Artificial Intelligence for the Diagnosis of De Novo Diabetic Retinopathy: A Scoping Review of Global Evidence

Artificial Intelligence for the Diagnosis of De Novo Diabetic Retinopathy: A Scoping Review of Global Evidence | 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 for the Diagnosis of De Novo Diabetic Retinopathy: A Scoping Review of Global Evidence Maria Paula Trujillo López, Maria Alejandra Cortés González, Andrea Valentina Gómez López, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8071096/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Introduction: Diabetic retinopathy (DR) requires early diagnosis, as it is a common complication of diabetes mellitus (DM) and a leading cause of blindness worldwide. Artificial intelligence (AI) is emerging as a promising tool for detecting DR, as it can analyse large volumes of data with high accuracy (ACC). Methodology: A scoping review of the literature from 2019–2025 was conducted in four databases, and inclusion and exclusion criteria were applied for the selection of articles. Data extraction was performed, and a synthesis of the results was generated. Results Sixty-two articles were reviewed, mostly experimental studies (43%) and deep learning (DL) models (82%). AI was mainly applied to retinography (59%) and other images, such as optical coherence tomography (OCT) and optical coherence tomography angiography (OCTA) (11.3%). It demonstrated high sensitivity (SE) and specificity (SP) and highlighted benefits such as mass screening capacity. Limitations, such as image quality and variability between devices, have also been identified, but there are future opportunities to revolutionise the diagnosis of DR. Conclusion AI is emerging as an effective tool for the de novo diagnosis of DR, with retinography as the main technique and other methods as options with great potential. Its benefits include automation, greater coverage, and a reduced healthcare burden, reinforcing its diagnostic value. However, some technical and validation challenges need to be overcome before these methods can be implemented in clinical practice. Biological sciences/Computational biology and bioinformatics Health sciences/Diseases Health sciences/Health care Health sciences/Medical research Diabetic retinopathy artificial intelligence automation early detection imaging devices Figures Figure 1 Figure 2 INTRODUCTION DR is one of the most common and characteristic microvascular complications of DM [ 1 ]. It can manifest as no proliferative or proliferative DR, the latter being the most advanced and posing the greatest risk to visual health. It is estimated that up to one-third of people with DM develop some degree of DR, making it one of the most prevalent complications of this condition and one of the leading causes of blindness worldwide [ 2 ]. By 2024, its estimated global prevalence was 22.3% [ 3 ]. Additionally, it is projected that by 2045, it will affect more than 161 million people [ 2 ]. Considering that these changes are irreversible and that the burden of DR can vary depending on geographical location, lifestyle, and patients' access to healthcare systems [ 4 ], improving detection methods to provide timely management is essential. The gold standard for diagnosing DR is the capture of stereoscopic photographs of seven 30° fields, as they offer greater SE and SP than ophthalmoscopic methods do, although their performance depends on image quality and adequate interpretability [ 5 ]. Despite this, fundoscopy with pupil dilation continues to be used in some settings because of its low cost, although it has inferior performance, requires more time, does not allow clinical records to be kept, and its effectiveness depends on the examiner's experience [ 6 ]. These limitations reflect a gap in the early identification of complications arising from DM and highlight the need for innovative methods that allow for a more efficient and coordinated approach [ 7 ]. In this context, the implementation of AI in medicine has emerged as a promising tool for addressing and supporting clinical decision-making. Algorithms have been shown to achieve high levels of diagnostic performance, promoting early detection of DR and improving the quality of health services in developing countries [ 8 ]. Therefore, a scoping review of the evidence for AI in the de novo diagnosis of DR in people over 18 years of age worldwide was conducted to map the literature and provide an updated global overview. MATERIALS AND METHODS A scoping review was conducted in September 2025 following the Joanna Briggs Institute methodology and PRISMA-ScR guidelines [9,10]. This allowed for rigorous mapping of the literature. The review was guided by the following research question: What is the available evidence on the implementation of AI for the de novo diagnosis of DR in adults worldwide? Search strategy The search was conducted in four databases: PubMed, Scopus, BIREME, and Web of Science. Articles related to the research question were identified, and studies from January 2019 to September 2025 in English, Spanish, and Portuguese were included. The search terms were designed using MeSH-DeCS descriptors and related keywords. The specific search strategies for each database, as well as the inclusion and exclusion criteria applied, are detailed in Table 1. The selection of databases was based on their relevance to answering the research question. PubMed was included for its biomedical focus and broad coverage, Scopus for its interdisciplinary scope, BIREME for its access to the literature from Latin America and the Caribbean, and Web of Science for its rigorous indexing and relevant studies from high-impact international journals. Study selection All the citations were imported and managed via the Rayyan platform, an online tool designed for managing systematic reviews, for efficient study selection and data management. For article selection, duplicates were first removed. Second, three assessors (MPTL, MACG, AVGL) independently selected titles and abstracts according to the inclusion and exclusion criteria. Any discrepancies were resolved jointly by the three evaluators to reach a final agreement. Finally, the reviewers conducted an in-depth reading of the selected articles to determine the final eligibility of the studies. Data extraction Data extraction was carried out via a table previously designed by the research team. This table included the following elements: title, authors, date of publication, country or location, journal, type of study, study objective, population included or number of articles (in the case of reviews), key results and findings, type of AI used, and its application in the diagnosis of DR. Similarly, the main results and key findings were identified, as were the advantages derived from the use of AI in diagnosis and the limitations noted. On the other hand, opportunities and recommendations for the future were compiled to continue strengthening research in this area. This information is available in the Zenodo repository [11]. Data analysis and synthesis The data obtained were analysed via a narrative and descriptive synthesis, organising the information according to the application of AI in the de novo diagnosis of DR in adults according to the type of biomedical data used. The different approaches were subsequently compared quantitatively and qualitatively to identify similarities and differences in their implementation. Finally, the advantages and limitations noted in the literature were integrated, as were future opportunities for the development and validation of these technologies in different clinical contexts. RESULTS A total of 1,852 articles were identified in the initial search. Of these, 345 (18.6%) were eliminated because they were duplicates. A total of 1,507 (81.3%) studies were reviewed, of which 1,409 were excluded after review of the title and abstract because they did not meet the inclusion criteria. Ninety-eight articles were selected for full-text reading, 13 articles could not be accessed, and 23 articles did not meet the inclusion criteria. A total of 62 studies were included in this review. The selection and flow of the included articles are presented via the PRISMA model in Figure 1. Data characteristics Sixty-two articles evaluating the use of AI for the de novo diagnosis of DR in adults were included. Among these, reviews (n=5) were identified (one scoping review and four systematic reviews). Most of these studies were experimental studies (n=27) that focused mainly on the development and validation of AI algorithms or models. Cross-sectional observational studies (n=21), including retrospective designs, cohort studies, and diagnostic validations in clinical practice, were also identified. Finally, prospective or methodological studies (n=9) were recorded, with a focus on clinical validations in real-world settings, prospective cohorts, and technical developments with ophthalmic imaging. In terms of temporal distribution, the studies included were published between 2019 and 2025. In 2019, the evidence was limited (n=3). An increase was subsequently observed in 