-training AI models may be necessary to optimise performance in devices from other manufacturers. 47 This may of course vary depending on the diversity of training data for each model. Nevertheless, ensuring that AIaMDs are robust across imaging devices from different manufacturers would benefit from extensive testing with diverse datasets. In addition, re-validation (with or without re-training) is essential to optimise AIaMD performance in new devices, and aligns with regulatory requirements, such as the FDA’s mandate to validate and re-certify each new device to ensure full regulatory compliance. However, this does not appear to be a mandatory requirement for the EU and Australia. Notably, the intended use statements for FDA-approved AIaMDs such as IDx-DR/ LumineticsCore, EyeArt, and AEYE-DS specify the imaging device(s) with which they are allowed to be used. This was not the case for the EU and Australia. The imaging device used was not always well-documented in the validation studies we identified as well. Oculomics is an emerging field, as evidenced by the 4 AIaMDs with regulatory approvals that we have identified. However, performing clinical validation for such AIaMDs may pose unique challenges. These models differ from standard diagnostic AI models in several key aspects, such as the need to handle more diverse and complex data types, including multimodal data combining ophthalmic imaging, systemic information or imaging, and/or genomic data. In addition, demonstrating the ability to predict a range of systemic conditions that may not have well-defined clinical endpoints (e.g., the presence or absence of a specific disease) renders establishing a ground truth more difficult. Additionally, they require integration with diverse clinical workflows in other fields beyond ophthalmology. The potential for these models to reveal previously unknown associations between ocular and systemic health raises questions about clinical interpretability, generalizability, and ethical considerations as well. Several challenges were encountered in the conduct of this scoping review, which highlight broader issues in the landscape of AIaMD evaluation. A substantial proportion of manufacturers (18/28, 64%) did not respond to requests for further information or clarification on their AIaMD(s). To mitigate this, the missing data was supplemented with publicly accessible sources wherever possible, and multiple methods of corroboration were employed, including conducting searches of manufacturers’ websites, evaluating peer-reviewed publications, and internet search engines. It is important to highlight that only the FDA has made a summary of regulatory documents publicly available for each AIaMD – this was not the case for the other regulatory agencies. The findings presented in this review are therefore contingent upon the quality and availability of data from these pragmatic methods, and reflect the most accurate information obtainable under these constraints. This is also likely to be the same evidence that decision-makers are presented with to make a decision on procurement. The search functionality of the databases was not well suited to identifying software medical devices with and without AI components, particularly class 1 devices, for which registration on EUDAMED is not currently mandated. The scope of this review also excluded AIiMD (as opposed to AIaMD) as there was no apparent means to construct a search strategy with meaningful sensitivity for such regulatory approvals in current databases. As such, the two hardware/software ‘system’ products with AI components for ophthalmic image analysis, SCANLY home monitoring (Notal Vision Inc.) and EyeLib (MIKAJAKI SA) which were identified through separate searches were therefore not included. Several studies did not explicitly name the AIaMD they were evaluating. This omission made determining the relevance of a given paper challenging on occasion, and a pragmatic approach in assessing eligibility was therefore taken. Some AIaMDs also undergo name changes across versions, or are marketed under different names in various jurisdictions. For example, the AIaMD originally known as the Iowa Detection Program was rebranded commercially as IDx-DR and subsequently LumineticsCore (depending on the jurisdiction). As these devices transition from academic to commercial products, clear documentation of naming as well as versioning would facilitate future research such as comparative studies and systematic reviews. Tracking the specific version of the AIaMD used in each study is also essential for assessing performance, particularly when updates or retraining could significantly impact clinical outcomes. Unfortunately, this information was frequently poorly recorded in the studies reviewed. This is a requirement of the CONSORT-AI extension 53 reporting guideline for RCTs involving AI models, and should be considered for other types of validation to improve transparency and replicability. The scoping review had an Anglocentric focus by design, and included only AIaMDs with regulatory approvals across three jurisdictions: Australia, Europe, and the USA. This was a pragmatic choice given that these jurisdictions possess centralised regulatory databases that facilitated our search process (albeit with certain limitations in their search functionalities and level of transparency), are members of the International Medical Device Regulators Forum, and have a well-established history of authorising AIaMDs for their markets. Exploring regulatory approvals in other regions such as Asia, South America, or the Middle East would offer valuable insights, especially considering the rapid advancements in AI health technologies there. Future research could aim to address this gap by exploring alternative data sources or collaborating with local experts to systematically map AIaMD development. Finally, this study focused on peer-reviewed publications identified through PubMed searches only, omitting evidence that exists only in preprints or conference abstracts. This was a pragmatic decision aimed at ensuring the reliability and scientific rigor of the included studies. In addition, some manufacturers may opt to submit evidence directly to regulatory bodies without pursuing publication in peer-reviewed journals, which would lead to underrepresentation in the academic literature, which is an inherent limitation of the current regulatory process. In summary, a growing number of ophthalmic imaging AIaMDs have passed regulatory approval for clinical use globally, though availability varies substantially between jurisdictions and identifying them can be challenging. These AIaMDs predominantly focus on the detection of posterior segment diseases from CFPs, particularly DR. There is scope for expanding AIaMD applications to other imaging modalities, ocular conditions, and use cases. Greater emphasis should be placed on accurate and transparent reporting of datasets to highlight risks of varied subgroup performance; this is critical to ensuring equitable performance as some populations may be underrepresented in the training data. The evidence available to evaluate the effectiveness of individual AIaMDs is extremely variable, with a focus on retrospective diagnostic accuracy study designs, but limited data on outcomes related to real-world implementation. A requirement for more high-quality prospective implementation studies may help promote transparency and confidence in performance for end-users. Finally, regulatory frameworks for AIaMDs may benefit from a more standardised approach to evidence reporting. This could provide clarity for manufacturers as they plan their clinical evaluation strategies, and provide potential adopters with more of the information they need to make responsible choices about AI innovation. METHODS In line with the primary objectives of this study, a scoping review was selected in preference to a systematic review. This was because our purpose in conducting this review was to identify relevant AIaMDs for ophthalmic imaging and map the available evidence for effectiveness to identify research gaps, instead of providing an unbiased and precise effect estimate. 54 Protocol and registration The review adheres to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) 55 framework where applicable. The protocol was registered at https://osf.io/cmkyv and published prior to full execution. 56 The methodology is summarised below. Eligibility The review focused on AIaMDs using ophthalmic imaging to help inform clinical management, which have regulatory approvals in the USA, Australia, and Europe. No restrictions were placed on the type of imaging modality or the intended use. AIaMDs were defined as having a partial or fully data-led mechanism, rather than an exclusively rule-based mechanism. 57 With regards to the evidence underpinning each AIaMD, only primary research evaluating performance in human participants was included. Eligible study types included randomised controlled trials (RCT), non-randomised interventional studies, ‘silent’ trials, or retrospective observational studies. Systematic reviews and meta-analyses, case series, case reports, commentaries, and expert opinions were not eligible. No date or language restrictions were applied to the electronic search. Only peer-reviewed publications were considered. Preprints and conference abstracts were ineligible. Search strategy and sources of information To identify potentially eligible AIaMDs, the following regulatory databases were searched: the Food and Drug Administration (FDA, USA) database, the Australian Register of Therapeutic Goods (ARTG, Australia), the Public Access Registration Database (PARD, United Kingdom), and the European Database on Medical Devices (EUDAMED, European Union (EU)). A tailored search strategy was designed to circumvent the challenges inherent in navigating existing regulatory databases (such as limited search functionality, transparency, and lack of AI-specific global medical device nomenclature limiting identification); this involved an exhaustive review of the product class codes and predicate devices (if applicable) with which each known eligible device was associated. The search commenced with a list of 15 AIaMD for ophthalmic imaging. This represented the sum of the authors’ awareness of regulated products at the start of the search process and a pragmatic search of relevant academic literature. 58 This strategy was adopted due to limitations in the search functionality of these regulatory databases. No AI tools were used in the search process. Following confirmation of AIaMD eligibility, PubMed was systematically searched up to 24 July 2024 for publications relevant to each AIaMD and its manufacturer by combining both search terms with an “OR” Boolean operator. Where appropriate, these searches were limited to relevant ophthalmology-specific studies using relevant key terms, for example “retin*” for AIaMDs relating to diabetic retinopathy. The search terms and number of hits are presented in the supplementary materials (Supplementary Table 2). Manual searches of reference lists and manufacturers’ websites were also conducted to identify additional peer-reviewed publications. The manufacturers of all eligible AIaMDs were contacted with a standardised email template (Supplementary Table 3) to provide clarification, corroboration, and/or additional peer-reviewed publications not identified in earlier searches. Three attempts were made to contact each manufacturer. This additional step was undertaken to help ensure that the data captured were as comprehensive as possible. Preliminary scoping searches had highlighted some areas of ambiguity, including instances where studies did not specify the name of the AIaMD or manufacturer, or cases where devices underwent a name change from one version to the next. Information on all eligible AIaMD was also collated from relevant publications identified from the above searches, and supplemented using an internet search engine (Google Search, Google). AIaMD and study selection Two authors (AK, HDJH) searched the regulatory databases independently and screened all identified AIaMDs for eligibility. Any disagreements were discussed, and if consensus could not be reached, these were resolved with recourse to a third author (ED) for arbitration. In instances where an AIaMD’s eligibility or its regulatory approval status could not be determined with publicly available evidence, the manufacturers were contacted to seek clarification (AYO). If no response was forthcoming, the ambiguity about the AIaMD’s eligibility and the rationale for including or excluding were duly recorded. Each title and abstract from the PubMed search were screened independently by two review authors (AYO and PT/MS) to determine eligibility. Full-text articles were reviewed according to the eligibility criteria set out in the protocol. At each stage, results were compared and consensus reached, with arbitration by a third reviewer (HDJH) as required. Data Extraction Data extraction was undertaken by AO (and verified by PT/ MS) in two phases, using standardised data extraction forms designed and piloted for the purposes of this review. Phase 1: The following outcomes were extracted for each eligible AIaMD: Jurisdiction under which regulatory approval was given Class assigned under FDA, TGA, UK MDR (Medical Devices Regulations 2002) and/or EU MDR (Regulation (EU) 2017/745) or MDD 93/42/EEC Intended use statement (IUS) (or manufacturer’s description of purpose when IUS was not available Ophthalmic imaging modality AI model type and architecture Phase 2: The following outcomes were extracted for each eligible study: Study information: title, author name, publication status, funding source, conflicts of interest, author affiliations with manufacturers Study methodology and outcomes: study duration, study design, internal/external validation, reference standards, comparison between AIaMD and humans, AIaMD version etc. Data set or cohort details: source of dataset, size of dataset or number of participants, setting, number of countries, number of centres, and participant demographics (age, gender, ethnicity) Data Synthesis The data for each AIaMD were synthesised to give an overview of its characteristics and that of its regulatory approval(s) through narrative and tabular approaches. Study- and cohort-level data were similarly synthesised and presented using descriptive statistics to outline the characteristics of the included studies. Differences from the protocol Two changes were made to the published protocol. 56 Firstly, although a quality assessment of eligible studies was initially planned, it was later determined that this did not align with the stated purpose of the scoping review, which sought to map the evidence for commercially available ophthalmic imaging AIaMDs. Secondly, extracting data on model performance was not carried out for similar reasons; the heterogeneity of AI models (even within the same use case), study types, study settings, populations, and technical factors (such as camera types) limited the feasibility and value of meta-analysis, even at the level of individual AIaMDs. Abbreviations AIaMD, Artificial intelligence as a medical device AIiMD, AI in a medical device AMD, Age-related macular degeneration ARTG, Australian Register of Therapeutic Goods DR, Diabetic retinopathy EU, European Union EUDAMED, European Database on Medical Devices FDA, Food and Drug Administration OCT, Optical coherence tomography PARD, Public Access Registration Database RCT, Randomised controlled trial Declarations DATA AVAILABILITY STATEMENT: Data sharing is not applicable to this article as no datasets were generated or analysed during the current study. CODE AVAILABILITY STATEMENT: Not applicable as no code was generated. ACKNOWLEDGEMENTS: This study received no direct funding. AYO is supported by a National Institute for Health Research (NIHR) - Moorfields Eye Charity (MEC) Doctoral Fellowship (NIHR303691). PAK is supported by a UK Research & Innovation Future Leaders Fellowship (MR/T019050/1) and The Rubin Foundation Charitable Trust. This research was supported by the NIHR Moorfields Biomedical Research Centre (BRC) and the NIHR Birmingham BRC. The views expressed in this publication are those of the author(s) and not necessarily those of the NHS, the NIHR, the Department of Health and Social Care, or any of the other funding bodies mentioned above, none of which have played any role in the research. AUTHOR CONTRIBUTIONS: AYO and HDJH conceptualised and designed the study. AYO, PT, MS, AUK, ERD, and HDJH acquired the data. AYO performed data analysis and interpretation. AYO prepared the first draft of the manuscript, which was critically reviewed and revised by all authors (AYO, PT, MS, AUK, ERD, TM, AK, GM, XL, PAK, AKD, HDJH). COMPETING INTERESTS: XL has received consulting fees from Hardian Health and Conceivable Life Sciences and was previously a Health Studies Scientist at Apple. PAK has acted as a consultant for Retina Consultants of America, Topcon, Roche, Boehringer-Ingleheim, and Bitfount and is an equity owner in Big Picture Medical. He has received speaker fees from Zeiss, Novartis, Gyroscope, Boehringer-Ingleheim, Apellis, Roche, Abbvie, Topcon, and Hakim Group. He has received travel support from Bayer, Topcon, and Roche. He has attended advisory boards for Topcon, Bayer, Boehringer-Ingleheim, RetinAI, and Novartis. The remaining authors do not have any conflicts of interest to declare. References Secinaro, S., Calandra, D., Secinaro, A., Muthurangu, V. & Biancone, P. The role of artificial intelligence in healthcare: a structured literature review. BMC Medical Informatics and Decision Making 21, 125 (2021). Foot, B. & MacEwen, C. Surveillance of sight loss due to delay in ophthalmic treatment or review: frequency, cause and outcome. Eye 31, 771–775 (2017). RCOphth. Facing workforce shortages and backlogs in the aftermath of COVID-19: The 2022 census of the ophthalmology consultant, trainee and SAS workforce. (2023). Arias, L. et al. Delay in treating age-related macular degeneration in Spain is associated with progressive vision loss. Eye 23, 326–333 (2009). Popescu, M. L. et al. Age-Related Eye Disease and Mobility Limitations in Older Adults. Investigative Ophthalmology & Visual Science 52, 7168–7174 (2011). Demmin, D. L. & Silverstein, S. M. Visual Impairment and Mental Health: Unmet Needs and Treatment Options. Clin Ophthalmol 14, 4229–4251 (2020). Gupta, P. et al. Different impact of early and late stages irreversible eye diseases on vision-specific quality of life domains. Sci Rep 12, 8465 (2022). Brown, M. M. et al. Age-related macular degeneration: economic burden and value-based medicine analysis. Canadian journal of ophthalmology 40, (2005). Pezzullo, L., Streatfeild, J., Simkiss, P. & Shickle, D. The economic impact of sight loss and blindness in the UK adult population. BMC Health Services Research 18, 63 (2018). Onitiu, D., Wachter, S. & Mittelstadt, B. How AI challenges the medical device regulation: patient safety, benefits, and intended uses. Journal of Law and the Biosciences lsae007 (2024) doi: 10.1093/jlb/lsae007 . Regulatory Horizons Council. The Regulation of Artificial Intelligence as a Medical Device. (2022). Abràmoff, M. D., Tobey, D. & Char, D. S. Lessons Learned About Autonomous AI: Finding a Safe, Efficacious, and Ethical Path Through the Development Process. Am J Ophthalmol 214, 134–142 (2020). Wagner, S. K. et al. Insights into Systemic Disease through Retinal Imaging-Based Oculomics. Transl Vis Sci Technol 9, 6 (2020). Mellor, J. et al. Prediction of retinopathy progression using deep learning on retinal images within the Scottish screening programme. British Journal of Ophthalmology (2024) doi: 10.1136/bjo-2023-323400 . Ribeiro, L. et al. Screening for Diabetic Retinopathy in the Central Region of Portugal. Added Value of Automated ‘Disease/No Disease’ Grading. Ophthalmologica 233, 96–103 (2014). World Bank. World Bank Country and Lending Groups – World Bank Data Help Desk. https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups?_gl=1*1q1vd9p*_gcl_au*ODkxMDMyMjkwLjE3MjYxNTYwODA . Khan, S. M. et al. A global review of publicly available datasets for ophthalmological imaging: barriers to access, usability, and generalisability. The Lancet Digital Health 3, e51–e66 (2021). Bourne, R. R. A. et al. Prevalence and causes of vision loss in high-income countries and in Eastern and Central Europe in 2015: magnitude, temporal trends and projections. British Journal of Ophthalmology 102, 575–585 (2018). Abou Taha, A., Dinesen, S., Vergmann, A. S. & Grauslund, J. Present and future screening programs for diabetic retinopathy: a narrative review. International Journal of Retina and Vitreous 10, 14 (2024). Patterson, E. J. et al. Oculomics: A Crusade Against the Four Horsemen of Chronic Disease. Ophthalmol Ther 13, 1427–1451 (2024). Macdonald, T. et al. Target Product Profile for a Machine Learning–Automated Retinal Imaging Analysis Software for Use in English Diabetic Eye Screening: Protocol for a Mixed Methods Study. JMIR Res Protoc 13, e50568 (2024). Wu, K. et al. Characterizing the Clinical Adoption of Medical AI Devices through U.S. Insurance Claims. NEJM AI 1, AIoa2300030 (2023). Van Norman, G. A. Drugs and Devices: Comparison of European and U.S. Approval Processes. JACC: Basic to Translational Science 1, 399–412 (2016). Reinstein, D. & Kanellopoulos, A. CE Mark Versus FDA Approval: Which System Has it Right? CRSTG | Europe Edition https://crstodayeurope.com/articles/2015-feb/ce-mark-versus-fda-approval-which-system-has-it-right/ . US FDA. Breakthrough Devices Program. FDA (2024). Ibrahim, H., Liu, X., Zariffa, N., Morris, A. D. & Denniston, A. K. Health data poverty: an assailable barrier to equitable digital health care. The Lancet Digital Health 3, e260–e265 (2021). STANDING Together. Recommendations for diversity, inclusivity, and generalisability in artificial intelligence health technologies and health datasets. (2023). Mitchell, M. et al. Model Cards for Model Reporting. in Proceedings of the Conference on Fairness, Accountability, and Transparency 220–229 (2019). doi: 10.1145/3287560.3287596 . Wu, E. et al. How medical AI devices are evaluated: limitations and recommendations from an analysis of FDA approvals. Nat Med 27, 582–584 (2021). McCradden, M. D., Joshi, S., Anderson, J. A. & London, A. J. A normative framework for artificial intelligence as a sociotechnical system in healthcare. PATTER 4, (2023). Beede, E. et al. A Human-Centered Evaluation of a Deep Learning System Deployed in Clinics for the Detection of Diabetic Retinopathy. in Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems 1–12 (Association for Computing Machinery, New York, NY, USA, 2020). doi: 10.1145/3313831.3376718 . Widner, K. et al. Lessons learned from translating AI from development to deployment in healthcare. Nat Med 1–3 (2023) doi: 10.1038/s41591-023-02293-9 . Wolf, R. M. et al. Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial. Nat Commun 15, 421 (2024). Huang, J. J. et al. Autonomous artificial intelligence for diabetic eye disease increases access and health equity in underserved populations. NPJ Digital Medicine 7, 196 (2024). Dow, E. R. et al. Artificial Intelligence Improves Patient Follow-Up in a Diabetic Retinopathy Screening Program. Clin Ophthalmol 17, 3323–3330 (2023). Raftery, J. et al. Theme 6: the cost of randomised trials, trends and determinants. in Clinical trial metadata: defining and extracting metadata on the design, conduct, results and costs of 125 randomised clinical trials funded by the National Institute for Health Research Health Technology Assessment programme (NIHR Journals Library, 2015). Lee, A. Y. et al. Multicenter, Head-to-Head, Real-World Validation Study of Seven Automated Artificial Intelligence Diabetic Retinopathy Screening Systems. Diabetes Care 44, 1168–1175 (2021). Tufail, A. et al. Automated Diabetic Retinopathy Image Assessment Software: Diagnostic Accuracy and Cost-Effectiveness Compared with Human Graders. Ophthalmology 124, 343–351 (2017). Fajtl, J. et al. Trustworthy Evaluation of Clinical AI for Analysis of Medical Images in Diverse Populations. NEJM AI 1, AIoa2400353 (2024). Fehr, J., Citro, B., Malpani, R., Lippert, C. & Madai, V. I. A trustworthy AI reality-check: the lack of transparency of artificial intelligence products in healthcare. Front. Digit. Health 6, (2024). Silkens, M. E. W. M., Ross, J., Hall, M., Scarbrough, H. & Rockall, A. The time is now: making the case for a UK registry of deployment of radiology artificial intelligence applications. Clinical Radiology 78, 107–114 (2023). Romion Health. Radiology Health AI Register. Radiology Health AI Register http://radiology.healthairegister.com/products/ . Royal College of Radiologists. AI Registry Listing | The Royal College of Radiologists. https://www.rcr.ac.uk/our-services/artificial-intelligence-ai/ai-registry/ . Benjamens, S., Dhunnoo, P. & Meskó, B. The state of artificial intelligence-based FDA-approved medical devices and algorithms: an online database. npj Digit. Med. 3, 1–8 (2020). Pencina, M. J., McCall, J. & Economou-Zavlanos, N. J. A Federated Registration System for Artificial Intelligence in Health. JAMA (2024) doi: 10.1001/jama.2024.14026 . Krause, J. et al. Grader Variability and the Importance of Reference Standards for Evaluating Machine Learning Models for Diabetic Retinopathy. Ophthalmology 125, 1264–1272 (2018). De Fauw, J. et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med 24, 1342–1350 (2018). Wang, Y. et al. Impact of Gold-Standard Label Errors on Evaluating Performance of Deep Learning Models in Diabetic Retinopathy Screening: Nationwide Real-World Validation Study. Journal of Medical Internet Research 26, e52506 (2024). Chen, P.-H. C., Mermel, C. H. & Liu, Y. Evaluation of artificial intelligence on a reference standard based on subjective interpretation. The Lancet Digital Health 3, e693–e695 (2021). Srinivasan, R., Surya, J., Ruamviboonsuk, P., Chotcomwongse, P. & Raman, R. Influence of Different Types of Retinal Cameras on the Performance of Deep Learning Algorithms in Diabetic Retinopathy Screening. Life (Basel) 12, 1610 (2022). Doğan, M. E. et al. Head to head comparison of diagnostic performance of three non-mydriatic cameras for diabetic retinopathy screening with artificial intelligence. Eye (Lond) 38, 1694–1701 (2024). He, S. et al. Cross-camera Performance of Deep Learning Algorithms to Diagnose Common Ophthalmic Diseases: A Comparative Study Highlighting Feasibility to Portable Fundus Camera Use. Curr Eye Res 48, 857–863 (2023). Liu, X., Cruz Rivera, S., Moher, D., Calvert, M. J. & Denniston, A. K. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Nat Med 26, 1364–1374 (2020). Munn, Z. et al. Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC Medical Research Methodology 18, 143 (2018). Tricco, A. C. et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med 169, 467–473 (2018). Ong, A. Y. et al. AI as a Medical Device for Ophthalmic Imaging in Europe, Australia, and the United States: Protocol for a Systematic Scoping Review of Regulated Devices. JMIR Res Protoc 13, e52602 (2024). OECD. Explanatory Memorandum on the Updated OECD Definition of an AI System . https://www.oecd-ilibrary.org/science-and-technology/explanatory-memorandum-on-the-updated-oecd-definition-of-an-ai-system_623da898-en (2024) doi: 10.1787/623da898-en . Grzybowski, A. & Brona, P. Approval and Certification of Ophthalmic AI Devices in the European Union. Ophthalmol Ther 12, 633–638 (2023). Tables Table 1: Characteristics of commercially available ophthalmic imaging AIaMDs and their regulatory approvals. AMD, age-related macular degeneration; DMO, diabetic macular oedema; DR, diabetic retinopathy; mtmDR, more than mild diabetic retinopathy; OCT, optical coherence tomography; NR, not recorded * Based on regulatory documents where possible; peer-reviewed literature or manufacturer website otherwise ** Incomplete availability of year of certification AIaMD Manufacturer Manufacturer HQ Imaging Modality Task* Regulatory Approvals (including year)** Model Type Image Quality Australia (TGA) USA (FDA) EU (EUDAMED) UK (MHRA) LumineticsCore (US) or IDx-DR (EU) Digital Diagnostics Inc. USA Fundus photograph Detects mtmDR (includes DMO) Breakthrough device (2018) Class II (2021, 2022) CE Class IIa (2013) Deep learning Yes Eyeart Eyenuk, Inc USA Fundus photograph Detects mtmDR and vision-threatening DR (including DMO) 510(k), Class II (2020, 2023) CE Class IIb (2015) Deep learning Yes Retmarker Screening or DAIRET (in Italy) Retmarker (part of METEDA S.r.l.) Portugal Fundus photograph Detects absence or presence of DR Class IIa (2018) CE Class IIa (2010) Deep learning Yes RetmarkerDR Retmarker (part of METEDA S.r.l.) Portugal Fundus photograph Tracks microaneurysm turnover to aid prediction of DR complications CE Class IIa (NR) NR NR SELENA+ eyRIS Pte. Ltd. Singapore Fundus photograph Detects mtmDR (including DMO), referable/non-referable glaucoma suspect and referable/non-referable AMD Class IIa (2023) CE Class IIa (2020) Deep learning Yes Automated Disease Assessment (ARDA) Verily Life Sciences USA Fundus photograph Detects DR and grades severity, detects DMO CE Class IIa (2018) Class IIa (2021) Deep learning Yes Medios AI (or Medios DR) Remidio Innovative Solutions Pvt. Ltd. India Fundus photograph Detects referable DR including DMO CE Class IIa (2023) Class IIa (2021) Deep learning Yes OphtAI Evolucare/ ADCIS (partnership) France Fundus photograph Detects DR (and grades severity), DMO, glaucoma, and AMD CE Class IIa (2019) Class IIa (2023) Deep learning Yes RetCAD Thirona Retina B.V. Netherlands Fundus photograph Detects DR, AMD, and glaucoma, and grades severity of DR and AMD. Class IIa (2020) CE Class IIa (2022) Class IIa (2022) Deep learning Yes DeepDee DeepDee Netherlands Fundus photograph Detects DR, AMD, and glaucoma. CE Class I (NR) Deep learning NR MONA DR MONA.health Belgium Fundus photograph Detects referrable DR including DMO CE Class I (NR) Deep learning Yes MONA GLC MONA.health Belgium Fundus photograph Screens for glaucoma CE Class I (2023) Deep learning Yes Retinalyze RetinaLyze System A/S (Ltd.) Denmark Fundus photograph Detects DR, AMD, and glaucoma CE Class I (2021) Support vector machine Yes AEYE-DS AEYE Health Israel Fundus photograph Detects mtmDR (including DMO) 510(k), Class II (2022) NR NR EyeCheckup URAL TELEKOMÜNİKASYON SAN. TİC. A.Ş Turkey Fundus photograph Detects mtmDR and vision-threatening DR (severe NPDR, PDR, DMO) CE Class IIa (NR) Deep learning Yes Reti-Eye Reti-CVD Medi Whale Inc. South Korea Fundus photograph Reti-Eye: Detects referrable retinal disease (DR, AMD, ERM etc.), glaucoma, and media opacities Reti-CVD: Cardiovascular risk assessment Class IIa (2021) CE Class IIa (2021) Class IIa (2022) Deep learning NR ITOS Mass Screening Voigtmann GmbH Germany Fundus photograph DR screening: DR absent, suspicion of DR, DR present CE Class IIa (2022) NR NR EyeWisdom MCS/ Nexy AI Visionary Intelligence Ltd. China Fundus photograph Detects presence of 13 retinal diseases including: DR, wet and dry AMD, glaucoma CE Class IIa (2024) Class IIa (NR) Deep learning Yes EyeWisdom DSS Visionary Intelligence Ltd. China Fundus photograph Detects presence of DR and grades severity CE Class IIa (2022) Class IIa (2021) Deep learning Yes RetInSight Fluid Monitor RetInSight GmbH Austria OCT macula OCT segmentation and measurement of fluid-related biomarkers, to facilitate monitoring of nAMD CE Class IIa (2022) Class IIa (2022) Deep learning Yes RetInSight GA Monitor RetInSight GmbH Austria OCT macula OCT segmentation and measurement of GA areas to facilitate visualisation and monitoring CE Class IIa (2023) Deep learning NR RetinAI Layer Segmentation Ikerian (formerly RetinAI Medical AG) Switzerland OCT macula OCT segmentation and measurement of retinal layers CE Class IIa (2024) Deep learning NR RetinAI Fluid Segmentation Ikerian (formerly RetinAI Medical AG) Switzerland OCT macula OCT segmentation and measurement of retinal fluid biomarkers CE Class IIa (2024) Deep learning NR RetinAI Macula Biomarkers Ikerian (formerly RetinAI Medical AG) Switzerland OCT macula OCT segmentation of macular biomarkers CE Class IIa (2024) Deep learning NR iPredict System iHealthScreen Inc; Arif Systems USA Fundus photograph iPredict-DR : Detects mtmDR or vision threatening DR iPredict-AMD: Detects referable AMD iPredict-Glaucoma: Detects glaucoma suspects Class IIa (2022) NR (2021) Class IIa (2023) Deep learning Yes TeleMedC DR grader TeleMedC PTE LTD Australia Fundus photograph Screens for DR, glaucoma, and AMD *indications vary according to jurisdictions Class IIa (2021) CE Class IIa (NR) Deep learning Yes Eyetelligence system (Assure Plus) Eyetelligence Pty Ltd; Optain Australia Fundus photograph Screens for referable eye diseases including DR, glaucoma, and AMD. Additionally, screens for CVD risks Class I (2021) CE Class I (NR) Class I (NR) Deep learning Yes Eyetelligence system (Assure) Eyetelligence Pty Ltd; Optain Australia Fundus photograph Screens for referable eye diseases including DR, glaucoma, and AMD. Class I (2019) CE Class I (NR) Deep learning Yes Diabetic Retinopathy Screening (DRISTi) Artificial Learning Systems India Private Limited (Artelus) India Fundus photograph Screens for the absence or presence of DR CE Class I (NR) Deep learning NR VUNO Med - Fundus AI VUNO South Korea Fundus photograph Identifies and locates the presence of 12 retinal abnormalities to support the diagnosis of retinal diseases CE Class IIa (2020) Class IIa (2023) Deep learning NR BioAge Toku Eyes New Zealand Fundus photograph Determines biological age to give an indication of overall health CE Class I (NR) Class I (NR) Deep learning NR CLAiR Toku Eyes New Zealand Fundus photograph Cardiovascular risk assessment Breakthrough device designation CE Class I (2024) Class I (NR) Deep learning Yes iGradingM or AutoGrader Medalytix Group Ltd → National Services Scotland UK Fundus photograph Detects absence or presence of DR (or whether image is ungradable) CE Class I (2012) Support vector machine Yes Altris AI Altris USA OCT macula OCT segmentation of retinal layers and biomarkers; detects retinal biomarkers and pathologies; measures segmentation volume and area; enables progression analysis CE Class IIa (2019) NR NR Ophthal mr-doc Italy OCT macula OCT segmentation of retinal layers and biomarkers to aid the monitoring of patients with DMO CE Class IIa (2023) Deep learning NR UMI DR ULMA Medical Technologies Spain Fundus photograph Detects absence or presence of DR CE Class IIa (2023) NR NR Table 2: Clinical evidence for each AIaMD available in the peer-reviewed literature from our searches. AIaMD, artificial intelligence as a medical device AIaMD Manufacturer Highest Level of Evidence ^ # Head-to-head comparison (against other AIaMDs) External validation Post-market evidence LumineticsCore (US) or IDx-DR (EU) Digital Diagnostics Inc. RCT, Prospective interventional Yes Multiple countries and settings Yes Eyeart Eyenuk, Inc Prospective interventional Yes Multiple countries and settings No Retmarker Screening or DAIRET (in Italy) Retmarker (part of METEDA S.r.l.) Post-deployment Yes Three countries Yes RetmarkerDR Retmarker (part of METEDA S.r.l.) Retrospective No Single country No SELENA+ eyRIS Pte. Ltd. Prospective silent No Four countries No Automated Disease Assessment (ARDA) Verily Life Sciences Prospective interventional No Four countries No Medios AI (or Medios DR) Remidio Innovative Solutions Pvt. Ltd. Prospective silent Yes Two countries No OphtAI Evolucare/ ADCIS (partnership) Retrospective No No No RetCAD Thirona Retina B.V. Prospective observational No Four countries No DeepDee DeepDee Not available N/A Not available No MONA DR MONA.health Retrospective No Single country No MONA GLC MONA.health Retrospective No Multiple countries No Retinalyze RetinaLyze System A/S (Ltd.) Prospective observational Yes Two countries No AEYE-DS AEYE Health Prospective observational Yes Single country No EyeCheckup TELEKOMÜNİKASYON SAN. TİC. A.