2020 (n=7), 2021 (n=6), 2022 (n=15), 2023 (n=7), 2024 (n=12), and 2025 (n=12). The geographical distribution of the studies, classified by continent, is illustrated in Figure 2. With respect to biomedical data, the most widely used tool was retinography (n=37), followed by multimodal tools (n=12). There were also studies based on OCT/OCTA (n=7). Finally, electronic devices (n=6), mainly nonmydriatic portable cameras and systems coupled to smartphones in telemedicine settings, were used. A summary of the quantitative results according to biomedical data is shown in Table 2. Types of artificial intelligence for detection Among the 62 articles reviewed, deep neural networks or DL (n=51) stood out, followed by articles based on machine learning (ML) algorithms (n=4). In addition, some articles used hybrid DL + ML models (n=7). Among the subtypes of DL applied were: Convolutional neural networks (CNN) (n=48); Transfer learning (n=18); Ensemble (n=12); Attention mechanisms (n=5); Transformer (n=2); Deep belief nets (n=2) and, finally, Recurrent neural networks (RNN) (n=1). On the other hand, the ML categories implemented were: Vector machine (n=7); Random Forest (n=5); Decision Tree (n=2); Multilayer perceptron (n=2); Naive bayes (n=1); Radial basis function classifier (n=2); and finally, Fuzzy methods (n=2). Artificial intelligence applied to retinography Thirty-seven studies were analysed, which together showed high performance, exceeding the performance of standard screening. DL models and hybrids stand out, maintaining good results even with low-quality images. Among the most relevant models are CNN systems, hybrid approaches, and methods that optimise processing times [12–14]. Several articles have evaluated validation in real clinical settings, where automated platforms and systems outperform previous algorithms [15–17]. Recent architectures with results close to those of retina specialists were also identified, as were models designed to improve the discrimination of specific lesions [18–48]. Artificial intelligence applied to multimodal models Three multimodal studies were evaluated, all of which described good diagnostic performance. Notably, an OCTA–fundus fusion model [49], an OCT-based model using morphology and reflectivity [50], and a comparison between OCT and retinography were used[51]. Nine studies comparing the use of AI for the diagnosis of DR with the criteria of trained ophthalmologists were also reviewed. The analysis of retinography on a telemedicine platform yielded consistent results [52], as did the use of conventional retinography compared with the criteria of ophthalmologists [53–56]. Finally, the use of portable cameras, compared with standard ophthalmological examinations, has also shown efficacy in both clinical settings and community hospitals [57,58]. Artificial intelligence applied to optical coherence tomography/optical coherence angiography Seven articles evaluated AI for the automatic detection of DR via OCT and OCTA. In a cohort of 664 patients, a DL-based GoogLeNet model proved useful in early stages [59]. A review of 32 studies revealed that CNN and Vision Transformers (ViTs) outperform traditional ML in terms of SE, SP, and area under the curve (AUC) [60]. In contrast, a study using logistic regression and classification trees obtained moderate performance [61], whereas a model using ViTs and data augmentation achieved ACC, SE, and SP values greater than 98% [62]. The integration of three-dimensional features (reflectivity + thickness) improved the results compared with those of SVM, KNN, naive Bayes, decision trees, and classic DL architectures such as GoogLeNet or ResNet-50 [63]. A ResNet101-based CNN externally validated with OCTA showed consistent performance [64], and a hybrid model with multifractal descriptors and multiple layers was able to detect subtle structural irregularities, offering robust and scalable classification in the early stages of DR [65]. Artificial intelligence applied to electronic devices Six studies evaluated the use of AI for DR detection based on images obtained with smartphones. One study used a smartphone with an adapted lens, obtaining 2,130 images and showing high concordance with expert ophthalmologists [66]. Another study compared smartphone images with conventional retinography images and reported comparable diagnostic performance [67]. In addition, the offline MediosAI tool, which operates on smartphones, demonstrated consistent performance when dilated fields of view were used [68] and, when applied to nonmydriatic images, proved useful in low-resource settings [69]. Finally, the use of portable cameras coupled with AI confirmed the feasibility of screening in community settings [70]. Benefits and limitations The benefits and limitations of AI for the de novo diagnosis of DR extracted from the articles evaluated are presented in Table 3. DISCUSSION Summary of results The global outlook on the use of AI for the de novo diagnosis of DR has expanded widely. Most of the studies included in this review were experimental or observational studies. There was a relatively high concentration of research in Asia, especially China and India, and low representation in Latin America and Africa. This distribution reflects an equity and access gap, highlighting the need to encourage research in these regions to ensure the validity and global applicability of the evidence. Additionally, there has been a surge in research since 2020, reflecting a sustained increase in interest in exploring this innovative tool. Regarding the type of biomedical data associated with AI, the most reported modality is retinography, followed by multimodal approaches, OCT/OCTA, and electronic devices. The most used type of AI was DL, mainly CNNs. The results confirm the high performance of AI using retinography, with an average ACC of 95.3%, SE of 92%, SP of 94% and an AUC close to 0.95. These results exceed the performance of conventional screening and remain high even in models validated in clinical practice. This finding indicates that retinography was the most robust tool alongside AI in the diagnosis of DR, although its effectiveness will depend on image quality and multicentre validations to ensure its generalisation. On the other hand, multimodal studies achieved SE ranges of 94–96% and SPs ranges of 94–99%%, indicating greater reliability when different imaging techniques are combined. Likewise, in direct comparisons with ophthalmologists, SE ranges of 79–100% and SPs of 86–100% were obtained, with better performance in retinography and higher-quality cameras. These data indicate high diagnostic performance that can match or exceed the TAs of specialists, although the TP depends on the device and the clinical context, posing the challenge of optimising its use in real-world settings. Studies based on OCT/OCTA have shown variable AI performance, with ACC ranges of 88–99%, SE of 63–99%, SP of 72–99%, and AUC of 0.74–0.97. Classic models offered moderate results, whereas more advanced models, such as ViTs and three-dimensional fusions, achieved performance close to 99%. This indicates high potential for early detection, although multicentre validation and image standardisation are required to ensure consistent results outside experimental settings. Finally, studies of electronic devices based on smartphones and portable cameras reported SEs of 83–100% and SPs of 85–96%, with AUC close to 0.90–0.91. Models such as MediosAI showed the best results. This evidence confirms the high potential for community screening given its accessibility and usefulness in low-resource settings. In conclusion, evidence shows that AI applied to retinography is the most far-reaching strategy for diagnosing DR, with high ACC values and clinical validity; OCT/OCTA-based models and multimodal approaches reinforce its potential for early detection, although its performance depends on image standardisation and multicentre validations. Similarly, mobile devices confirm the feasibility of implementing screening in communities with limited resources. These findings are consistent with previous systematic reviews, which have documented robust diagnostic performance of AI, with SE and SP comparable to or even superior to those of trained clinicians, reinforcing its potential for integration into large-scale clinical practice [ 71 ]. From this perspective, AI is seen as an effective tool with diagnostic potential in various contexts, but its impact will depend on multicentre validation and an approach that minimises the risks of bias arising from the use of public databases and high-quality images that are not representative of actual clinical practice. Explainability and impact on clinical confidence in artificial intelligence In addition to diagnostic performance, a key aspect identified in the literature is the explainability of AI models. Tools such as gradient-weighted class activation mapping (Grad-CAM) allow heatmaps to be generated on retinography or OCTs, highlighting relevant lesions (microaneurysms, exudates, and haemorrhages) that support the algorithm's decision. Similarly, ViTs, in addition to achieving performance close to 99% in OCT/OCTA [ 62 ], offer self-attention mechanisms that allow visualisation of which areas of the image influence the classification. However, most of the studies included still report models with limited interpretability, which reduces professional confidence and hinders their validation in real-world scenarios. The evidence highlights that more interpretable and visually understandable models favour both medical acceptance and patient confidence and underscore the need for regulatory frameworks and clinical guidelines that define minimum standards of explainability for safe integration into DR diagnosis. Clinical relevance AI offers substantial advantages in the de novo diagnosis of DR. Its high SE and SP metrics support its reliability for screening by detecting early microvascular lesions that may go unnoticed in conventional assessments [ 22 , 28 , 51 , 60 , 66 ]. Automation streamlines the analysis of large volumes of data, reduces the workload of ophthalmologists, and facilitates early detection [ 21 , 23 , 26 , 31 , 56 , 57 ]. In addition, the availability of portable, low-cost systems facilitates screening in primary, rural, or even home settings by nonspecialised personnel, expanding coverage and improving equity in access to care [ 25 , 52 , 58 , 69 , 70 , 72 ]. These characteristics reinforce the relevance of AI in optimising early detection, reducing underdiagnosis, and strengthening the sustainability of visual health programmes [ 45 , 48 , 52 , 56 , 59 , 70 , 73 ]. Future opportunities. Future opportunities highlight the need for external validation in diverse, multicentre cohorts, which is key to ensuring the generalisation of models [ 22 , 26 , 37 , 49 , 61 , 69 ][ 22 , 23 , 26 , 30 , 38 , 53 , 67 ]. The Another priority is the optimisation of models to improve robustness against noise and artefacts and adjust diagnostic thresholds [ 17 , 29 , 33 , 47 , 63 ]. The multimodal fusion of fundus images, OCT/OCTA, and clinical data is emerging as a strategy to increase diagnostic ACC [ 32 , 50 , 57 , 60 , 62 ]. In addition, emphasis is placed on advancing explainability and interpretability through techniques such as heatmaps or SHAP/LIME, which promote clinical confidence and acceptance by patients and professionals [ 32 , 46 , 47 , 51 ], and exploring transfer learning and advanced architectures to strengthen generalisation and standardisation protocols between datasets [ 16 , 25 , 42 , 47 , 57 , 74 ]. Ethical considerations This review used only published information, so it did not require ethical approval or informed consent. However, the studies analysed involve the handling of clinical and biomedical data, which require strict confidentiality, data protection, and governance frameworks to regulate ownership, access, and responsible use. According to the Ethics and governance of artificial intelligence for health: WHO guidance [ 75 ], it is essential to protect privacy through anonymization and transparent governance, with regulatory frameworks that allow auditing and accountability. It is also important to ensure equity in the implementation of these technologies, avoiding the exclusion of vulnerable populations or regions with less digital infrastructure, which can perpetuate health gaps. The WHO highlights the need for transparency and explainability in algorithms so that their decisions are auditable and understandable by health professionals [ 76 ]. Furthermore, under the principles of nonmaleficence and shared responsibility, maintaining human oversight in clinical interpretation and establishing monitoring mechanisms to identify biases and prevent potential harm are recommended, promoting effective and socially responsible adoption of AI. Limitations of the article Despite the promising results of AI for the de novo diagnosis of DR, this study has several limitations. First, much of the evidence collected comes from public databases or specific populations, which leads to heterogeneity and lower external validation. In addition, there was a high dependence on optimal high-quality images obtained in controlled environments, with variability in the devices used, which limits generalisation to real clinical contexts. There is a lack of longitudinal and multicentre studies in diverse populations, which restricts the evaluation of the performance of the models needed in routine practice. Additional methodological limitations, such as publication bias, should also be considered. Finally, the rapid evolution of AI techniques requires periodic updates of the evidence and continuous evaluations to ensure that the models remain valid and relevant in real clinical settings. CONCLUSION The application of AI in the de novo diagnosis of DR is establishing itself as an effective and versatile tool, with retinography standing out as the most robust technique and multimodal approaches, OCT/OCTA, and electronic devices as alternatives with great potential for early detection and screening, even in communities with limited resources. Its benefits include automation, expanded coverage, and a reduced healthcare burden, reinforcing its value as an innovative diagnostic support tool. However, its implementation faces challenges related to image quality, protocol standardisation, the need for multicentre validation, and improved interpretability. In this context, the evidence positions AI as a promising complement to clinical practice, whose integration will depend on overcoming current limitations and consolidating its application in a safe and generalisable manner through multicentre, longitudinal studies with greater methodological rigour. Abbreviations DR- Diabetic retinopathy DM- Diabetes mellitus AI - Artificial intelligence DL - Deep learning OCT- Optical coherence tomography OCTA - Optical coherence tomography angiography SE - Sensitivity SP- Specificity ML - Machine learning CNN - Convolutional neural network RNN - Recurrent neural networks ACC - Accuracy AUC - Area under the curve ViTs - Vision Transformers Declarations AUTHORS' CONTRIBUTIONS MPTL: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Writing – original draft. MACG: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Visualization, Writing – original draft. AVGL: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Visualization, Writing – original draft. EHHR: Funding acquisition, Methodology, Project administration, Supervision, Validation, Writing – review & editing. Ethical approval: This study was approved by the Ethics and Research Committee of the Faculty of Medicine of the University of La Sabana. Data availability: The datasets analysed during the current study are available in the Zenodo repository [11]. Conflicts of interest: All authors declare no financial or non-financial competing interests. Funding: Research derived from the MED-341-2023 project, developed by the Family Medicine and Population Health Research Group at Universidad de La Sabana, Colombia. Acknowledgements: Not applicable. References Shah J, Cheong ZY, Tan B, Wong D, Liu X, Chua J. Dietary Intake and Diabetic Retinopathy: A Systematic Review of the Literature. Nutrients [Internet]. MDPI; 2022 [cited 2025 Sep 7];14:5021. https://doi.org/10.3390/NU14235021/S1 Seo H, Park SJ, Song M. Diabetic Retinopathy (DR): Mechanisms, Current Therapies, and Emerging Strategies. Cells [Internet]. Multidisciplinary Digital Publishing Institute (MDPI); 2025 [cited 2025 Sep 7];14. https://doi.org/10.3390/CELLS14050376, Bravo González R, Cuartas Mesa JJ, Garzón Duque MO, Bravo Acosta JD, Vásquez MCF, Escobar Howard M, et al. Factores relacionados con la prevalencia de retinopatía diabética. Revista Med [Internet]. Universidad Militar Nueva Granada; 2024 [cited 2025 Sep 7];32:73–91. https://doi.org/10.18359/RMED.7374 Diabetic Retinopathy - Latin America - American Academy of Ophthalmology [Internet]. [cited 2025 Sep 7]. https://www.aao.org/education/topic-detail/diabetic-retinopathy-latin-america?utm. Accessed 7 Sep 2025 Kumari S, Venkatesh P, Tandon N, Chawla R, Takkar B, Kumar A. Selfie fundus imaging for diabetic retinopathy screening. Eye [Internet]. Springer Nature; 2021 [cited 2025 Sep 20];36:1988. https://doi.org/10.1038/S41433-021-01804-7 Yuribia Karina Millán-Gámez 2Niels H. Wacher-Rodarte,3Juan Carlos Bravo-Ortiz, 4Néstor Hugo Garrido-Gaspar,5Rosaura Mendoza-Topete, Adriana Marcela Martínez-Ruiz,Guillermo Miller-Arrevillaga, Alfonso Ramírez-Nájera. Diabetic retinopathy screening in the emerging era of artificial intelligence. Diabetologia [Internet]. Springer Science and Business Media Deutschland GmbH; 2022 [cited 2025 Sep 20];65:1415–23. https://doi.org/10.1007/S00125-022-05727-0/FIGURES/3 Gunasekeran D V., Ting DSW, Tan GSW, Wong TY. Artificial intelligence for diabetic retinopathy screening, prediction and management. Curr Opin Ophthalmol [Internet]. Lippincott Williams and Wilkins; 2020 [cited 2025 Sep 7];31:357–65. https://doi.org/10.1097/ICU.0000000000000693 Castrillón-Lozano JA, Arango-Cárdenas D, Marulanda-Márquez DE. Inteligencia artificial y tamizaje de la discapacidad visual relacionada con retinopatía diabética