Ş Prospective observational No Single country No Reti-Eye Reti-CVD Medi Whale Inc. Prospective observational (Reti-Eye); Retrospective (Reti-CVD) No Four countries No ITOS Mass Screening Voigtmann GmbH Not available N/A Not available No EyeWisdom MCS/ Nexy AI Visionary Intelligence Ltd. Prospective observational Yes Single country No EyeWisdom DSS Visionary Intelligence Ltd. Prospective observational Yes Single country No RetInSight Fluid Monitor RetInSight GmbH Retrospective No Multiple countries No RetInSight GA Monitor RetInSight GmbH Retrospective No No No RetinAI Layer Segmentation Ikerian (formerly RetinAI Medical AG) Retrospective No Single country No RetinAI Fluid Segmentation Ikerian (formerly RetinAI Medical AG) Retrospective No Single country No RetinAI Macula Biomarkers Ikerian (formerly RetinAI Medical AG) Retrospective No Single country No iPredict System iHealthScreen Inc; Arif Systems Prospective observational Yes Three countries No TeleMedC DR grader TeleMedC PTE LTD Prospective observational No Three countries No Eyetelligence system (Assure Plus) Eyetelligence Pty Ltd; Optain Prospective observational No Three countries No Eyetelligence system (Assure) Eyetelligence Pty Ltd; Optain Prospective observational No No Diabetic Retinopathy Screening (DRISTi) Artificial Learning Systems India Private Limited (Artelus) Not available N/A Not available No VUNO Med - Fundus AI VUNO Retrospective No Four countries No BioAge Toku Eyes Retrospective No No No CLAiR Toku Eyes Retrospective No Single country No iGradingM or AutoGrader Medalytix Group Ltd → National Services Scotland Post-deployment No Two countries Yes Altris AI Altris Not available N/A Not available No Ophthal mr-doc Not available N/A Not available No UMI DR ULMA Medical Technologies Not available N/A Not available No Note: This table presents the best available evidence identified from our search of the peer-reviewed literature in July 2024. The availability and level of evidence are presented, but the quality and methodological rigour of this evidence is not assessed (out of scope). ^ ‘We have used the following definitions for study types: Prospective observational study: Clinical data are collected prospectively, which allows for subsequent retrospective evaluation of AIaMD performance on prospectively gathered data. Prospective silent trial (also known as shadow, translational trial): The AIaMD is run in real time on live data, but its predictions are not visible to clinicians and do not affect patient care. The goal is to assess how the model performs in the target clinical environment, simulating deployment without clinical impact. Prospective interventional study: The AIaMD is prospectively deployed with outputs shown to clinicians, who may incorporate them into care decisions. This design evaluates how the AI affects clinical workflows, behaviour, and potentially patient outcomes. Randomized controlled trial (RCT): a type of prospective interventional study wherein p atients, clinicians, or clinical episodes are randomized to either an AI-assisted arm (where AI output informs care) or a control arm (usual care) to evaluate causal impact. Post-deployment monitoring: Ongoing surveillance after regulatory approval and integration of an AIaMD into routine clinical practice. # Distinguishing ‘silent’ trials with certainty in this cohort was not always possible due to the ambiguous descriptions of study methodology in several instances. In such cases, we have inferred the methodology from the available evidence provided, and have adopted a conservative approach in doing so. Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterials.pdf PRISMAScRFillableChecklist10Sept2019.pdf Cite Share Download PDF Status: Published Journal Publication published 29 May, 2025 Read the published version in npj Digital Medicine → Version 1 posted Editorial decision: Revision requested 16 Apr, 2025 Reviews received at journal 14 Apr, 2025 Reviews received at journal 12 Apr, 2025 Reviewers agreed at journal 06 Apr, 2025 Reviewers agreed at journal 05 Apr, 2025 Reviewers invited by journal 04 Apr, 2025 Submission checks completed at journal 04 Apr, 2025 First submitted to journal 26 Mar, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6026482","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":439158075,"identity":"2c0a9af1-f7d8-4830-9e68-b02afa192121","order_by":0,"name":"Ariel Yuhan Ong","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzUlEQVRIiWNgGAWjYFACxgaGByCavYGNBC0JIJrnAEQLD1G6wFokEojUwj/7cOODBAabPPnIt8ce/mC4I2dPSIvEucRmgwSGtGLD23npxjwMz4wJO+wMY5tEAsPhxI2zc8ykGYCMHkI65M8wtv8Aa5l5xkzyB8PheoJaDIC2MIC0zJfgMZPgYTicQNBhhmcYmyUSDNISN/AAHcZjcNiw5wABLXJn2B9++FBhkzi/HeSwisPy7A2ErIE4D4gOQBnEA3niDB8Fo2AUjIKRCABlTzvvL9lQpwAAAABJRU5ErkJggg==","orcid":"","institution":"Moorfields Eye Hospital NHS Foundation Trust","correspondingAuthor":true,"prefix":"","firstName":"Ariel","middleName":"Yuhan","lastName":"Ong","suffix":""},{"id":439158078,"identity":"2ead3e3f-7c78-4c55-ab77-8e3d93ceeb81","order_by":1,"name":"Priyal Taribagil","email":"","orcid":"","institution":"Moorfields Eye Hospital NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Priyal","middleName":"","lastName":"Taribagil","suffix":""},{"id":439158080,"identity":"2ae7fded-a5fa-40c0-97ce-420b7e4cc8f7","order_by":2,"name":"Mertcan Sevgi","email":"","orcid":"","institution":"Moorfields Eye Hospital NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Mertcan","middleName":"","lastName":"Sevgi","suffix":""},{"id":439158082,"identity":"7063d335-30ec-4b21-93e0-9320a15c9ea7","order_by":3,"name":"Aditya U Kale","email":"","orcid":"","institution":"University Hospitals Birmingham NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Aditya","middleName":"U","lastName":"Kale","suffix":""},{"id":439158088,"identity":"b96e2b94-8dbc-49b6-aa70-34e5b4b20acd","order_by":4,"name":"Eliot R. Dow","email":"","orcid":"","institution":"Retinal Consultants Medical Group","correspondingAuthor":false,"prefix":"","firstName":"Eliot","middleName":"R.","lastName":"Dow","suffix":""},{"id":439158089,"identity":"7bf40780-ecb9-4d96-885f-2a30fa4dd1f8","order_by":5,"name":"Trystan Macdonald","email":"","orcid":"","institution":"University Hospitals Birmingham NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Trystan","middleName":"","lastName":"Macdonald","suffix":""},{"id":439158090,"identity":"dd053f6c-c863-4082-adc7-513f959a7105","order_by":6,"name":"Ashley Kras","email":"","orcid":"","institution":"Sydney Eye Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ashley","middleName":"","lastName":"Kras","suffix":""},{"id":439158092,"identity":"a40592ef-1564-4ac9-8470-31f3bda3854d","order_by":7,"name":"Gregory Maniatopoulos","email":"","orcid":"","institution":"University of Leicester","correspondingAuthor":false,"prefix":"","firstName":"Gregory","middleName":"","lastName":"Maniatopoulos","suffix":""},{"id":439158093,"identity":"6d2dd8ac-c74d-463f-a6d6-132a60c8331f","order_by":8,"name":"Xiaoxuan Liu","email":"","orcid":"","institution":"University of Birmingham","correspondingAuthor":false,"prefix":"","firstName":"Xiaoxuan","middleName":"","lastName":"Liu","suffix":""},{"id":439158094,"identity":"2bf620de-9082-4bbb-9287-c42c55a2115e","order_by":9,"name":"Pearse A Keane","email":"","orcid":"","institution":"Moorfields Eye Hospital NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Pearse","middleName":"A","lastName":"Keane","suffix":""},{"id":439158095,"identity":"eb56ac95-6a74-4809-a206-5f426720561d","order_by":10,"name":"Alastair K Denniston","email":"","orcid":"","institution":"University Hospitals Birmingham NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Alastair","middleName":"K","lastName":"Denniston","suffix":""},{"id":439158096,"identity":"0df93163-c098-44db-ae56-3c4667847307","order_by":11,"name":"Henry David Jeffry Hogg","email":"","orcid":"","institution":"University of Birmingham","correspondingAuthor":false,"prefix":"","firstName":"Henry","middleName":"David Jeffry","lastName":"Hogg","suffix":""}],"badges":[],"createdAt":"2025-02-14 01:38:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6026482/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6026482/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41746-025-01726-8","type":"published","date":"2025-05-29T15:57:26+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":80579651,"identity":"151dc207-f7cb-4714-864e-be8dd63626db","added_by":"auto","created_at":"2025-04-14 23:09:44","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":242725,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTask(s) performed by commercially available ophthalmic imaging AIaMDs.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis figure showcases the range of tasks performed by the 36 ophthalmic image analysis AIaMDs with regulatory approvals in the European Union, Australia, and the United States of America. These AIaMDs predominantly focus on the detection of posterior segment diseases, particularly DR. (\u003cem\u003eAMD, age-related macular degeneration; DR, diabetic retinopathy; GA, geographic atrophy; OCT, optical coherence tomography)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Figure1300dpi.png","url":"https://assets-eu.researchsquare.com/files/rs-6026482/v1/015cbdc2083a5994992f935e.png"},{"id":80579650,"identity":"de337aa6-3c42-41b3-b318-936e05a26695","added_by":"auto","created_at":"2025-04-14 23:09:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":151855,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRegulatory approvals for ophthalmic imaging AIaMDs across three jurisdictions.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis figure shows the distribution of regulatory approvals for ophthalmic image analysis AIaMDs across one or more jurisdictions in the EU, Australia, and the USA, as well as the regulatory classes of AIaMDs in each jurisdiction. The majority of AIaMDs were approved for use in the EU, with a proportion of these also having been approved in a second jurisdiction. (\u003cem\u003eEU, European Union; USA, United States of America; AIaMD, Artificial intelligence as a medical device)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Figure2300dpi.png","url":"https://assets-eu.researchsquare.com/files/rs-6026482/v1/9b0f22392e864e8f76783bdb.png"},{"id":80582140,"identity":"b9fc9871-5e74-4de4-9a0c-b6fb38d0df92","added_by":"auto","created_at":"2025-04-14 23:25:44","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":183708,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGeographical distributions of datasets\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis figure shows the geographical distributions of dataset instances used in validation studies of ophthalmic imaging artificial intelligence as a medical device commercially available in Europe, Australia, and the United States of America (USA). The USA and China were the most common sources.\u003c/p\u003e","description":"","filename":"Figure3300dpi.png","url":"https://assets-eu.researchsquare.com/files/rs-6026482/v1/74d5c30bb0090ebeace202e1.png"},{"id":83782875,"identity":"bcdb64e5-c2ad-49de-b8ac-87a978ec1979","added_by":"auto","created_at":"2025-06-02 16:07:52","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1815318,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6026482/v1/00ac9bb7-f0e6-4db6-9866-ccf1cde80bbf.pdf"},{"id":80579658,"identity":"dbf609c6-2626-4891-884f-e053d503fdbd","added_by":"auto","created_at":"2025-04-14 23:09:44","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":258320,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterials.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6026482/v1/0634b57a98525a71c001db1d.pdf"},{"id":80582557,"identity":"f0baf555-fd33-408c-9b53-bfe3bc99e74d","added_by":"auto","created_at":"2025-04-14 23:33:44","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":107931,"visible":true,"origin":"","legend":"","description":"","filename":"PRISMAScRFillableChecklist10Sept2019.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6026482/v1/bf40640f42eb1cdae3a87b08.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Artificial intelligence as a medical device for ophthalmic image analysis: a scoping review of regulated devices","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eArtificial intelligence (AI) represents a rapidly evolving frontier in healthcare which offers transformative potential across multiple specialties. In ophthalmology, AI can help enhance diagnostic accuracy, provide insights into systemic diseases, streamline clinical and research workflows, optimise treatment, with the ultimate goal of improving patient outcomes.\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e It can potentially help in addressing challenges such as variability in subjective human interpretation, the increasing volume of complex imaging data, and the global shortage of ophthalmic specialists. This shortage has resulted in a capacity-demand imbalance that risks irreversible sight loss from treatment delays,\u003csup\u003e\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e affecting quality of life for patients and carers,\u003csup\u003e\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e and posing a significant economic burden to individuals, healthcare services, and society.\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eHowever, while AI promises significant advancements, it may also introduce new complexities in the evaluation of safety, effectiveness, and equity/bias, which are essential to ensure they achieve their intended purpose for their target population.\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e At present, the evidence requirements to support the regulatory approval of artificial intelligence as a medical device (AIaMD) are less transparent compared to more established interventions such as drugs, which follow well-established and rigorous pathways for evaluation prior to reaching the market. Evidence standards for AIaMDs are informed by regulations which are designed to be broadly applicable across all medical devices and clinical contexts. This therefore requires some level of abstraction, which allows for varied interpretations when applied to specific AIaMDs and use cases. This raises questions about what constitutes a \u0026lsquo;sufficient\u0026rsquo; level of evidence for regulatory approval or real-world deployment, particularly for AIaMDs intended to support or replace clinicians in their decision-making processes.\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eGiven this context, understanding the level and variation of evidence underpinning AIaMDs which have received regulatory approval for clinical use may help to identify best practices and opportunities for improvement in AIaMD evidence generation and appraisal. This would support the use of AIaMDs that are safe, effective, and beneficial for the populations they aim to serve. As such, this scoping review focuses specifically on ophthalmic imaging AIaMDs that help inform clinical management and which have received regulatory approval in three jurisdictions with established regulatory pathways \u0026ndash; Europe, Australia, and the United States of America (USA).\u003c/p\u003e \u003cp\u003eThe study objectives were:\u003c/p\u003e\u003cp\u003e1. To identify ophthalmic imaging AIaMDs with regulatory approval for clinical use in Europe, Australia, and the USA (covering all forms of market approval within that jurisdiction);\u003c/p\u003e\u003cp\u003e2. To describe the characteristics of these AIaMDs and the regulatory approvals granted to them;\u003c/p\u003e \u003cp\u003e3. To report and characterise the available evidence on model performance and clinical outcomes for these AIaMDs.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003eCharacteristics of eligible ophthalmic imaging AIaMDs\u003c/h2\u003e\n \u003cp\u003eForty-four potentially eligible AIaMDs for ophthalmic imaging were identified. Eight AIaMDs were excluded for the following reasons: they focused on image quality or denoising alone without impacting clinical care, were AI in a medical device (AIiMD) rather than AIaMDs, or were regulator-approved image management systems or platforms which may support AI models that are not themselves approved for commercial use (Supplementary Table\u0026nbsp;1).\u003c/p\u003e\n \u003cp\u003eIn total, there were 36 eligible AIaMDs from 28 manufacturers. The 28 manufacturers were headquartered across a range of regions: Europe (12/28, 43%), Asia (6/28, 21%), the USA (5/28, 18%), Australasia (3/28, 11%), and the Middle East (2/28, 7%).\u003c/p\u003e\n \u003cp\u003eIn terms of task or intended purpose, 36% (13/36) were designed for diabetic retinopathy (DR) screening or detection alone. 28% (10/36) could detect multiple fundus pathologies - of these, eight focused on three common conditions (DR, age-related macular degeneration (AMD) and glaucoma, depending on the jurisdiction), one detected these conditions and 9 other diseases, and one highlighted pathological findings and diseases on fundus images. One (3%) (1/36) performed glaucoma screening. 19% (7/36) performed optical coherence tomography (OCT) segmentation for detecting or monitoring diseases and/or biomarkers. The remainder were designed for oculomics tasks (inferences about systemic health from via ophthalmic biomarkers, most commonly obtained through retinal imaging\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e) alone (2/36, 6%), oculomics tasks plus detection of DR, AMD, and glaucoma (2/36, 6%), or assessing microaneurysm turnover in DR to aid prediction and monitoring (1/36, 3%) (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eInput ophthalmic imaging modalities were either colour fundus photographs (CFP) (29/36, 81%) or retinal OCT scans (7/36, 19%). One AIaMD (ARDA, Verily Health) which was trained for DR screening using colour fundus images as inputs has also been used in ultrawidefield pseudocolour images as Optos AI (unable to confirm status of CE mark). For AIaMDs using CFPs, three were approved for clinical use or tested on images captured on handheld cameras - two on both tabletop and handheld devices (AEYE-DS, AEYE Health; SELENA+, EyRIS), and one on a handheld device only (Medios AI, Remidio). The remainder utilised a range of standard tabletop imaging devices. 58% (21/36) were paired with an image quality assessment system; the status was unclear in the remainder.