y edema macular. Gac Med Mex. 2024;160:217–8. https://doi.org/10.24875/GMM.M24000881 Aromataris E et al. JBI Manual for Evidence Synthesis. Aromataris E, Lockwood C, Porritt K, Pilla B, Jordan Z, editors. JBI; 2024. https://doi.org/10.46658/JBIMES-24-01 Tricco AC, Lillie E, Zarin W, O’Brien KK, Colquhoun H, Levac D, et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med. 2018;169:467–73. https://doi.org/10.7326/M18-0850 Trujillo Lopez MP. Dataset: Artificial Intelligence for the Diagnosis of De Novo Diabetic Retinopathy A Scoping Review of Global Evidence. Zenodo; 2025. https://doi.org/10.5281/zenodo.17600994 Baget-Bernaldiz M, Fontoba-Poveda B, Romero-Aroca P, Navarro-Gil R, Hernando-Comerma A, Bautista-Perez A, et al. Artificial Intelligence-Based Screening System for Diabetic Retinopathy in Primary Care. Diagnostics 2024, Vol 14, Page 1992 [Internet]. Multidisciplinary Digital Publishing Institute; 2024 [cited 2025 Oct 4];14:1992. https://doi.org/10.3390/DIAGNOSTICS14171992 Naz H, Nijhawan R, Ahuja NJ. An automated unsupervised deep learning–based approach for diabetic retinopathy detection. Med Biol Eng Comput [Internet]. Springer Science and Business Media Deutschland GmbH; 2022 [cited 2025 Oct 4];60:3635–54. https://doi.org/10.1007/S11517-022-02688-9/FIGURES/17 Omer HK. Diabetic retinopathy detection using Bilayered Neural Network classification model with resubstitution validation. MethodsX [Internet]. Elsevier; 2024 [cited 2025 Oct 4];12:102705. https://doi.org/10.1016/J.MEX.2024.102705 Abreu-Gonzalez R, Susanna-González G, Blair JPM, Vitar RML, Ciller C, Apostolopoulos S, et al. Validation of artificial intelligence algorithm LuxIA for screening of diabetic retinopathy from a single 45° retinal colour fundus images: the CARDS study. BMJ Open Ophthalmol [Internet]. BMJ Publishing Group Ltd; 2025 [cited 2025 Oct 4];10:2109. https://doi.org/10.1136/BMJOPHTH-2024-002109 Hu CL, Wang YC, Wu WF, Xi Y. Evaluation of AI-enhanced non-mydriatic fundus photography for diabetic retinopathy screening. Photodiagnosis Photodyn Ther [Internet]. Elsevier; 2024 [cited 2025 Oct 4];49:104331. https://doi.org/10.1016/J.PDPDT.2024.104331 Antaki F, Hammana I, Tessier M-C, Boucher A, David Jetté ML, Beauchemin C, et al. Implementation of Artificial Intelligence-Based Diabetic Retinopathy Screening in a Tertiary Care Hospital in Quebec: Prospective Validation Study. JMIR Diabetes [Internet]. JMIR Diabetes; 2024 [cited 2025 Oct 4];9:e59867. https://doi.org/10.2196/59867 Naveen K V., Anoop BN, Siju KS, Kar MK, Venugopal V. EffNet-SVM: A Hybrid Model for Diabetic Retinopathy Classification Using Retinal Fundus Images. IEEE Access. Institute of Electrical and Electronics Engineers Inc.; 2025;13:79793–804. https://doi.org/10.1109/ACCESS.2025.3566073 Zhang G, Sun B, Chen Z, Gao Y, Zhang Z, Li K, et al. Diabetic Retinopathy Grading by Deep Graph Correlation Network on Retinal Images Without Manual Annotations. Front Med (Lausanne) [Internet]. Frontiers Media S.A.; 2022 [cited 2025 Oct 4];9:872214. https://doi.org/10.3389/FMED.2022.872214/BIBTEX Muthusamy D, Palani P. Deep neural network model for diagnosing diabetic retinopathy detection: An efficient mechanism for diabetic management. Biomed Signal Process Control [Internet]. Elsevier; 2025 [cited 2025 Oct 4];100:107035. https://doi.org/10.1016/J.BSPC.2024.107035 Butt MM, Iskandar DNFA, Abdelhamid SE, Latif G, Alghazo R. Diabetic Retinopathy Detection from Fundus Images of the Eye Using Hybrid Deep Learning Features. Diagnostics 2022, Vol 12, Page 1607 [Internet]. Multidisciplinary Digital Publishing Institute; 2022 [cited 2025 Oct 4];12:1607. https://doi.org/10.3390/DIAGNOSTICS12071607 Tavakoli M, Mehdizadeh A, Aghayan A, Shahri RP, Ellis T, Dehmeshki J. Automated Microaneurysms Detection in Retinal Images Using Radon Transform and Supervised Learning: Application to Mass Screening of Diabetic Retinopathy. IEEE Access [Internet]. Institute of Electrical and Electronics Engineers Inc.; 2021 [cited 2025 Oct 4];9:67302–14. https://doi.org/10.1109/ACCESS.2021.3074458 Thomas GAS, Robinson YH, Julie EG, Shanmuganathan V, Rho S, Nam Y. Intelligent Prediction Approach for Diabetic Retinopathy Using Deep Learning Based Convolutional Neural Networks Algorithm by Means of Retina Photographs. Computers, Materials & Continua [Internet]. Tech Science Press; 2020 [cited 2025 Oct 4];66:1613–29. https://doi.org/10.32604/CMC.2020.013443 Mutawa AM, Hemalakshmi GR, Prakash NB, Murugappan M. Randomization-Driven Hybrid Deep Learning for Diabetic Retinopathy Detection. IEEE Access [Internet]. Institute of Electrical and Electronics Engineers Inc.; 2025 [cited 2025 Oct 4];13:38901–13. https://doi.org/10.1109/ACCESS.2025.3546359 Modi P, Kumar Y. An Effective Diabetic Retinopathy Detection System using Deep Belief Nets and Adaptive Learning in Cloud Environment. Scalable Computing: Practice and Experience [Internet]. Scalable Computing: Practice and Experience; 2023 [cited 2025 Oct 4];24:97–114. https://doi.org/10.12694/SCPE.V24I2.2117 Tran J, Estevez JJ, Howard NJ, Kumar S. Barriers and Enablers Influencing the Implementation of Artificial Intelligence for Diabetic Retinopathy Screening in Clinical Practice: A Scoping Review. Clin Exp Ophthalmol [Internet]. John Wiley & Sons, Ltd; 2025 [cited 2025 Oct 4];0:1–12. https://doi.org/10.1111/CEO.14567 Wang Y, Han X, Li C, Luo L, Yin Q, Zhang J, et al. Impact of Gold-Standard Label Errors on Evaluating Performance of Deep Learning Models in Diabetic Retinopathy Screening: Nationwide Real-World Validation Study. J Med Internet Res [Internet]. JMIR Publications Inc.; 2024 [cited 2025 Oct 4];26:e52506. https://doi.org/10.2196/52506 Selvin Charles A, Akila C. Enhanced detection of diabetic retinopathy using machine learning based feature selection and ensemble classifiers. AIP Adv [Internet]. AIP Publishing; 2025 [cited 2025 Oct 4];15:75029. https://doi.org/10.1063/5.0279598 Romero-Oraá R, García M, Oraá-Pérez J, López-Gálvez MI, Hornero R. Effective Fundus Image Decomposition for the Detection of Red Lesions and Hard Exudates to Aid in the Diagnosis of Diabetic Retinopathy. Sensors 2020, Vol 20, Page 6549 [Internet]. Multidisciplinary Digital Publishing Institute; 2020 [cited 2025 Oct 4];20:6549. https://doi.org/10.3390/S20226549 KHAN ZAID, GAIDHANE AM, SINGH M, GANESAN S, KAUR M, SHARMA GC, et al. Diagnostic Accuracy of IDX-DR for Detecting Diabetic Retinopathy: A Systematic Review and Meta-Analysis. Am J Ophthalmol [Internet]. Elsevier Inc.; 2025 [cited 2025 Oct 4];273:192–204. https://doi.org/10.1016/j.ajo.2025.02.022 JOSEPH S, SELVARAJ J, MANI I, KUMARAGURUPARI T, SHANG X, MUDGIL P, et al. Diagnostic Accuracy of Artificial Intelligence-Based Automated Diabetic Retinopathy Screening in Real-World Settings: A Systematic Review and Meta-Analysis. Am J Ophthalmol [Internet]. Elsevier Inc.; 2024 [cited 2025 Oct 4];263:214–30. https://doi.org/10.1016/j.ajo.2024.02.012 Mohammedhasan M, Uğuz H. A New Early Stage Diabetic Retinopathy Diagnosis Model Using Deep Convolutional Neural Networks and Principal Component Analysis. [cited 2025 Oct 4]; https://doi.org/10.18280/ts.370503 Tan-Torres A, Praveen PA, Jeji D, Brant A, Yin X, Yang L, et al. Validation of a Deep Learning Model for Diabetic Retinopathy on Patients with Young-Onset Diabetes. Ophthalmol Ther [Internet]. Adis; 2025 [cited 2025 Oct 4];14:1147–55. https://doi.org/10.1007/S40123-025-01116-Z/FIGURES/1 Zhang G-H, Zhuo G-P, Zhang Z-X, Sun B, Yang W-H, Zhang S-C, et al. Diabetic retinopathy identification based on multi-source-free domain adaptation. Int J Ophthalmol [Internet]. International Journal of Ophthalmology (c/o Editorial Office); 2024 [cited 2025 Oct 4];17:1193–204. https://doi.org/10.18240/IJO.2024.07.03 Hao S, Liu C, Li N, Wu Y, Li D, Gao Q, et al. Clinical evaluation of AI-assisted screening for diabetic retinopathy in rural areas of midwest China. PLoS One [Internet]. Public Library of Science; 2022 [cited 2025 Oct 4];17:e0275983. https://doi.org/10.1371/JOURNAL.PONE.0275983 Zhang Y, Shi J, Peng Y, Zhao Z, Zheng Q, Wang Z, et al. Artificial intelligence-enabled screening for diabetic retinopathy: a real-world, multicenter and prospective study. BMJ Open Diabetes Res Care [Internet]. American Diabetes Association; 2020 [cited 2025 Oct 4];8:1596. https://doi.org/10.1136/BMJDRC-2020-001596 Bellemo V, Lim ZW, Lim G, Nguyen QD, Xie Y, Yip MYT, et al. Artificial intelligence using deep learning to screen for referable and vision-threatening diabetic retinopathy in Africa: a clinical validation study. Lancet Digit Health [Internet]. Elsevier Ltd; 2019 [cited 2025 Oct 4];1:e35–44. https://doi.org/10.1016/S2589-7500(19)30004-4 Karakaya M, Aygun RS, Sallam AB. Collaborative Deep Learning for Privacy Preserving Diabetic Retinopathy Detection. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS. Institute of Electrical and Electronics Engineers Inc.; 2022;2022-July:2181–4. https://doi.org/10.1109/EMBC48229.2022.9871617 Aquino-Brítez D, Gómez JA, Noguera JLV, García-Torres M, Román JCM, Gardel-Sotomayor PE, et al. Automatic Diagnosis of Diabetic Retinopathy from Fundus Images Using Neuro-Evolutionary Algorithms. Stud Health Technol Inform [Internet]. IOS Press; 2022 [cited 2025 Oct 4];290:689–93. https://doi.org/10.3233/SHTI220166 Ruamviboonsuk P, Tiwari R, Sayres R, Nganthavee V, Hemarat K, Kongprayoon A, et al. Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study. Lancet Digit Health [Internet]. Elsevier Ltd; 2022 [cited 2025 Oct 4];4:e235–44. https://doi.org/10.1016/S2589-7500(22)00017-6 Wang J, Jiao W, Zheng Y, Ullah MZ, Song J. Multicolor image classification using the multimodal information bottleneck network (MMIB-Net) for detecting diabetic retinopathy. Optics Express, Vol 29, Issue 14, pp 22732-22748 [Internet]. Optica Publishing Group; 2021 [cited 2025 Oct 4];29:22732–48. https://doi.org/10.1364/OE.430508 Vives-Boix V, Ruiz-Fernández D. Diabetic retinopathy detection through convolutional neural networks with synaptic metaplasticity. Comput Methods Programs Biomed [Internet]. Elsevier; 2021 [cited 2025 Oct 4];206:106094. https://doi.org/10.1016/J.CMPB.2021.106094 Palaniswamy T, Vellingiri M. Internet of Things and Deep Learning Enabled Diabetic Retinopathy Diagnosis Using Retinal Fundus Images. IEEE Access. Institute of Electrical and Electronics Engineers Inc.; 2023;11:27590–601. https://doi.org/10.1109/ACCESS.2023.3257988 Mazar Pasha LT, Rajashekar JS. Image Based Edge Weighted Linked Segmentation Model Using Deep Learning for Detection of Diabetic Retinopathy. Traitement du Signal [Internet]. International Information and Engineering Technology Association; 2022 [cited 2025 Oct 4];39:165–72. https://doi.org/10.18280/TS.390116 Pareja-Ríos A, Ceruso S, Romero-Aroca P, Bonaque-González S. A New Deep Learning Algorithm with Activation Mapping for Diabetic Retinopathy: Backtesting after 10 Years of Tele-Ophthalmology. Journal of Clinical Medicine 2022, Vol 11, Page 4945 [Internet]. Multidisciplinary Digital Publishing Institute; 2022 [cited 2025 Oct 4];11:4945. https://doi.org/10.3390/JCM11174945 Jabbar A, Liaqat H Bin, Akram A, Sana MU, Azpiroz ID, Diez IDLT, et al. A Lesion-Based Diabetic Retinopathy Detection Through Hybrid Deep Learning Model. IEEE Access. Institute of Electrical and Electronics Engineers Inc.; 2024;12:40019–36. https://doi.org/10.1109/ACCESS.2024.3373467 Luo X, Pu Z, Xu Y, Wong WK, Su J, Dou X, et al. MVDRNet: Multi-view diabetic retinopathy detection by combining DCNNs and attention mechanisms. Pattern Recognit [Internet]. Pergamon; 2021 [cited 2025 Oct 4];120:108104. https://doi.org/10.1016/J.PATCOG.2021.108104 Butt MM, Iskandar DNFA, Abdelhamid SE, Latif G, Alghazo R. Diabetic Retinopathy Detection from Fundus Images of the Eye Using Hybrid Deep Learning Features. Diagnostics 2022, Vol 12, Page 1607 [Internet]. Multidisciplinary Digital Publishing Institute; 2022 [cited 2025 Oct 4];12:1607. https://doi.org/10.3390/DIAGNOSTICS12071607 Bidwai P, Gite S, Pahuja N, Pahuja K, Kotecha K, Jain N, et al. Multimodal image fusion for the detection of diabetic retinopathy using optimized explainable AI-based Light GBM classifier. Information Fusion [Internet]. Elsevier B.V.; 2024 [cited 2025 Oct 4];111. https://doi.org/10.1016/j.inffus.2024.102526 Sharafeldeen A, Elsharkawy M, Khalifa F, Soliman A, Ghazal M, AlHalabi M, et al. Precise higher-order reflectivity and morphology models for early diagnosis of diabetic retinopathy using OCT images. Sci Rep [Internet]. Nature Research; 2021 [cited 2025 Oct 4];11:1–16. https://doi.org/10.1038/S41598-021-83735-7;SUBJMETA Bi Z, Li J, Liu Q, Fang Z. Deep learning-based optical coherence tomography and retinal images for detection of diabetic retinopathy: a systematic and meta analysis. Front Endocrinol (Lausanne) [Internet]. Frontiers Media SA; 2025 [cited 2025 Oct 4];16:1485311. https://doi.org/10.3389/FENDO.2025.1485311/BIBTEX Boyle J, Vignarajan J, Saha S. Automated Diabetic Retinopathy Diagnosis for Improved Clinical Decision Support. Stud Health Technol Inform [Internet]. IOS Press; 2024 [cited 2025 Oct 4];310:1490–1. https://doi.org/10.3233/SHTI231259 Ibañez-Bruron MC, Cruzat A, Ordenes-Caviere G, Coria M, Ibañez-Bruron MC, Cruzat A, et al. Use of Artificial Intelligence in Diabetic Retinopathy Screening: Experience in a Health Service in Santiago, Chile. Rev Med Chil [Internet]. Sociedad Médica de Santiago; 2024 [cited 2025 Oct 4];152:1148–53. https://doi.org/10.4067/S0034-98872024001101148 Riotto E, Gasser S, Potic J, Sherif M, Stappler T, Schlingemann R, et al. Accuracy of Autonomous Artificial Intelligence-Based Diabetic Retinopathy Screening in Real-Life Clinical Practice. Journal of Clinical Medicine 2024, Vol 13, Page 4776 [Internet]. Multidisciplinary Digital Publishing Institute; 2024 [cited 2025 Oct 4];13:4776. https://doi.org/10.3390/JCM13164776 He J, Cao T, Xu F, Wang S, Tao H, Wu T, et al. Artificial intelligence-based screening for diabetic retinopathy at community hospital. Eye (Basingstoke) [Internet]. Springer Nature; 2020 [cited 2025 Oct 4];34:572–6. https://doi.org/10.1038/S41433-019-0562-4;SUBJMETA Lim JI, Regillo CD, Sadda SVR, Ipp E, Bhaskaranand M, Ramachandra C, et al. Artificial Intelligence Detection of Diabetic Retinopathy: Subgroup Comparison of the EyeArt System with Ophthalmologists’ Dilated Examinations. Ophthalmology Science [Internet]. Elsevier Inc.; 2023 [cited 2025 Oct 4];3. https://doi.org/10.1016/j.xops.2022.100228 Dong X, Du S, Zheng W, Cai C, Liu H, Zou J. Evaluation of an Artificial Intelligence System for the Detection of Diabetic Retinopathy in Chinese Community Healthcare Centers. Front Med (Lausanne) [Internet]. Frontiers Media S.A.; 2022 [cited 2025 Oct 4];9:883462. https://doi.org/10.3389/FMED.2022.883462/BIBTEX Tomić M, Vrabec R, Hendelja Đ, Kolarić V, Bulum T, Rahelić D. Diagnostic Accuracy of Hand-Held Fundus Camera and Artificial Intelligence in Diabetic Retinopathy Screening. Biomedicines 2024, Vol 12, Page 34 [Internet]. Multidisciplinary Digital Publishing Institute; 2023 [cited 2025 Oct 4];12:34. https://doi.org/10.3390/BIOMEDICINES12010034 Singh R, Singuri S, Batoki J, Lin K, Luo S, Hatipoglu D, et al. Deep Learning Algorithm Detects Presence of Disorganization of Retinal Inner Layers (DRIL)–An Early Imaging Biomarker in Diabetic Retinopathy. Transl Vis Sci Technol. Association for Research in Vision and Ophthalmology Inc.; 2023;12. https://doi.org/10.1167/tvst.12.7.6 Hayati A, Abdol Homayuni MR, Sadeghi R, Asadigandomani H, Dashtkoohi M, Eslami S, et al. Advancing Diabetic Retinopathy Screening: A Systematic Review of Artificial Intelligence and Optical Coherence Tomography Angiography Innovations. Diagnostics [Internet]. Multidisciplinary Digital Publishing Institute (MDPI); 2025 [cited 2025 Oct 4];15:737. https://doi.org/10.3390/DIAGNOSTICS15060737/S1 Yao H, Wu S, Zhan Z, Li Z. A Classification Tree Model with Optical Coherence Tomography Angiography Variables to Screen Early-Stage Diabetic Retinopathy in Diabetic Patients. J Ophthalmol [Internet]. John Wiley & Sons, Ltd; 2022 [cited 2025 Oct 4];2022:9681034. https://doi.org/10.1155/2022/9681034 Zhou Z, Yu H, Zhao J, Wang X, Wu Q, Dai C. Automatic diagnosis of diabetic retinopathy using vision transformer based on wide-field optical coherence tomography angiography. https://doi.org/101142/S1793545823500190 [Internet]. World Scientific Publishing Company ; 2023 [cited 2025 Oct 4];17:2350019. https://doi.org/10.1142/S1793545823500190 Elgafi M, Sharafeldeen A, Elnakib A, Elgarayhi A, Alghamdi NS, Sallah M, et al. Detection of Diabetic Retinopathy Using Extracted 3D Features from OCT Images. Sensors 2022, Vol 22, Page 7833 [Internet]. Multidisciplinary Digital Publishing Institute; 2022 [cited 2025 Oct 4];22:7833. https://doi.org/10.3390/S22207833 Ryu G, Lee K, Park D, Park SH, Sagong M. A deep learning model for identifying diabetic retinopathy using optical coherence tomography angiography. Sci Rep [Internet]. Nature Research; 2021 [cited 2025 Oct 4];11:1–9. https://doi.org/10.1038/S41598-021-02479-6;SUBJMETA Aziz A, Tezel NS, Kaçmaz S, Attallah Y. Early Diabetic Retinopathy Detection from OCT Images Using Multifractal Analysis and Multi-Layer Perceptron Classification. Diagnostics 2025, Vol 15, Page 1616 [Internet]. Multidisciplinary Digital Publishing Institute; 2025 [cited 2025 Oct 4];15:1616. https://doi.org/10.3390/DIAGNOSTICS15131616 Wroblewski JJ, Sanchez-Buenfil E, Inciarte M, Berdia J, Blake L, Wroblewski S, et al. Diabetic Retinopathy Screening Using Smartphone-Based Fundus Photography and Deep-Learning Artificial Intelligence in the Yucatan Peninsula: A Field Study. J Diabetes Sci Technol [Internet]. SAGE PublicationsSage CA: Los Angeles, CA; 2025 [cited 2025 Oct 4];19:370–6. https://doi.org/10.1177/19322968231194644 Karakaya M, Hacisoftaoglu RE. Comparison of smartphone-based retinal imaging systems for diabetic retinopathy detection using deep learning. BMC Bioinformatics [Internet]. BioMed Central; 2020 [cited 2025 Oct 4];21:1–18. https://doi.org/10.1186/S12859-020-03587-2/TABLES/3 Sosale B, Sosale A, Murthy H, Sengupta S, Naveenam M. Medios-An offline, smartphone-based artificial intelligence algorithm for the diagnosis of diabetic retinopathy. Indian J Ophthalmol [Internet]. Wolters Kluwer Medknow Publications; 