\u003c/p\u003e\n \u003cp\u003eDeep learning models constituted the majority (29/36, 81%), of which 18 utilised convolutional neural networks and 11 did not specify the model architecture. Support vector machines represented a smaller proportion (2/36, 6%). For the remaining five AIaMDs, the model type could not be ascertained as this information was not provided by the manufacturer nor available publicly.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eCharacteristics of regulatory approvals\u003c/h3\u003e\n\u003cp\u003eAlmost all (35/36, 97%) were approved for use in the EU, and only 22% (8/36) and 8% (3/36) in Australia and the USA respectively (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e and Table 1). 72% (26/36) were approved in a single jurisdiction \u0026ndash; 67% (24/36) in the EU alone and 3% (1/36) in the USA alone; the remainder were approved across two jurisdictions \u0026ndash; the EU and USA (6%, 2/36), or the EU and Australia (22%, 8/36). None were approved across all three jurisdictions.\u003c/p\u003e\n\u003cp\u003eWhile there is some variation in regulatory classification across the three jurisdictions, broadly speaking, the EU, ARTG (Australia), and FDA (USA) have three classes of medical devices, and the class assigned increases with the perceived risk level of the device. Class I AIaMDs pose the lowest risk to patient safety, and class III represents the highest risk. For AIaMDs approved for use in the EU, the majority were qualified as CE class IIa (23/35, 66%), followed by class I (10/35, 29%), and class IIb (1/35, 3%). For Australia, the AIaMDs were class IIa (6/8, 75%) or class I (2/8, 25%) only (Table\u0026nbsp;1). All regulatory approvals in the USA were class II.\u003c/p\u003e\n\u003cp\u003eThe UK is a separate jurisdiction within Europe that accepts the CE mark. All 13 ophthalmic imaging AIaMDs registered on PARD (UK) were also approved for commercial use in the EU. These products had the same regulatory classes in both jurisdictions and have therefore not been considered separately.\u003c/p\u003e\n\u003cp\u003eDetails on pivotal trials supporting regulatory approval were only available from summary documents on the FDA (USA) website; clinical evidence supporting regulatory approval was not available in the public domain for all other regulatory bodies.\u003c/p\u003e\n\u003ch3\u003eStudy characteristics\u003c/h3\u003e\n\u003cp\u003eThe PubMed search identified 1164 studies, and manual searches (reference lists, correspondence with manufacturers, information on manufacturer websites) identified an additional 37 unique studies. Following de-duplication and abstract screening, 152 papers underwent full text review, resulting in 131 studies eligible for inclusion in the scoping review. The search strategy for each AIaMD is presented in Supplementary Table\u0026nbsp;1, and the PRISMA flow diagram for study selection in Supplementary Fig.\u0026nbsp;1.\u003c/p\u003e\n\u003cp\u003eOverall, the 36 AIaMDs were supported by 131 clinical evaluation studies (range 0\u0026ndash;22, median 2, interquartile range (IQR) 1\u0026ndash;6). Overall, 19% (7/36) of commercially available AIaMDs did not have published peer-reviewed evidence supporting their efficacy. 22% (8/36) AIaMDs were supported by one validation study only.\u003c/p\u003e\n\u003cp\u003eIn total, only 37% (49/131) of studies were conducted independently of the manufacturer. The remaining studies were directly funded by the manufacturer (14/131, 11%), were co-authored by researchers affiliated with the manufacturer (80/131, 61%), or both (79/131, 60%).\u003c/p\u003e\n\u003cp\u003eModel version was generally poorly reported across all studies (27%, 35/131). On a study-level, 22% (29/131) included comparisons of the AIaMD against human performance with no additional reference standard. Only 8% (10/131) of studies performed head-to-head comparisons of two or more AIaMDs. Sample size calculations were performed in 22% (29/131), of which 5 did not meet the required sample size.\u003c/p\u003e\n\u003cp\u003eOnly 11 studies (8%) were interventional, meaning that the AIaMD impacted clinical care. Of these, 3 were post-deployment studies where data from routine clinical care was analysed retrospectively, and 8 were experimental (7 non-randomised prospective studies, 1 RCT). These studies encompassed 7 different AIaMDs with a DR screening use case; of these, 2 (iGradingM\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e and Retmarker/ DAIRET\u003csup\u003e15\u003c/sup\u003e) have been deployed in the Scottish and Portuguese national DR screening services respectively for over a decade. The remaining studies were non-interventional, and were predominantly retrospective in nature (71/120, 59%). Distinguishing \u0026lsquo;silent\u0026rsquo; trials (also known as translational trials) with certainty in this cohort was not always possible due to the ambiguous descriptions of study methodology in many cases.\u003c/p\u003e\n\u003cp\u003eThese data are summarised in Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\n\u003ch3\u003eDataset characteristics\u003c/h3\u003e\n\u003cp\u003eThe 131 studies described 192 datasets or patient cohorts across 31 countries, most commonly the USA (39), China (27), the UK (15), India (15), France (13), and Singapore (13) (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). 25% (48/192) of the datasets were from low- and middle-income countries (LMICs) (based on the World Bank\u0026rsquo;s Classification).\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e The datasets were mostly from multiple sites (107/192, 56%).\u003c/p\u003e\n\u003cp\u003eDataset size ranged from 19 to 30,000 patients for datasets where the numbers of patients were reported; this could not be summarised due to the heterogeneity of the unit of reporting (patient, visit, eye, or image). Demographic subgroups were poorly reported across the 192 datasets \u0026ndash; age was reported in 52% (101/192), sex in 51% (97/192), and ethnicity in 21% (40/192). Study duration (or duration of data collection) was reported in 54% (103/192) only.\u003c/p\u003e\n\u003cp\u003e45% (87/192) of the datasets used for validation were from a range of publicly available datasets with different levels of data accessibility,\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e such as Messidor/ Messidor-2 (8 instances); datasets from pre-existing epidemiological studies such as the Singapore Epidemiology of Eye Diseases study (8 instances), AREDS study (3 instances), or the UK Biobank (3 instances); or landmark RCTs such as the HARBOR trial (2 instances), or the HAWK, HARRIER, and FILLY trials (1 instance of each).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eReference standard setting\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eReference standard setting was variable. For the 167 datasets used to evaluate AIaMD diagnostic accuracy, reference standards were typically determined by experienced human graders grading the same image used as inputs for the AIaMD, although a small subset used the findings from routine clinical care (e.g. dilated fundus examination), or different imaging protocols (e.g. 7-field ETDRS or 4-wide field photography protocol for DR screening), or both, as the reference standard. Datasets were labelled by 1 grader (29/167, 17%), 2 graders (43/167, 26%), 3 or more graders (41/167, 25%), or not specified in the remainder.\u003c/p\u003e\n\u003cp\u003eThe approach to adjudication varied considerably as well. Single grader studies did not require adjudication, although some elected to adjudicate those cases where the AIaMD and the human grader disagreed. For disagreements between 2 or more graders, many did not require additional adjudicators, instead opting for consensus discussion, a majority voting rule, or re-review in a round robin fashion until consensus was achieved. Others sought the input of an additional senior clinician to arbitrate.\u003c/p\u003e\n\u003cp\u003eThe majority (84%, 141/167) provided some description of the graders involved in setting the reference standard, predominantly by stating the profession (e.g. ophthalmologist, retinal specialist, non-ophthalmologist grader). The graders\u0026rsquo; level of experience was not well characterised overall, with many citing \u0026ldquo;trained graders\u0026rdquo;, \u0026ldquo;experienced graders\u0026rdquo;, or \u0026ldquo;experts\u0026rdquo;, without elucidating the number of years of experience or familiarity with the specific task.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003e This scoping review has identified and described the characteristics of ophthalmic imaging AIaMDs with regulatory approval for clinical use. The available evidence for the effectiveness of these AIaMDs has also been curated and characterised.\u003c/p\u003e \u003cp\u003eThirty-six ophthalmic imaging AIaMDs with regulatory approvals in Europe, Australia, and the USA were identified. They serve four main intended uses: detection or screening of 1) DR screening or 2) DR and other fundus pathologies, 3) OCT segmentation for biomarker and/or disease detection or progression, and 4) oculomics tasks.\u003c/p\u003e \u003cp\u003eThe heavy emphasis on DR aligns with a significant public health need, given that DR is a leading cause of preventable blindness in the working-age population, and early detection and intervention can reduce the risk of vision loss.\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e As diabetes becomes more common globally, there is an opportunity for AIaMDs to help improve the scalability and efficiency of screening processes, alleviate some of the burden on healthcare systems, and improve access to care. Existing national or regional DR screening programmes lend themselves well to AI integration due to their standardised nature and pre-existing quality assurance frameworks, particularly as many mandate double-reader screening for a subset of cases.\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eHowever, there is significant scope for expanding AIaMD applications to other imaging modalities, ocular conditions, and use cases as well. In particular, there is rising interest in further oculomics applications to detect or predict the risk of chronic systemic diseases with a significant morbidity and mortality burden, including neurodegenerative diseases such as Alzheimer\u0026rsquo;s dementia or Parkinson's disease.\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e Tools such as target product profiles, which are well-established in other fields and are in development for AIaMDs,\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e can guide product development and evaluation by laying out the requirements necessary for successful implementation. This may help accelerate the development of AIaMDs that align with stakeholders\u0026rsquo; needs.\u003c/p\u003e \u003cp\u003eIt is also important to note how the interplay between regulatory approval, development costs, and reimbursement structures may shape the commercialisation strategies for AIaMDs, affecting both their availability and the scope of applications pursued by manufacturers.\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e Notably, nearly all 36 AIaMDs were commercially available in the EU, but only three were approved for use in the USA. This discrepancy is likely to be multifactorial. We speculate that key contributors may include the varied value propositions and reimbursement structures for tools across different healthcare systems, as well as differing regulatory frameworks across jurisdictions.\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e,\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eFor example, the clinical evidence requirements appear to differ substantially - all FDA-authorised ophthalmic imaging AIaMDs to date have been supported by pivotal trials, whereas several EU MDR approvals have been based on retrospective observational data, which has obvious time and financial implications. In addition, the EU market comprises multiple different healthcare systems with diverse reimbursement models, whereas AIaMDs in the USA must secure reimbursement through Medicare, a process that can be particularly challenging in a fee-for-service paradigm. Notably, the FDA has designed a \u0026lsquo;Breakthrough Device Designation\u0026rsquo; pathway to expedite regulatory review and facilitate increased regulator interaction and support with commercialisation for eligible devices, potentially leading to faster market access and hence patient benefit, over conventional pathways.\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e One AIaMD we have identified (IDx-DR/ LumineticsCore, Digital Diagnostics) has previously benefited from this, and another (CLaiR, Toku Eyes) has latterly received this designation. This pathway was established in 2015 but does not appear to have contributed significantly to addressing the discrepancy, suggesting that market factors may play a more significant role. Future work should consider qualitative research to elucidate the true underlying reasons for these differences, and to consider how the regulation of AI health technologies can balance safety and maximise patient benefit.\u003c/p\u003e \u003cp\u003eThis study found that many clinical validation studies were predominantly or solely conducted on existing datasets. These included retrospective open access datasets, epidemiological studies, and data repurposed from previous RCTs, which tend to have strict eligibility criteria and may not reflect real-world practice settings. It has previously been reported that publicly available ophthalmic imaging datasets tend towards inadequate reporting of basic demographic characteristics (age, sex, ethnicity), disparities in representation of different population and disease groups, and uneven geographical distributions, highlighting issues of health data poverty that may encode biases into AI models.\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e In addition, their differing disease prevalence and relatively high image quality may not reflect real world clinical practice, potentially affecting their suitability for robust clinical evaluation of AIaMDs. To mitigate this, future validation studies should consider the STANDING Together (STANdards for data Diversity, INclusivity, \u0026amp; Generalisability) recommendations for documenting and using health datasets in developing and testing AI health technologies,\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e as well as model cards or similar initiatives that encourage transparency of model reporting, including details on training datasets where feasible, to enhance accountability and mitigate biases while respecting proprietary constraints.\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eIn addition, we demonstrate that the evidence base for ophthalmic imaging AIaMDs with regulatory approvals remains heavily weighted towards retrospective and observational studies. This mirrors findings from a 2021 review of FDA-authorised AIaMDs, which found that few regulatory submissions reported prospective data.\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e While leveraging large retrospective datasets is resource- and cost-effective, it has become increasingly recognised that this is merely an initial step, and that AI deployment requires a sociotechnical approach to inform safe integration into current clinical workflows.\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e Testing the fragility of AIaMD performance in prospective implementation-focused trials (either silent or interventional) is essential to identify challenges that may not be apparent \u003cem\u003ein silico\u003c/em\u003e, and may help drive improvements in model design, training, and deployment strategies.\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e This is particularly important for AIaMDs that are intended for use as clinical decision support tools, in which incorporation and evaluation of human-computer interaction is essential.\u003c/p\u003e \u003cp\u003e Our review found that few studies of commercially available AIaMDs examined their performance in a real-world clinical workflow. There was significant variation in the number and depth of validation studies across the AIaMDs under study. \u003cem\u003eIDx-DR/ LumineticsCore\u003c/em\u003e (Digital Diagnostics Inc.) exemplifies high quality evidence, with external validation across a wide range of countries, population groups, and study types demonstrating real-world clinical effectiveness. Beyond diagnostic performance, this AIaMD has been tested in a RCT demonstrating improved adherence to follow-up compared to traditional referral routes.\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e Post-deployment studies have also demonstrated the utility of AI-driven point-of-care screening in improving patient access to DR screening and closing the health equity gap,\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e while also improving ophthalmology follow-up rates for patients with referrable DR, potentially by reducing the time taken to receive their screening results.\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e While RCTs are the gold standard for generating evidence in many fields of medicine, whether they are necessarily the best method of validating AIaMDs\u0026rsquo; safety and effectiveness remains to be determined, given that the problems AIaMDs address often lack a reference standard, and human-computer interactions and explainability issues may limit replicability and reproducibility. At the very minimum, for diagnostic AI, moving beyond diagnostic accuracy metrics to real-world evidence including patient-centered and implementation-related outcomes will be instrumental in making the case for real-world deployment and integration into the clinical workflow.\u003c/p\u003e \u003cp\u003eOne-fifth of ophthalmic imaging AIaMDs did not have publicly available peer-reviewed evidence supporting their effectiveness. This does not necessarily equate to an absence of evidence, as some manufacturers choose not to publish results of studies submitted to regulatory bodies or conferences. However, this raises important questions about the incentives for manufacturers to invest in, conduct, or publish rigorous studies on their AIaMDs beyond regulatory requirements, particularly given the significant financial, logistical and time costs,\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e especially for small and medium-sized enterprises with limited resources. Without strong incentives \u0026ndash; whether regulatory, financial, or reputational \u0026ndash; manufacturers may not necessarily prioritise evidence generation for real-world deployment. This pushes the due diligence on to cross-functional AI adopter teams, which may have varying levels of resources and different processes for obtaining and critically appraising this evidence.\u003c/p\u003e \u003cp\u003eAlternatively, evidence of AIaMD performance can be generated independently of the manufacturer, either by facilitating participation in research led by academic institutions or conducting post-deployment studies. This can be helpful in providing objective evidence of performance, but was only the case for one-third of studies identified. For example, in three researcher-led head-to-head comparison studies of multiple AIaMDs, several manufacturers either did not respond to enquiries or ultimately withdrew from participation, citing commercial or unspecified reasons. \u003csup\u003e\u003cspan additionalcitationids=\"CR38\" citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eFacilitating greater transparency from vendors is a key step in building trust in AIaMDs among stakeholder groups. Possible strategies could include regulatory mandates for public disclosure of clinical evidence from development through to post-market surveillance (particularly for more mature AIaMDs),\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e supported by additional funding, which could help align AI development and deployment with the ethical imperatives of safety, inclusivity, and equity. Conducting this scoping review has surfaced several challenges in navigating regulatory databases due to limited access and/or search functionality, data fragmentation, and a dearth of useful information. This presents a real challenge to healthcare provider organisations considering AI implementation, who are unlikely to have the resources or expertise to identify all regulated AIaMDs that may meet their needs. To mitigate this, establishing public-facing databases could facilitate stakeholder access to information about available products, their performance, and safety risks.\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e This approach has been led by the field of radiology, with examples such as the Health AI Register listing regulator-approved AI products,\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e or the Royal College of Radiologists\u0026rsquo; AI registry featuring AIaMDs being deployed or tested in the UK.\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e Other groups have developed an open-access database summarising information about FDA-approved AIaMDs.\u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e National or international registries, for example through a federated registration approach,\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e could also help standardise the reporting and evaluation of AIaMDs, and ensure that information is accessible, consistent, and reliable to inform successful implementation.\u003c/p\u003e \u003cp\u003eThis study identified significant variability in reference standard setting, in terms of the number and experience level of graders as well as the arbitration process. Image-based reference standards are subjective by nature, and interpretation may sometimes differ even between experts, potentially leading to inconsistencies in the labelling and ground-truthing process.\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e,\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e Any variation in the reference standards against which AIaMDs are evaluated can influence performance metrics and affect the perceived effectiveness of these models.\u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e To address this, researchers should consider increasing the number and experience of graders required, and ensure an unbiased arbitration process, all while carefully balancing the trade-off between the quality of labelling and the resources required.\u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e In any case, transparency in this process is a valuable safety mechanism, but information on reference standard setting was not always clearly documented in the studies identified.\u003c/p\u003e \u003cp\u003eAnother key consideration is whether and how AIaMD performance may be influenced by the imaging device used. Differences in hardware may produce variations in image resolution, size, field of view, and quality. Several AIaMDs identified in our searches have reported differences in model performance across some types of cameras used to capture colour fundus images,\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e,\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e which was not necessarily the case across all AIaMDs.\u003csup\u003e\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e\u003c/sup\u003e For other modalities such as OCT scans, re-training AI models may be necessary to optimise performance in devices from other manufacturers.\u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e This may of course vary depending on the diversity of training data for each model. Nevertheless, ensuring that AIaMDs are robust across imaging devices from different manufacturers would benefit from extensive testing with diverse datasets. In addition, re-validation (with or without re-training) is essential to optimise AIaMD performance in new devices, and aligns with regulatory requirements, such as the FDA\u0026rsquo;s mandate to validate and re-certify each new device to ensure full regulatory compliance. However, this does not appear to be a mandatory requirement for the EU and Australia. Notably, the intended use statements for FDA-approved AIaMDs such as IDx-DR/ LumineticsCore, EyeArt, and AEYE-DS specify the imaging device(s) with which they are allowed to be used. This was not the case for the EU and Australia. The imaging device used was not always well-documented in the validation studies we identified as well.\u003c/p\u003e \u003cp\u003eOculomics is an emerging field, as evidenced by the 4 AIaMDs with regulatory approvals that we have identified. However, performing clinical validation for such AIaMDs may pose unique challenges. These models differ from standard diagnostic AI models in several key aspects, such as the need to handle more diverse and complex data types, including multimodal data combining ophthalmic imaging, systemic information or imaging, and/or genomic data. In addition, demonstrating the ability to predict a range of systemic conditions that may not have well-defined clinical endpoints (e.g., the presence or absence of a specific disease) renders establishing a ground truth more difficult. Additionally, they require integration with diverse clinical workflows in other fields beyond ophthalmology. The potential for these models to reveal previously unknown associations between ocular and systemic health raises questions about clinical interpretability, generalizability, and ethical considerations as well.\u003c/p\u003e \u003cp\u003eSeveral challenges were encountered in the conduct of this scoping review, which highlight broader issues in the landscape of AIaMD evaluation.\u003c/p\u003e \u003cp\u003eA substantial proportion of manufacturers (18/28, 64%) did not respond to requests for further information or clarification on their AIaMD(s). To mitigate this, the missing data was supplemented with publicly accessible sources wherever possible, and multiple methods of corroboration were employed, including conducting searches of manufacturers\u0026rsquo; websites, evaluating peer-reviewed publications, and internet search engines. It is important to highlight that only the FDA has made a summary of regulatory documents publicly available for each AIaMD \u0026ndash; this was not the case for the other regulatory agencies. The findings presented in this review are therefore contingent upon the quality and availability of data from these pragmatic methods, and reflect the most accurate information obtainable under these constraints. This is also likely to be the same evidence that decision-makers are presented with to make a decision on procurement.\u003c/p\u003e \u003cp\u003eThe search functionality of the databases was not well suited to identifying software medical devices with and without AI components, particularly class 1 devices, for which registration on EUDAMED is not currently mandated. The scope of this review also excluded AIiMD (as opposed to AIaMD) as there was no apparent means to construct a search strategy with meaningful sensitivity for such regulatory approvals in current databases. As such, the two hardware/software \u0026lsquo;system\u0026rsquo; products with AI components for ophthalmic image analysis, SCANLY home monitoring (Notal Vision Inc.) and EyeLib (MIKAJAKI SA) which were identified through separate searches were therefore not included.\u003c/p\u003e \u003cp\u003eSeveral studies did not explicitly name the AIaMD they were evaluating. This omission made determining the relevance of a given paper challenging on occasion, and a pragmatic approach in assessing eligibility was therefore taken. Some AIaMDs also undergo name changes across versions, or are marketed under different names in various jurisdictions. For example, the AIaMD originally known as the \u003cem\u003eIowa Detection Program\u003c/em\u003e was rebranded commercially as \u003cem\u003eIDx-DR\u003c/em\u003e and subsequently \u003cem\u003eLumineticsCore\u003c/em\u003e (depending on the jurisdiction). As these devices transition from academic to commercial products, clear documentation of naming as well as versioning would facilitate future research such as comparative studies and systematic reviews. Tracking the specific version of the AIaMD used in each study is also essential for assessing performance, particularly when updates or retraining could significantly impact clinical outcomes. Unfortunately, this information was frequently poorly recorded in the studies reviewed. This is a requirement of the CONSORT-AI extension\u003csup\u003e\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e reporting guideline for RCTs involving AI models, and should be considered for other types of validation to improve transparency and replicability.\u003c/p\u003e \u003cp\u003e The scoping review had an Anglocentric focus by design, and included only AIaMDs with regulatory approvals across three jurisdictions: Australia, Europe, and the USA. This was a pragmatic choice given that these jurisdictions possess centralised regulatory databases that facilitated our search process (albeit with certain limitations in their search functionalities and level of transparency), are members of the International Medical Device Regulators Forum, and have a well-established history of authorising AIaMDs for their markets. Exploring regulatory approvals in other regions such as Asia, South America, or the Middle East would offer valuable insights, especially considering the rapid advancements in AI health technologies there. Future research could aim to address this gap by exploring alternative data sources or collaborating with local experts to systematically map AIaMD development.\u003c/p\u003e \u003cp\u003eFinally, this study focused on peer-reviewed publications identified through PubMed searches only, omitting evidence that exists only in preprints or conference abstracts. This was a pragmatic decision aimed at ensuring the reliability and scientific rigor of the included studies. In addition, some manufacturers may opt to submit evidence directly to regulatory bodies without pursuing publication in peer-reviewed journals, which would lead to underrepresentation in the academic literature, which is an inherent limitation of the current regulatory process.\u003c/p\u003e \u003cp\u003eIn summary, a growing number of ophthalmic imaging AIaMDs have passed regulatory approval for clinical use globally, though availability varies substantially between jurisdictions and identifying them can be challenging. These AIaMDs predominantly focus on the detection of posterior segment diseases from CFPs, particularly DR. There is scope for expanding AIaMD applications to other imaging modalities, ocular conditions, and use cases. Greater emphasis should be placed on accurate and transparent reporting of datasets to highlight risks of varied subgroup performance; this is critical to ensuring equitable performance as some populations may be underrepresented in the training data. The evidence available to evaluate the effectiveness of individual AIaMDs is extremely variable, with a focus on retrospective diagnostic accuracy study designs, but limited data on outcomes related to real-world implementation. A requirement for more high-quality prospective implementation studies may help promote transparency and confidence in performance for end-users. Finally, regulatory frameworks for AIaMDs may benefit from a more standardised approach to evidence reporting. This could provide clarity for manufacturers as they plan their clinical evaluation strategies, and provide potential adopters with more of the information they need to make responsible choices about AI innovation.\u003c/p\u003e "},{"header":"METHODS","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003cp\u003e In line with the primary objectives of this study, a scoping review was selected in preference to a systematic review. This was because our purpose in conducting this review was to identify relevant AIaMDs for ophthalmic imaging and map the available evidence for effectiveness to identify research gaps, instead of providing an unbiased and precise effect estimate.\u003csup\u003e\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eProtocol and registration\u003c/h3\u003e\n\u003cp\u003eThe review adheres to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews)\u003csup\u003e\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e\u003c/sup\u003e framework where applicable. The protocol was registered at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://osf.io/cmkyv\u003c/span\u003e\u003cspan address=\"https://osf.io/cmkyv\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e and published prior to full execution.\u003csup\u003e\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e The methodology is summarised below.\u003c/p\u003e\n\u003ch3\u003eEligibility\u003c/h3\u003e\n\u003cp\u003e The review focused on AIaMDs using ophthalmic imaging to help inform clinical management, which have regulatory approvals in the USA, Australia, and Europe. No restrictions were placed on the type of imaging modality or the intended use. AIaMDs were defined as having a partial or fully data-led mechanism, rather than an exclusively rule-based mechanism.\u003csup\u003e\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eWith regards to the evidence underpinning each AIaMD, only primary research evaluating performance in human participants was included. Eligible study types included randomised controlled trials (RCT), non-randomised interventional studies, \u0026lsquo;silent\u0026rsquo; trials, or retrospective observational studies. Systematic reviews and meta-analyses, case series, case reports, commentaries, and expert opinions were not eligible. No date or language restrictions were applied to the electronic search. Only peer-reviewed publications were considered. Preprints and conference abstracts were ineligible.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eSearch strategy and sources of information\u003c/h2\u003e \u003cp\u003eTo identify potentially eligible AIaMDs, the following regulatory databases were searched: the Food and Drug Administration (FDA, USA) database, the Australian Register of Therapeutic Goods (ARTG, Australia), the Public Access Registration Database (PARD, United Kingdom), and the European Database on Medical Devices (EUDAMED, European Union (EU)). A tailored search strategy was designed to circumvent the challenges inherent in navigating existing regulatory databases (such as limited search functionality, transparency, and lack of AI-specific global medical device nomenclature limiting identification); this involved an exhaustive review of the product class codes and predicate devices (if applicable) with which each known eligible device was associated. The search commenced with a list of 15 AIaMD for ophthalmic imaging. This represented the sum of the authors\u0026rsquo; awareness of regulated products at the start of the search process and a pragmatic search of relevant academic literature.\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e This strategy was adopted due to limitations in the search functionality of these regulatory databases. No AI tools were used in the search process.\u003c/p\u003e \u003cp\u003eFollowing confirmation of AIaMD eligibility, PubMed was systematically searched up to 24 July 2024 for publications relevant to each AIaMD and its manufacturer by combining both search terms with an \u0026ldquo;OR\u0026rdquo; Boolean operator. Where appropriate, these searches were limited to relevant ophthalmology-specific studies using relevant key terms, for example \u0026ldquo;retin*\u0026rdquo; for AIaMDs relating to diabetic retinopathy. The search terms and number of hits are presented in the supplementary materials (Supplementary Table\u0026nbsp;2). Manual searches of reference lists and manufacturers\u0026rsquo; websites were also conducted to identify additional peer-reviewed publications.