2020 [cited 2025 Oct 4];68:391–5. https://doi.org/10.4103/IJO.IJO_1203_19 Sosale B, Aravind SR, Murthy H, Narayana S, Sharma U, Gowda SGV, et al. Simple, Mobile-based Artificial Intelligence Algorithm in the detection of Diabetic Retinopathy (SMART) study. BMJ Open Diabetes Res Care [Internet]. American Diabetes Association; 2020 [cited 2025 Oct 4];8:892. https://doi.org/10.1136/BMJDRC-2019-000892 Malerbi FK, Andrade RE, Morales PH, Stuchi JA, Lencione D, de Paulo JV, et al. Diabetic Retinopathy Screening Using Artificial Intelligence and Handheld Smartphone-Based Retinal Camera. J Diabetes Sci Technol [Internet]. SAGE PublicationsSage CA: Los Angeles, CA; 2022 [cited 2025 Oct 4];16:716–23. https://doi.org/10.1177/1932296820985567 Alqahtani AS, Alshareef WM, Aljadani HT, Hawsawi WO, Shaheen MH. The efficacy of artificial intelligence in diabetic retinopathy screening: a systematic review and meta-analysis. Int J Retina Vitreous [Internet]. BioMed Central Ltd; 2025 [cited 2025 Oct 4];11:1–12. https://doi.org/10.1186/S40942-025-00670-9/FIGURES/8 Natarajan S, Jain A, Krishnan R, Rogye A, Sivaprasad S. Diagnostic Accuracy of Community-Based Diabetic Retinopathy Screening With an Offline Artificial Intelligence System on a Smartphone. JAMA Ophthalmol [Internet]. American Medical Association; 2019 [cited 2025 Oct 4];137:1182–8. https://doi.org/10.1001/JAMAOPHTHALMOL.2019.2923 Dow ER, Khan NC, Chen KM, Mishra K, Perera C, Narala R, et al. AI-Human Hybrid Workflow Enhances Teleophthalmology for the Detection of Diabetic Retinopathy. Ophthalmology Science [Internet]. Elsevier Inc.; 2023 [cited 2025 Oct 4];3. https://doi.org/10.1016/j.xops.2023.100330 Zhang G, Lin JW, Wang J, Ji J, Cen LP, Chen W, et al. Automated multidimensional deep learning platform for referable diabetic retinopathy detection: a multicentre, retrospective study. BMJ Open [Internet]. British Medical Journal Publishing Group; 2022 [cited 2025 Oct 4];12:e060155. https://doi.org/10.1136/BMJOPEN-2021-060155 World Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance: executive summary. Https://IrisWhoInt/Bitstream/Handle/10665/350567/9789240037403-EngPdf. 2021;1–148. Artificial Intelligence for Health [Internet]. [cited 2025 Oct 5]. https://www.who.int/publications/m/item/artificial-intelligence-for-health. Accessed 5 Oct 2025 Tables Table 1 Search strategies and inclusion and exclusion criteria Inclusion criteria Studies analyzing the use of artificial intelligence for automatic detection of new-onset diabetic retinopathy in adults worldwide Publications between January 2 019 and September 2 025. Articles in English, Spanish, or Portuguese. Scientific articles, systematic or scoping reviews, qualitative, quantitative, or mixed studies, and technical reports. Adult patients over 18 years old. Exclusion criteria Editorials, letters to the editor, opinions without empirical support, narrative reviews, case reports. Studies published outside the established time frame. Exclusively clinical or technological studies that do not address information about the central topic. Publications focused solely on clinical or surgical interventions, unrelated to the use of artificial intelligence. Publications focused solely on classification, stratification, treatment, prognosis, and complications associated with diabetic retinopathy. Patients with other conditions or comorbidities. Search strategies PubMed ("Diabetic Retinopathy"[Mesh] OR "diabetic retinopathy"[tiab] OR "retinal diabetic"[tiab] OR "retinal microaneurysm*"[tiab]) AND ("Artificial Intelligence"[Mesh] OR "Machine Learning"[Mesh] OR "Deep Learning"[Mesh] OR "Neural Networks (Computer)"[Mesh] OR "Algorithms"[Mesh] OR "artificial intelligence"[tiab] OR "machine learning"[tiab] OR "deep learning"[tiab] OR "neural network*"[tiab] OR "convolutional neural network*"[tiab] OR CNN[tiab] OR "support vector machine*"[tiab] OR SVM[tiab] OR "transfer learning"[tiab] OR "automated detection"[tiab] OR "automatic diagnos*"[tiab] OR "computer-aided diagnos*"[tiab] OR CAD[tiab]) AND (diagnos*[tiab] OR screen*[tiab] OR detect*[tiab] OR "early detection"[tiab]) NOT (pediatr*[tiab] OR child*[tiab] OR infant*[tiab]) AND ("2019/01/01"[Date - Publication] : "2025/09/30"[Date - Publication]) Scopus TITLE-ABS-KEY("diabetic retinopathy" OR "retinopatía diabética" OR "retinopatia diabética") AND TITLE-ABS-KEY("artificial intelligence" OR "machine learning" OR "deep learning" OR "neural network*" OR "computer-assisted diagnosis" OR "computer-aided diagnosis" OR "automated detection" OR "automatic detection" OR "AI" OR "IA") AND TITLE-ABS-KEY(adult OR adults OR "≥18 years") AND NOT TITLE-ABS-KEY(pediatr* OR child* OR infant*) AND TITLE-ABS-KEY(ophthalmology OR oftalmología OR "eye" OR "ocular" OR "retina") AND NOT TITLE-ABS-KEY(treatment OR therapy OR prognosis OR pronóstico OR complication* OR complicación*) AND PUBYEAR > 2018 AND PUBYEAR < 2026 AND (LIMIT-TO (DOCTYPE,"ar") OR LIMIT-TO (DOCTYPE,"re")) AND (LIMIT-TO (LANGUAGE,"English") OR LIMIT-TO (LANGUAGE,"Spanish") OR LIMIT-TO (LANGUAGE,"Portuguese")) Web of Science (diabetic retinopathy OR retinopatía diabética OR retinopathia diabética) (All Fields) and (artificial intelligence OR machine learning OR deep learning OR neural network* OR computer-aided diagnosis OR computer assisted diagnosis OR AI OR data mining) (All Fields) and (automatic detection OR detection OR diagnos* OR screening OR identificat*) (All Fields) not (pediatr* OR child* OR infant* OR niño* OR infantil) (All Fields) and (ophthalmology OR oftalmología OR ocular OR eye OR retina) (All Fields) not (treatment OR therapy OR tratamiento OR pronóstico OR prognosis OR complicación* OR complication*) (All Fields) and 2025 or 2024 or 2023 or 2022 or 2021 or 2020 (Publication Years) and Article or Review Article or Early Access (Document Types) BIREME (LILACS + SciELO) (((tw:("retinopatía diabética") OR mh:"Retinopatía Diabética" OR tw:"diabetic retinopathy")) AND ((tw:"inteligencia artificial" OR mh:"Inteligencia Artificial" OR tw:"artificial intelligence" OR tw:"machine learning" OR tw:"aprendizaje automático" OR tw:"deep learning" OR tw:"red neuronal" OR tw:"neural network*" OR tw:"support vector machine*" OR tw:"SVM" OR tw:"convolutional neural network*" OR tw:"CNN")) AND ((tw:"detección" OR tw:"diagnóstico" OR tw:"screening" OR tw:"automatic detection" OR tw:"computer-aided diagnosis" OR tw:"CAD")) AND NOT (tw:pediatr* OR tw:niño* OR tw:infantil OR tw:child* OR tw:infant*) AND (tw:oftalmología OR mh:"Oftalmología" OR tw:ophthalmology OR tw:ocular OR tw:retina) AND NOT (tw:tratamiento OR tw:therapy OR tw:pronóstico OR tw:prognosis OR tw:complicación* OR tw:complication*) Table 2 Synthesis of the results according to the biomedical data Modality/Signal # Articles Average SE Average SP Average ACC Average AUC Other Metrics References Retinography 37 92.9% 93.9% 93.9% 95% F1 Score: 95% Kappa: 70% Precision: 90% [11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47] Multimodal models + comparison with ophthalmologists 12 90% 91% 90% 95% F1 Score: 92% PPV: 39% NPV: 98% Kappa: 86% [48, 49, 50, 51 52, 53, 54, 55, 56, 57, 72, 73] OCT y OCTA 7 88.5% 92.1% 91.6% 88.2% PPV: 88.3% F1 score: 94.1% [58, 59, 60, 61, 63, 64, 71] Electronic devices 6 90.2% 87.4 % 70.6 % 86.6 % PPV: 82.6 % NPV: 93.9 % [65, 66, 67, 68, 69, 71]. SE: sensitivity; SP: specificity; ACC: accuracy; AUC: area under the curve; PPV: positive predictive value; NPV: negative predictive value Table 3 Benefits and limitations of using artificial intelligence in the diagnosis of diabetic retinopathy Category Description References Benefits Superior SE and SP metrics support AI as a reliable tool for screening and clinical support. [11, 19, 21, 26, 34, 39, 54, 59] Automating image analysis allows large volumes to be processed in less time, reducing the ophthalmologist's workload and promoting early detection programs. [20, 22, 25, 30, 55, 56, 72] Portable, low-cost systems facilitate mass screening in rural and primary care settings, expanding coverage in populations with limited resources. [24, 51, 57, 68, 69, 71] Models using techniques such as Grad-CAM allow visualization of the regions of the image used in the prediction, strengthening clinical confidence. [16, 43, 44, 58] Limitations Algorithms require high-quality images; variability between devices and capture conditions affects SE and SP in real-world practice. [17, 21, 22, 28, 54, 59, 72] Some studies are validated in public databases, with limited multicenter evaluation or in diverse populations. [11, 23, 42, 45, 55, 58, 63] High rates of false positives are reported, leading to unnecessary referrals and reducing SP, especially in mild cases or cases with other eye conditions. [51, 52, 53, 55, 69, 72] There are no uniform acquisition protocols; furthermore, it is difficult to detect early lesions or macular edema with a single image. [15, 32, 34, 48, 68, 73] Transparency and clinical validation of algorithms must be strengthened to ensure safe integration into healthcare practice. [18, 50, 59] SE: sensitivity; SP: specificity; AI: artificial