\u003c/p\u003e \u003cp\u003eThe manufacturers of all eligible AIaMDs were contacted with a standardised email template (Supplementary Table\u0026nbsp;3) to provide clarification, corroboration, and/or additional peer-reviewed publications not identified in earlier searches. Three attempts were made to contact each manufacturer. This additional step was undertaken to help ensure that the data captured were as comprehensive as possible. Preliminary scoping searches had highlighted some areas of ambiguity, including instances where studies did not specify the name of the AIaMD or manufacturer, or cases where devices underwent a name change from one version to the next. Information on all eligible AIaMD was also collated from relevant publications identified from the above searches, and supplemented using an internet search engine (Google Search, Google).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eAIaMD and study selection\u003c/h2\u003e \u003cp\u003eTwo authors (AK, HDJH) searched the regulatory databases independently and screened all identified AIaMDs for eligibility. Any disagreements were discussed, and if consensus could not be reached, these were resolved with recourse to a third author (ED) for arbitration. In instances where an AIaMD\u0026rsquo;s eligibility or its regulatory approval status could not be determined with publicly available evidence, the manufacturers were contacted to seek clarification (AYO). If no response was forthcoming, the ambiguity about the AIaMD\u0026rsquo;s eligibility and the rationale for including or excluding were duly recorded.\u003c/p\u003e \u003cp\u003eEach title and abstract from the PubMed search were screened independently by two review authors (AYO and PT/MS) to determine eligibility. Full-text articles were reviewed according to the eligibility criteria set out in the protocol. At each stage, results were compared and consensus reached, with arbitration by a third reviewer (HDJH) as required.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eData Extraction\u003c/h2\u003e \u003cp\u003e Data extraction was undertaken by AO (and verified by PT/ MS) in two phases, using standardised data extraction forms designed and piloted for the purposes of this review.\u003c/p\u003e \u003cp\u003ePhase 1: The following outcomes were extracted for each eligible AIaMD:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eJurisdiction under which regulatory approval was given\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eClass assigned under FDA, TGA, UK MDR (Medical Devices Regulations 2002) and/or EU MDR (Regulation (EU) 2017/745) or MDD 93/42/EEC\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIntended use statement (IUS) (or manufacturer\u0026rsquo;s description of purpose when IUS was not available\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eOphthalmic imaging modality\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAI model type and architecture\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003ePhase 2: The following outcomes were extracted for each eligible study:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eStudy information: title, author name, publication status, funding source, conflicts of interest, author affiliations with manufacturers\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eStudy methodology and outcomes: study duration, study design, internal/external validation, reference standards, comparison between AIaMD and humans, AIaMD version etc.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eData set or cohort details: source of dataset, size of dataset or number of participants, setting, number of countries, number of centres, and participant demographics (age, gender, ethnicity)\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eData Synthesis\u003c/h2\u003e \u003cp\u003eThe data for each AIaMD were synthesised to give an overview of its characteristics and that of its regulatory approval(s) through narrative and tabular approaches. Study- and cohort-level data were similarly synthesised and presented using descriptive statistics to outline the characteristics of the included studies.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eDifferences from the protocol\u003c/h2\u003e \u003cp\u003eTwo changes were made to the published protocol.\u003csup\u003e\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e Firstly, although a quality assessment of eligible studies was initially planned, it was later determined that this did not align with the stated purpose of the scoping review, which sought to map the evidence for commercially available ophthalmic imaging AIaMDs. Secondly, extracting data on model performance was not carried out for similar reasons; the heterogeneity of AI models (even within the same use case), study types, study settings, populations, and technical factors (such as camera types) limited the feasibility and value of meta-analysis, even at the level of individual AIaMDs.\u003c/p\u003e \u003c/div\u003e "},{"header":"Abbreviations","content":"\u003cp\u003eAIaMD, Artificial intelligence as a medical device\u003c/p\u003e\n\u003cp\u003eAIiMD, AI in a medical device\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAMD, Age-related macular degeneration\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eARTG, Australian Register of Therapeutic Goods\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDR, Diabetic retinopathy\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEU, European Union\u003c/p\u003e\n\u003cp\u003eEUDAMED, European Database on Medical Devices\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFDA, Food and Drug Administration\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOCT, Optical coherence tomography\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePARD, Public Access Registration Database\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRCT, Randomised controlled trial\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eDATA AVAILABILITY STATEMENT:\u003c/p\u003e\n\u003cp\u003eData sharing is not applicable to this article as no datasets were generated or analysed during the current study.\u003c/p\u003e\n\u003cp\u003eCODE AVAILABILITY STATEMENT:\u003c/p\u003e\n\u003cp\u003eNot applicable as no code was generated.\u003c/p\u003e\n\u003cp\u003eACKNOWLEDGEMENTS:\u003c/p\u003e\n\u003cp\u003eThis study received no direct funding. AYO is supported by a National Institute for Health Research (NIHR) - Moorfields Eye Charity (MEC) Doctoral Fellowship (NIHR303691). PAK is supported by a UK Research \u0026amp; Innovation Future Leaders Fellowship (MR/T019050/1) and The Rubin Foundation Charitable Trust. This research was supported by the NIHR Moorfields Biomedical Research Centre (BRC) and the NIHR Birmingham BRC.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe views expressed in this publication are those of the author(s) and not necessarily those of the NHS, the NIHR, the Department of Health and Social Care, or any of the other funding bodies mentioned above, none of which have played any role in the research.\u003c/p\u003e\n\u003cp\u003eAUTHOR CONTRIBUTIONS:\u003c/p\u003e\n\u003cp\u003eAYO and HDJH conceptualised and designed the study. AYO, PT, MS, AUK, ERD, and HDJH acquired the data. AYO performed data analysis and interpretation. AYO prepared the first draft of the manuscript, which was critically reviewed and revised by all authors (AYO, PT, MS, AUK, ERD, TM, AK, GM, XL, PAK, AKD, HDJH).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCOMPETING INTERESTS:\u003c/p\u003e\n\u003cp\u003eXL has received consulting fees from Hardian Health and Conceivable Life Sciences and was previously a Health Studies Scientist at Apple. PAK has acted as a consultant for Retina Consultants of America, Topcon, Roche, Boehringer-Ingleheim, and Bitfount and is an equity owner in Big Picture Medical. He has received speaker fees from Zeiss, Novartis, Gyroscope, Boehringer-Ingleheim, Apellis, Roche, Abbvie, Topcon, and Hakim Group. He has received travel support from Bayer, Topcon, and Roche. He has attended advisory boards for Topcon, Bayer, Boehringer-Ingleheim, RetinAI, and Novartis.\u003c/p\u003e\n\u003cp\u003eThe remaining authors do not have any conflicts of interest to declare.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSecinaro, S., Calandra, D., Secinaro, A., Muthurangu, V. \u0026amp; Biancone, P. The role of artificial intelligence in healthcare: a structured literature review. BMC Medical Informatics and Decision Making 21, 125 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFoot, B. \u0026amp; MacEwen, C. Surveillance of sight loss due to delay in ophthalmic treatment or review: frequency, cause and outcome. Eye 31, 771\u0026ndash;775 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRCOphth. Facing workforce shortages and backlogs in the aftermath of COVID-19: The 2022 census of the ophthalmology consultant, trainee and SAS workforce. (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArias, L. \u003cem\u003eet al.\u003c/em\u003e Delay in treating age-related macular degeneration in Spain is associated with progressive vision loss. Eye 23, 326\u0026ndash;333 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePopescu, M. L. \u003cem\u003eet al.\u003c/em\u003e Age-Related Eye Disease and Mobility Limitations in Older Adults. Investigative Ophthalmology \u0026amp; Visual Science 52, 7168\u0026ndash;7174 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDemmin, D. L. \u0026amp; Silverstein, S. M. Visual Impairment and Mental Health: Unmet Needs and Treatment Options. Clin Ophthalmol 14, 4229\u0026ndash;4251 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGupta, P. \u003cem\u003eet al.\u003c/em\u003e Different impact of early and late stages irreversible eye diseases on vision-specific quality of life domains. Sci Rep 12, 8465 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrown, M. M. \u003cem\u003eet al.\u003c/em\u003e Age-related macular degeneration: economic burden and value-based medicine analysis. Canadian journal of ophthalmology 40, (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePezzullo, L., Streatfeild, J., Simkiss, P. \u0026amp; Shickle, D. The economic impact of sight loss and blindness in the UK adult population. BMC Health Services Research 18, 63 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOnitiu, D., Wachter, S. \u0026amp; Mittelstadt, B. How AI challenges the medical device regulation: patient safety, benefits, and intended uses. Journal of Law and the Biosciences lsae007 (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/jlb/lsae007\u003c/span\u003e\u003cspan address=\"10.1093/jlb/lsae007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRegulatory Horizons Council. The Regulation of Artificial Intelligence as a Medical Device. (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbr\u0026agrave;moff, M. D., Tobey, D. \u0026amp; Char, D. S. Lessons Learned About Autonomous AI: Finding a Safe, Efficacious, and Ethical Path Through the Development Process. Am J Ophthalmol 214, 134\u0026ndash;142 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWagner, S. K. \u003cem\u003eet al.\u003c/em\u003e Insights into Systemic Disease through Retinal Imaging-Based Oculomics. Transl Vis Sci Technol 9, 6 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMellor, J. \u003cem\u003eet al.\u003c/em\u003e Prediction of retinopathy progression using deep learning on retinal images within the Scottish screening programme. British Journal of Ophthalmology (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/bjo-2023-323400\u003c/span\u003e\u003cspan address=\"10.1136/bjo-2023-323400\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRibeiro, L. \u003cem\u003eet al.\u003c/em\u003e Screening for Diabetic Retinopathy in the Central Region of Portugal. Added Value of Automated \u0026lsquo;Disease/No Disease\u0026rsquo; Grading. \u003cem\u003eOphthalmologica\u003c/em\u003e 233, 96\u0026ndash;103 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Bank. World Bank Country and Lending Groups \u0026ndash; World Bank Data Help Desk. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups?_gl=1*1q1vd9p*_gcl_au*ODkxMDMyMjkwLjE3MjYxNTYwODA\u003c/span\u003e\u003cspan address=\"https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups?_gl=1*1q1vd9p*_gcl_au*ODkxMDMyMjkwLjE3MjYxNTYwODA\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhan, S. M. \u003cem\u003eet al.\u003c/em\u003e A global review of publicly available datasets for ophthalmological imaging: barriers to access, usability, and generalisability. The Lancet Digital Health 3, e51\u0026ndash;e66 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBourne, R. R. A. \u003cem\u003eet al.\u003c/em\u003e Prevalence and causes of vision loss in high-income countries and in Eastern and Central Europe in 2015: magnitude, temporal trends and projections. British Journal of Ophthalmology 102, 575\u0026ndash;585 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbou Taha, A., Dinesen, S., Vergmann, A. S. \u0026amp; Grauslund, J. Present and future screening programs for diabetic retinopathy: a narrative review. International Journal of Retina and Vitreous 10, 14 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatterson, E. J. \u003cem\u003eet al.\u003c/em\u003e Oculomics: A Crusade Against the Four Horsemen of Chronic Disease. Ophthalmol Ther 13, 1427\u0026ndash;1451 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMacdonald, T. \u003cem\u003eet al.\u003c/em\u003e Target Product Profile for a Machine Learning\u0026ndash;Automated Retinal Imaging Analysis Software for Use in English Diabetic Eye Screening: Protocol for a Mixed Methods Study. JMIR Res Protoc 13, e50568 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, K. \u003cem\u003eet al.\u003c/em\u003e Characterizing the Clinical Adoption of Medical AI Devices through U.S. Insurance Claims. NEJM AI 1, AIoa2300030 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan Norman, G. A. Drugs and Devices: Comparison of European and U.S. Approval Processes. JACC: Basic to Translational Science 1, 399\u0026ndash;412 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReinstein, D. \u0026amp; Kanellopoulos, A. CE Mark Versus FDA Approval: Which System Has it Right? \u003cem\u003eCRSTG | Europe Edition\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://crstodayeurope.com/articles/2015-feb/ce-mark-versus-fda-approval-which-system-has-it-right/\u003c/span\u003e\u003cspan address=\"https://crstodayeurope.com/articles/2015-feb/ce-mark-versus-fda-approval-which-system-has-it-right/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUS FDA. Breakthrough Devices Program. \u003cem\u003eFDA\u003c/em\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIbrahim, H., Liu, X., Zariffa, N., Morris, A. D. \u0026amp; Denniston, A. K. Health data poverty: an assailable barrier to equitable digital health care. The Lancet Digital Health 3, e260\u0026ndash;e265 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSTANDING Together. Recommendations for diversity, inclusivity, and generalisability in artificial intelligence health technologies and health datasets. (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMitchell, M. \u003cem\u003eet al.\u003c/em\u003e Model Cards for Model Reporting. in \u003cem\u003eProceedings of the Conference on Fairness, Accountability, and Transparency\u003c/em\u003e 220\u0026ndash;229 (2019). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/3287560.3287596\u003c/span\u003e\u003cspan address=\"10.1145/3287560.3287596\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, E. \u003cem\u003eet al.\u003c/em\u003e How medical AI devices are evaluated: limitations and recommendations from an analysis of FDA approvals. Nat Med 27, 582\u0026ndash;584 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcCradden, M. D., Joshi, S., Anderson, J. A. \u0026amp; London, A. J. A normative framework for artificial intelligence as a sociotechnical system in healthcare. \u003cem\u003ePATTER\u003c/em\u003e 4, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBeede, E. \u003cem\u003eet al.\u003c/em\u003e A Human-Centered Evaluation of a Deep Learning System Deployed in Clinics for the Detection of Diabetic Retinopathy. in \u003cem\u003eProceedings of the\u003c/em\u003e 2020 \u003cem\u003eCHI Conference on Human Factors in Computing Systems\u003c/em\u003e 1\u0026ndash;12 (Association for Computing Machinery, New York, NY, USA, 2020). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/3313831.3376718\u003c/span\u003e\u003cspan address=\"10.1145/3313831.3376718\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWidner, K. \u003cem\u003eet al.\u003c/em\u003e Lessons learned from translating AI from development to deployment in healthcare. Nat Med 1\u0026ndash;3 (2023) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41591-023-02293-9\u003c/span\u003e\u003cspan address=\"10.1038/s41591-023-02293-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWolf, R. M. \u003cem\u003eet al.\u003c/em\u003e Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial. Nat Commun 15, 421 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, J. J. \u003cem\u003eet al.\u003c/em\u003e Autonomous artificial intelligence for diabetic eye disease increases access and health equity in underserved populations. NPJ Digital Medicine 7, 196 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDow, E. R. \u003cem\u003eet al.\u003c/em\u003e Artificial Intelligence Improves Patient Follow-Up in a Diabetic Retinopathy Screening Program. Clin Ophthalmol 17, 3323\u0026ndash;3330 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRaftery, J. \u003cem\u003eet al.\u003c/em\u003e Theme 6: the cost of randomised trials, trends and determinants. in \u003cem\u003eClinical trial metadata: defining and extracting metadata on the design, conduct, results and costs of 125 randomised clinical trials funded by the National Institute for Health Research Health Technology Assessment programme\u003c/em\u003e (NIHR Journals Library, 2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, A. Y. \u003cem\u003eet al.\u003c/em\u003e Multicenter, Head-to-Head, Real-World Validation Study of Seven Automated Artificial Intelligence Diabetic Retinopathy Screening Systems. Diabetes Care 44, 1168\u0026ndash;1175 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTufail, A. \u003cem\u003eet al.\u003c/em\u003e Automated Diabetic Retinopathy Image Assessment Software: Diagnostic Accuracy and Cost-Effectiveness Compared with Human Graders. Ophthalmology 124, 343\u0026ndash;351 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFajtl, J. \u003cem\u003eet al.\u003c/em\u003e Trustworthy Evaluation of Clinical AI for Analysis of Medical Images in Diverse Populations. NEJM AI 1, AIoa2400353 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFehr, J., Citro, B., Malpani, R., Lippert, C. \u0026amp; Madai, V. I. A trustworthy AI reality-check: the lack of transparency of artificial intelligence products in healthcare. Front. Digit. Health 6, (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilkens, M. E. W. M., Ross, J., Hall, M., Scarbrough, H. \u0026amp; Rockall, A. The time is now: making the case for a UK registry of deployment of radiology artificial intelligence applications. Clinical Radiology 78, 107\u0026ndash;114 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRomion Health. Radiology Health AI Register. Radiology Health AI Register \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://radiology.healthairegister.com/products/\u003c/span\u003e\u003cspan address=\"http://radiology.healthairegister.com/products/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoyal College of Radiologists. AI Registry Listing | The Royal College of Radiologists. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.rcr.ac.uk/our-services/artificial-intelligence-ai/ai-registry/\u003c/span\u003e\u003cspan address=\"https://www.rcr.ac.uk/our-services/artificial-intelligence-ai/ai-registry/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenjamens, S., Dhunnoo, P. \u0026amp; Mesk\u0026oacute;, B. The state of artificial intelligence-based FDA-approved medical devices and algorithms: an online database. npj Digit. Med. 3, 1\u0026ndash;8 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePencina, M. J., McCall, J. \u0026amp; Economou-Zavlanos, N. J. A Federated Registration System for Artificial Intelligence in Health. JAMA (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jama.2024.14026\u003c/span\u003e\u003cspan address=\"10.1001/jama.2024.14026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrause, J. \u003cem\u003eet al.\u003c/em\u003e Grader Variability and the Importance of Reference Standards for Evaluating Machine Learning Models for Diabetic Retinopathy. Ophthalmology 125, 1264\u0026ndash;1272 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Fauw, J. \u003cem\u003eet al.\u003c/em\u003e Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med 24, 1342\u0026ndash;1350 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, Y. \u003cem\u003eet al.\u003c/em\u003e Impact of Gold-Standard Label Errors on Evaluating Performance of Deep Learning Models in Diabetic Retinopathy Screening: Nationwide Real-World Validation Study. Journal of Medical Internet Research 26, e52506 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, P.-H. C., Mermel, C. H. \u0026amp; Liu, Y. Evaluation of artificial intelligence on a reference standard based on subjective interpretation. The Lancet Digital Health 3, e693\u0026ndash;e695 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSrinivasan, R., Surya, J., Ruamviboonsuk, P., Chotcomwongse, P. \u0026amp; Raman, R. Influence of Different Types of Retinal Cameras on the Performance of Deep Learning Algorithms in Diabetic Retinopathy Screening. Life (Basel) 12, 1610 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDoğan, M. E. \u003cem\u003eet al.\u003c/em\u003e Head to head comparison of diagnostic performance of three non-mydriatic cameras for diabetic retinopathy screening with artificial intelligence. Eye (Lond) 38, 1694\u0026ndash;1701 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe, S. \u003cem\u003eet al.\u003c/em\u003e Cross-camera Performance of Deep Learning Algorithms to Diagnose Common Ophthalmic Diseases: A Comparative Study Highlighting Feasibility to Portable Fundus Camera Use. Curr Eye Res 48, 857\u0026ndash;863 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, X., Cruz Rivera, S., Moher, D., Calvert, M. J. \u0026amp; Denniston, A. K. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Nat Med 26, 1364\u0026ndash;1374 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMunn, Z. \u003cem\u003eet al.\u003c/em\u003e Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC Medical Research Methodology 18, 143 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTricco, A. C. \u003cem\u003eet al.\u003c/em\u003e PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med 169, 467\u0026ndash;473 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOng, A. Y. \u003cem\u003eet al.\u003c/em\u003e AI as a Medical Device for Ophthalmic Imaging in Europe, Australia, and the United States: Protocol for a Systematic Scoping Review of Regulated Devices. JMIR Res Protoc 13, e52602 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOECD. \u003cem\u003eExplanatory Memorandum on the Updated OECD Definition of an AI System\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.oecd-ilibrary.org/science-and-technology/explanatory-memorandum-on-the-updated-oecd-definition-of-an-ai-system_623da898-en\u003c/span\u003e\u003cspan address=\"https://www.oecd-ilibrary.org/science-and-technology/explanatory-memorandum-on-the-updated-oecd-definition-of-an-ai-system_623da898-en\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1787/623da898-en\u003c/span\u003e\u003cspan address=\"10.1787/623da898-en\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrzybowski, A. \u0026amp; Brona, P. Approval and Certification of Ophthalmic AI Devices in the European Union. Ophthalmol Ther 12, 633\u0026ndash;638 (2023).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTable 1: Characteristics of commercially available ophthalmic imaging AIaMDs and their regulatory approvals.\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAMD, age-related macular degeneration; DMO, diabetic macular oedema; DR, diabetic retinopathy; mtmDR, more than mild diabetic retinopathy; OCT, optical coherence tomography; NR, not recorded\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e* Based on regulatory documents where possible; peer-reviewed literature or manufacturer website otherwise\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e** Incomplete availability of year of certification\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"928\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 93px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAIaMD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eManufacturer\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eManufacturer HQ\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eImaging Modality\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTask*\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 313px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRegulatory Approvals (including year)**\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 71px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Type\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eImage Quality\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAustralia (TGA)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUSA (FDA)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eEU (EUDAMED)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUK (MHRA)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eLumineticsCore (US) or IDx-DR (EU)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDigital Diagnostics Inc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eUSA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects mtmDR (includes DMO)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003eBreakthrough device (2018) \u003cimg height=\"21\" src=\"data:image/png;base64,R0lGODlhCwAVAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAABwALAAUAggAAAAAAAAAAOma225Db/7ZmOtuQZv+2ZgMRCLpgxBCUAU64+IoXJe2OkgAAOw==\" alt=\"image\" width=\"11\"\u003e\u0026nbsp;Class II (2021, 2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2013)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eEyeart\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eEyenuk, Inc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eUSA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects mtmDR and vision-threatening DR (including DMO)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e510(k), Class II (2020, 2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIb (2015)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetmarker Screening or DAIRET (in Italy)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eRetmarker (part of METEDA S.r.l.)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003ePortugal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects absence or presence of DR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003eClass IIa (2018)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2010)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetmarkerDR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eRetmarker (part of METEDA S.r.l.)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003ePortugal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eTracks microaneurysm turnover to aid prediction of DR complications\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eSELENA+\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eeyRIS Pte. Ltd.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eSingapore\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects mtmDR (including DMO), referable/non-referable glaucoma suspect and referable/non-referable AMD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003eClass IIa (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2020)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eAutomated Disease Assessment (ARDA)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eVerily Life Sciences\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eUSA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects DR and grades severity, detects DMO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2018)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eMedios AI (or Medios DR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eRemidio Innovative Solutions Pvt. Ltd.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eIndia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects referable DR including DMO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eOphtAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eEvolucare/ ADCIS (partnership)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eFrance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects DR (and grades severity), DMO, glaucoma, and AMD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2019)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetCAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eThirona Retina B.V.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eNetherlands\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects DR, AMD, and glaucoma, and grades severity of DR and AMD.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003eClass IIa (2020)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eDeepDee\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDeepDee\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eNetherlands\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects DR, AMD, and glaucoma.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eMONA DR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eMONA.health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eBelgium\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects referrable DR including DMO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eMONA GLC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eMONA.health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eBelgium\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eScreens for glaucoma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetinalyze\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eRetinaLyze System A/S (Ltd.)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eDenmark\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects DR, AMD, and glaucoma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eSupport vector machine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eAEYE-DS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eAEYE Health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eIsrael\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects mtmDR (including DMO)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e510(k), Class II (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eEyeCheckup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eURAL TELEKOM\u0026Uuml;NİKASYON SAN. TİC. A.Ş\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eTurkey\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects mtmDR and vision-threatening DR (severe NPDR, PDR, DMO)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eReti-Eye\u003cbr\u003e\u0026nbsp;Reti-CVD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eMedi Whale Inc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eSouth Korea\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u003cem\u003eReti-Eye:\u003c/em\u003e Detects referrable retinal disease (DR, AMD, ERM etc.), glaucoma, and media opacities\u003cbr\u003e\u003cem\u003eReti-CVD:\u003c/em\u003e Cardiovascular risk assessment\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003eClass IIa (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eITOS Mass Screening\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eVoigtmann GmbH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eGermany\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDR screening: DR absent, suspicion of DR, DR present\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eEyeWisdom MCS/ Nexy AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eVisionary Intelligence Ltd.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eChina\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects presence of 13 retinal diseases including: DR, wet and dry AMD, glaucoma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2024)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eEyeWisdom DSS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eVisionary Intelligence Ltd.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eChina\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects presence of DR and grades severity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetInSight Fluid Monitor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eRetInSight GmbH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eAustria\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eOCT macula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eOCT segmentation and measurement of fluid-related biomarkers, to facilitate monitoring of nAMD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetInSight GA Monitor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eRetInSight GmbH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eAustria\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eOCT macula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eOCT segmentation and measurement of GA areas to facilitate visualisation and monitoring\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetinAI Layer Segmentation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eIkerian (formerly RetinAI Medical AG)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eSwitzerland\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eOCT macula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eOCT segmentation and measurement of retinal layers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2024)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetinAI Fluid Segmentation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eIkerian (formerly RetinAI Medical AG)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eSwitzerland\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eOCT macula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eOCT segmentation and measurement of retinal fluid biomarkers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2024)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eRetinAI Macula Biomarkers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eIkerian (formerly RetinAI Medical AG)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eSwitzerland\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eOCT macula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eOCT segmentation of macular biomarkers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2024)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eiPredict System\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eiHealthScreen Inc; Arif Systems\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eUSA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u003cem\u003eiPredict-DR\u003c/em\u003e: Detects mtmDR or vision threatening DR\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eiPredict-AMD:\u0026nbsp;\u003c/em\u003eDetects referable AMD\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eiPredict-Glaucoma:\u0026nbsp;\u003c/em\u003eDetects glaucoma suspects\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003eClass IIa (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eNR (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eTeleMedC DR grader\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eTeleMedC PTE LTD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eAustralia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eScreens for DR, glaucoma, and AMD *indications vary according to jurisdictions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003eClass IIa (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eEyetelligence system (Assure Plus)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eEyetelligence Pty Ltd; Optain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eAustralia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eScreens for referable eye diseases including DR, glaucoma, and AMD. Additionally, screens for CVD risks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003eClass I (2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eEyetelligence system (Assure)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eEyetelligence Pty Ltd; Optain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eAustralia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eScreens for referable eye diseases including DR, glaucoma, and AMD.