intelligence Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-8071096","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":557653721,"identity":"21bf33d4-d592-475e-933c-b7c596547065","order_by":0,"name":"Maria Paula Trujillo López","email":"","orcid":"","institution":"Universidad de La Sabana","correspondingAuthor":false,"prefix":"","firstName":"Maria","middleName":"Paula Trujillo","lastName":"López","suffix":""},{"id":557653722,"identity":"3b79168e-492e-4e61-8322-5c23d7f95661","order_by":1,"name":"Maria Alejandra Cortés González","email":"","orcid":"","institution":"Universidad de La Sabana","correspondingAuthor":false,"prefix":"","firstName":"Maria","middleName":"Alejandra Cortés","lastName":"González","suffix":""},{"id":557653723,"identity":"16b17c9d-b089-4133-90ab-82cd1a441883","order_by":2,"name":"Andrea Valentina Gómez López","email":"","orcid":"","institution":"Universidad de La Sabana","correspondingAuthor":false,"prefix":"","firstName":"Andrea","middleName":"Valentina Gómez","lastName":"López","suffix":""},{"id":557653724,"identity":"1599f31e-96e1-4099-b244-02add3eedc76","order_by":3,"name":"Erwin Hernando Hernández Rincón","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9ElEQVRIiWNgGAWjYJCCA4wNDHIMDEASiMBAghgtxqRpAalMbIAyCGvhlz5jeLhwx730Dbebmz/83GGXx9/AfPA2D4OdHC4tkn05BodnninO3XDnYJtk75nkYokDbMnWPAzJxri0GJzh3XCYty0hd8ONxDYG3jbmxIYDPGbSPAwHEhtwaLGHakk3uJHY/PFvW33i/AP83/BqMeCBaEkAammQ5m07nLjhAA8bXi0SZ/g/gLQYzgQ6TFr2zPHEjYfZjC3nGOD2C38PW/JnoBZ5vhvpjz++3VGdOO9488MbbypwhxgWwAx2MAkaRsEoGAWjYBRgAADT1VsZJ3cicQAAAABJRU5ErkJggg==","orcid":"","institution":"Universidad de La Sabana","correspondingAuthor":true,"prefix":"","firstName":"Erwin","middleName":"Hernando Hernández","lastName":"Rincón","suffix":""}],"badges":[],"createdAt":"2025-11-09 20:23:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8071096/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8071096/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":97932259,"identity":"60f70a2e-a695-4c1b-b872-a696695c6325","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":53706,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/decda1474b7a9a35a5f570c2.docx"},{"id":98421936,"identity":"aed91a08-fd8d-4e2e-b2bc-a8405cee3559","added_by":"auto","created_at":"2025-12-17 16:29:57","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":566960,"visible":true,"origin":"","legend":"","description":"","filename":"FINALARTICLERDNPJ2.docx","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/90a52c421f537ed39eccf1f8.docx"},{"id":97932254,"identity":"9bfeb05a-9873-4f31-944e-74d01b6751f4","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":305252,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.docx","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/3484328d646082ac639220b5.docx"},{"id":98421572,"identity":"614e804d-affc-4b15-bf56-6cfa1785567d","added_by":"auto","created_at":"2025-12-17 16:28:32","extension":"json","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6408,"visible":true,"origin":"","legend":"","description":"","filename":"df7a83cfe1b74772a0f7a26f92c37975.json","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/023cbffdce32e32c1f5ff56b.json"},{"id":97932261,"identity":"5bca6830-7989-4477-8c8d-1de6ed7bb958","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"xml","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":185409,"visible":true,"origin":"","legend":"","description":"","filename":"df7a83cfe1b74772a0f7a26f92c379751enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/ae4d4378c2def34b97a3ea71.xml"},{"id":98421588,"identity":"4e49fd9f-27bb-4fbe-88cb-2994936c9843","added_by":"auto","created_at":"2025-12-17 16:28:39","extension":"eps","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":539,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/2b8ffec79a43a1f1468f2883.eps"},{"id":98422379,"identity":"640c201d-c011-4c2d-8d06-f2bef1fab995","added_by":"auto","created_at":"2025-12-17 16:30:56","extension":"eps","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":549,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage2.eps","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/ab8ec0298f2f52923748c77c.eps"},{"id":98422800,"identity":"f182770e-6e10-44d7-b726-2435d6969c77","added_by":"auto","created_at":"2025-12-17 16:31:31","extension":"eps","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":549,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage2.eps","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/b9e6e3418813529456928142.eps"},{"id":98422832,"identity":"1165a76f-4d57-4197-a2f2-112e56e65cdf","added_by":"auto","created_at":"2025-12-17 16:31:33","extension":"eps","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":549,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage2.eps","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/73e0bc51e29edf6560e0c283.eps"},{"id":97932256,"identity":"00c690a1-dc0c-4e32-97c5-3a74c5124c47","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"eps","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":522,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage5.eps","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/e12d3c1c78cf7e5dd52de713.eps"},{"id":98421305,"identity":"e0f8d0bd-b9a4-4d59-b40a-e8473fac0bed","added_by":"auto","created_at":"2025-12-17 16:26:28","extension":"eps","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":72234,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage6.eps","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/6b5fbdeb6608aa8a6cb184f1.eps"},{"id":98422846,"identity":"efb84f56-70d6-4e1e-920c-86ba2c243d34","added_by":"auto","created_at":"2025-12-17 16:31:34","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":60341,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/4f7587a3e8603980826db245.jpeg"},{"id":97932268,"identity":"ecc358ed-c969-4cfe-bb14-1fb7a4ac8170","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":635251,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/15ec8173a21ca530b626fbab.jpeg"},{"id":98421452,"identity":"6beccd2a-1307-4ba9-b665-35d98edafa8f","added_by":"auto","created_at":"2025-12-17 16:27:36","extension":"jpeg","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":35975,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/4d59bbc28929a26fa5603845.jpeg"},{"id":98421712,"identity":"d7e960b2-973e-424a-9365-8c08b1a10ca0","added_by":"auto","created_at":"2025-12-17 16:29:02","extension":"jpeg","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":28690,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/de0f125ef78f9b5ecc63dad2.jpeg"},{"id":98421509,"identity":"59df55fc-b051-4da7-915f-e929a220facc","added_by":"auto","created_at":"2025-12-17 16:28:10","extension":"jpeg","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26375,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/ab0037820de4e8d29ca199c6.jpeg"},{"id":98421839,"identity":"c909315c-b345-4db4-8acc-02ad1e5ecbb9","added_by":"auto","created_at":"2025-12-17 16:29:36","extension":"jpeg","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25205,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/b8a853b819ed4c719e65d9f5.jpeg"},{"id":97932272,"identity":"baa23d5f-fa5f-4ca8-91e3-feef54368352","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":295707,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/23e419931e4e2dddb768cf11.png"},{"id":98421620,"identity":"719eb372-9632-457b-ad0a-23b058ea4c86","added_by":"auto","created_at":"2025-12-17 16:28:42","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11095,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/04b6e2d28577270a115cb360.png"},{"id":98422850,"identity":"f9f1cae5-4ce8-4b1b-8426-820aa57bf795","added_by":"auto","created_at":"2025-12-17 16:31:34","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":132877,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/0882f71bee622419f888edea.png"},{"id":97932264,"identity":"41fbae53-cb9d-4c69-8cd5-9160f98aa605","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6635,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/750b2cfce50a36523160e3b2.png"},{"id":98421423,"identity":"07751d05-22e4-4ebb-ab73-487b7662ec6a","added_by":"auto","created_at":"2025-12-17 16:27:14","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5893,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/bbb8d6f8e445b866ad55710d.png"},{"id":98421600,"identity":"115f17ed-86ce-4977-8a6e-a8a789eb11b2","added_by":"auto","created_at":"2025-12-17 16:28:39","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4960,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/9203934fa032258af22b8ce6.png"},{"id":98421973,"identity":"bba8549d-57dd-4370-b67c-37ed809893b6","added_by":"auto","created_at":"2025-12-17 16:30:05","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6156,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/82701814ec99206031b98ba7.png"},{"id":98421790,"identity":"0878c680-9402-4c73-9ae8-aacf3b896582","added_by":"auto","created_at":"2025-12-17 