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003eClass I (2019)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eDiabetic Retinopathy Screening (DRISTi)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eArtificial Learning Systems India Private Limited (Artelus)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eIndia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eScreens for the absence or presence of DR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eVUNO Med - Fundus AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eVUNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eSouth Korea\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eIdentifies and locates the presence of 12 retinal abnormalities to support the diagnosis of retinal diseases\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2020)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass IIa (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eBioAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eToku Eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eNew Zealand\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetermines biological age to give an indication of overall health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eCLAiR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eToku Eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eNew Zealand\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eCardiovascular risk assessment\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003eBreakthrough device designation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (2024)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003eClass I (NR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eiGradingM or AutoGrader\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eMedalytix Group Ltd \u0026rarr; National Services Scotland\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eUK\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects absence or presence of DR (or whether image is ungradable)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class I (2012)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eSupport vector machine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eAltris AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eAltris\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eUSA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eOCT macula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eOCT segmentation of retinal layers and biomarkers; detects retinal biomarkers and pathologies; measures segmentation volume and area; enables progression analysis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2019)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eOphthal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003emr-doc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eItaly\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eOCT macula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eOCT segmentation of retinal layers and biomarkers to aid the monitoring of patients with DMO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 93px;\"\u003e\n \u003cp\u003eUMI DR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 84px;\"\u003e\n \u003cp\u003eULMA Medical Technologies\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 73px;\"\u003e\n \u003cp\u003eSpain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003eFundus photograph\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 159px;\"\u003e\n \u003cp\u003eDetects absence or presence of DR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003eCE Class IIa (2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 71px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 65px;\"\u003e\n \u003cp\u003eNR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTable 2: Clinical evidence for each AIaMD available in the peer-reviewed literature from our searches.\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAIaMD, artificial intelligence as a medical device\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"628\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAIaMD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eManufacturer\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHighest Level of Evidence \u003csup\u003e^ #\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHead-to-head comparison (against other AIaMDs)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExternal validation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePost-market evidence\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eLumineticsCore (US) or IDx-DR (EU)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eDigital Diagnostics Inc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRCT, Prospective interventional\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMultiple countries and settings\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyeart\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyenuk, Inc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective interventional\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMultiple countries and settings\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetmarker Screening or DAIRET (in Italy)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetmarker (part of METEDA S.r.l.)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003ePost-deployment\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eThree countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetmarkerDR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetmarker (part of METEDA S.r.l.)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eSELENA+\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eeyRIS Pte. Ltd.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective silent\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eFour countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eAutomated Disease Assessment (ARDA)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eVerily Life Sciences\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective interventional\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eFour countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eMedios AI (or Medios DR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRemidio Innovative Solutions Pvt. Ltd.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective silent\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTwo countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eOphtAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEvolucare/ ADCIS (partnership)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetCAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eThirona Retina B.V.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eFour countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eDeepDee\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eDeepDee\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eMONA DR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eMONA.health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eMONA GLC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eMONA.health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMultiple countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetinalyze\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetinaLyze System A/S (Ltd.)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTwo countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eAEYE-DS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eAEYE Health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyeCheckup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eTELEKOM\u0026Uuml;NİKASYON SAN. TİC. A.Ş\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eReti-Eye\u003cbr\u003e\u0026nbsp;Reti-CVD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eMedi Whale Inc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational (Reti-Eye); Retrospective (Reti-CVD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eFour countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eITOS Mass Screening\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eVoigtmann GmbH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyeWisdom MCS/ Nexy AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eVisionary Intelligence Ltd.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyeWisdom DSS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eVisionary Intelligence Ltd.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetInSight Fluid Monitor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetInSight GmbH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMultiple countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetInSight GA Monitor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetInSight GmbH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetinAI Layer Segmentation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eIkerian (formerly RetinAI Medical AG)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetinAI Fluid Segmentation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eIkerian (formerly RetinAI Medical AG)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetinAI Macula Biomarkers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eIkerian (formerly RetinAI Medical AG)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eiPredict System\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eiHealthScreen Inc; Arif Systems\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eThree countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eTeleMedC DR grader\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eTeleMedC PTE LTD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eThree countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyetelligence system (Assure Plus)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyetelligence Pty Ltd; Optain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eThree countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyetelligence system (Assure)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eEyetelligence Pty Ltd; Optain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eProspective observational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eDiabetic Retinopathy Screening (DRISTi)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eArtificial Learning Systems India Private Limited (Artelus)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eVUNO Med - Fundus AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eVUNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eFour countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eBioAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eToku Eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eCLAiR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eToku Eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRetrospective\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSingle country\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eiGradingM or AutoGrader\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eMedalytix Group Ltd \u0026rarr; National Services Scotland\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003ePost-deployment\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTwo countries\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eAltris AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eAltris\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eOphthal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003emr-doc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eUMI DR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eULMA Medical Technologies\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNot available\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eNote: This table presents the best available evidence identified from our search of the peer-reviewed literature in July 2024. The availability and level of evidence are presented, but the quality and methodological rigour of this evidence is not assessed (out of scope).\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e^\u0026nbsp;\u003c/sup\u003e\u0026lsquo;We have used the following definitions for study types:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eProspective observational study:\u003c/strong\u003e Clinical data are collected prospectively, which allows for subsequent retrospective evaluation of AIaMD performance on prospectively gathered data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eProspective silent trial (also known as shadow, translational trial):\u003c/strong\u003e The AIaMD is run in real time on live data, but its predictions are not visible to clinicians and do not affect patient care. The goal is to assess how the model performs in the target clinical environment, simulating deployment without clinical impact.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eProspective interventional study:\u003c/strong\u003e The AIaMD is prospectively deployed with outputs shown to clinicians, who may incorporate them into care decisions. This design evaluates how the AI affects clinical workflows, behaviour, and potentially patient outcomes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRandomized controlled trial (RCT): a type of prospective interventional study wherein p\u003c/strong\u003eatients, clinicians, or clinical episodes are randomized to either an AI-assisted arm (where AI output informs care) or a control arm (usual care) to evaluate causal impact.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePost-deployment monitoring:\u003c/strong\u003e Ongoing surveillance after regulatory approval and integration of an AIaMD into routine clinical practice.\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e#\u003c/sup\u003e Distinguishing \u0026lsquo;silent\u0026rsquo; trials with certainty in this cohort was not always possible due to the ambiguous descriptions of study methodology in several instances. In such cases, we have inferred the methodology from the available evidence provided, and have adopted a conservative approach in doing so.\u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6026482/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6026482/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e This scoping review aims to identify regulator-approved ophthalmic image analysis AIaMDs in three jurisdictions, examine their characteristics and regulatory approvals, and evaluate the available evidence underpinning them, as a step towards identifying best practice and areas for improvement. 36 AIaMDs from 28 manufacturers were identified \u0026minus;\u0026thinsp;97% (35/36) approved in the EU, 22% (8/36) in Australia, and 8% (3/36) in the USA. Most targeted diabetic retinopathy detection. 19% (7/36) did not have published evidence describing performance. For the remainder, 131 clinical evaluation studies (range 1\u0026ndash;22/AIaMD) describing 192 datasets/cohorts were identified. Demographics were poorly reported (age recorded in 52%, sex 51%, ethnicity 21%). On a study-level, few included head-to-head comparisons against other AIaMDs (8%,10/131) or humans (22%, 29/131), and 37% (49/131) were conducted independently of the manufacturer. Only 11 studies (8%) were interventional. There is scope for expanding AIaMD applications to other ophthalmic imaging modalities, conditions, and use cases. Facilitating greater transparency from manufacturers, better dataset reporting, validation across diverse populations, and high-quality interventional studies with implementation-focused outcomes are key steps towards building user confidence and supporting clinical integration.\u003c/p\u003e","manuscriptTitle":"Artificial intelligence as a medical device for ophthalmic image analysis: a scoping review of regulated devices","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-14 23:09:39","doi":"10.21203/rs.3.rs-6026482/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-04-16T12:05:43+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-14T04:15:00+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-12T16:57:12+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"208902403057158020702441183584459488364","date":"2025-04-06T23:31:37+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"130216608584659939079072128021651814109","date":"2025-04-05T06:58:28+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-04T23:30:50+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-04T12:55:59+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Digital Medicine","date":"2025-03-26T15:27:42+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b177c4ff-1007-420c-9688-2c924860f40f","owner":[],"postedDate":"April 14th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":46750981,"name":"Health sciences/Health care"},{"id":46750982,"name":"Health sciences/Health care/Diagnosis"}],"tags":[],"updatedAt":"2025-06-02T16:01:06+00:00","versionOfRecord":{"articleIdentity":"rs-6026482","link":"https://doi.org/10.1038/s41746-025-01726-8","journal":{"identity":"npj-digital-medicine","isVorOnly":false,"title":"npj Digital Medicine"},"publishedOn":"2025-05-29 15:57:26","publishedOnDateReadable":"May 29th, 2025"},"versionCreatedAt":"2025-04-14 23:09:39","video":"","vorDoi":"10.1038/s41746-025-01726-8","vorDoiUrl":"https://doi.org/10.1038/s41746-025-01726-8","workflowStages":[]},"version":"v1","identity":"rs-6026482","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6026482","identity":"rs-6026482","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}