16:29:28","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":43751,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/35a0ded771845fd2da82291d.png"},{"id":97932277,"identity":"ef2456fd-fdd0-4004-b782-d0a33880e9cb","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"xml","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":186460,"visible":true,"origin":"","legend":"","description":"","filename":"df7a83cfe1b74772a0f7a26f92c379751structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/4ef30eb25a1ffa7e23b928d6.xml"},{"id":98422908,"identity":"b01bfe5f-d2db-4039-a46f-c88d1a03cee9","added_by":"auto","created_at":"2025-12-17 16:31:38","extension":"html","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":200808,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/419f741ad68318edfddb209c.html"},{"id":97932248,"identity":"38ffc1fa-4f66-49f7-bb6e-f0c12b878412","added_by":"auto","created_at":"2025-12-11 00:43:38","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":60293,"visible":true,"origin":"","legend":"\u003cp\u003ePRISMA flowchart\u003c/p\u003e\n\u003cp\u003e\u003csup\u003eSource: Page MJ, et al. BMJ 2021;372:n71. doi: 10.1136/bmj.n71\u003c/sup\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/e05b12fd1d656298084ab398.png"},{"id":97932249,"identity":"87448b2c-d315-41d2-8df6-574a324e09e2","added_by":"auto","created_at":"2025-12-11 00:43:39","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":348569,"visible":true,"origin":"","legend":"\u003cp\u003eGeographic distribution of studies\u003c/p\u003e\n\u003cp\u003e\u003csup\u003eA predominance was observed in Asia: China (n=14), India (n=9), South Korea (n=1), Iran (n=1), Saudi Arabia (n=1), Turkey (n=2), Iraq (n=1) and Thailand (n=1). In Europe: Spain (n=5), Switzerland (n=1) and Croatia (n=1). In America: United States (n=6), Mexico (n=1), Chile (n=1), Brazil (n=1), and Canada (n=1). In Africa: Zambia (n=1).\u0026nbsp; In Oceania: Australia (n=2). Multinational studies were also recorded (n=12).\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003csup\u003eMapChart. World map creator. 2025. Available from: \u003c/sup\u003e\u003ca href=\"https://www.mapchart.net/world.html\"\u003e\u003csup\u003ehttps://www.mapchart.net/world.html\u003c/sup\u003e\u003c/a\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/a03b1b45b59c03b2a8f61989.png"},{"id":98775877,"identity":"f4c2ec6f-8901-481b-9bb3-0eb8e212de17","added_by":"auto","created_at":"2025-12-22 12:21:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1297881,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8071096/v1/b0446998-9199-480e-bae8-75725d94dab4.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Artificial Intelligence for the Diagnosis of De Novo Diabetic Retinopathy: A Scoping Review of Global Evidence","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eDR is one of the most common and characteristic microvascular complications of DM [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. It can manifest as no proliferative or proliferative DR, the latter being the most advanced and posing the greatest risk to visual health. It is estimated that up to one-third of people with DM develop some degree of DR, making it one of the most prevalent complications of this condition and one of the leading causes of blindness worldwide [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. By 2024, its estimated global prevalence was 22.3% [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Additionally, it is projected that by 2045, it will affect more than 161\u0026nbsp;million people [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Considering that these changes are irreversible and that the burden of DR can vary depending on geographical location, lifestyle, and patients' access to healthcare systems [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], improving detection methods to provide timely management is essential.\u003c/p\u003e\u003cp\u003eThe gold standard for diagnosing DR is the capture of stereoscopic photographs of seven 30\u0026deg; fields, as they offer greater SE and SP than ophthalmoscopic methods do, although their performance depends on image quality and adequate interpretability [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Despite this, fundoscopy with pupil dilation continues to be used in some settings because of its low cost, although it has inferior performance, requires more time, does not allow clinical records to be kept, and its effectiveness depends on the examiner's experience [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. These limitations reflect a gap in the early identification of complications arising from DM and highlight the need for innovative methods that allow for a more efficient and coordinated approach [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn this context, the implementation of AI in medicine has emerged as a promising tool for addressing and supporting clinical decision-making. Algorithms have been shown to achieve high levels of diagnostic performance, promoting early detection of DR and improving the quality of health services in developing countries [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Therefore, a scoping review of the evidence for AI in the de novo diagnosis of DR in people over 18 years of age worldwide was conducted to map the literature and provide an updated global overview.\u003c/p\u003e"},{"header":"MATERIALS AND METHODS","content":"\u003cp\u003eA scoping review was conducted in September 2025 following the Joanna Briggs Institute methodology and PRISMA-ScR guidelines [9,10]. This allowed for rigorous mapping of the literature. The review was guided by the following research question: What is the available evidence on the implementation of AI for the de novo diagnosis of DR in adults worldwide?\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSearch strategy\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe search was conducted in four databases: PubMed, Scopus, BIREME, and Web of Science. Articles related to the research question were identified, and studies from January 2019 to September 2025 in English, Spanish, and Portuguese were included. The search terms were designed using MeSH-DeCS descriptors and related keywords.\u003c/p\u003e\n\u003cp\u003eThe specific search strategies for each database, as well as the inclusion and exclusion criteria applied, are detailed in \u003cstrong\u003eTable 1.\u0026nbsp;\u003c/strong\u003eThe selection of databases was based on their relevance to answering the research question. PubMed was included for its biomedical focus and broad coverage, Scopus for its interdisciplinary scope, BIREME for its access to the literature from Latin America and the Caribbean, and Web of Science for its rigorous indexing and relevant studies from high-impact international journals.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudy selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll the citations were imported and managed via the Rayyan platform, an online tool designed for managing systematic reviews, for efficient study selection and data management. For article selection, duplicates were first removed. Second, three assessors (MPTL, MACG, AVGL) independently selected titles and abstracts according to the inclusion and exclusion criteria. Any discrepancies were resolved jointly by the three evaluators to reach a final agreement. Finally, the reviewers conducted an in-depth reading of the selected articles to determine the final eligibility of the studies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData extraction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData extraction was carried out via a table previously designed by the research team. This table included the following elements: title, authors, date of publication, country or location, journal, type of study, study objective, population included or number of articles (in the case of reviews), key results and findings, type of AI used, and its application in the diagnosis of DR. Similarly, the main results and key findings were identified, as were the advantages derived from the use of AI in diagnosis and the limitations noted. On the other hand, opportunities and recommendations for the future were compiled to continue strengthening research in this area. This information is available in the Zenodo repository [11].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData analysis and synthesis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data obtained were analysed via a narrative and descriptive synthesis, organising the information according to the application of AI in the de novo diagnosis of DR in adults according to the type of biomedical data used. The different approaches were subsequently compared quantitatively and qualitatively to identify similarities and differences in their implementation. Finally, the advantages and limitations noted in the literature were integrated, as were future opportunities for the development and validation of these technologies in different clinical contexts.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003eA total of 1,852 articles were identified in the initial search. Of these, 345 (18.6%) were eliminated because they were duplicates. A total of 1,507 (81.3%) studies were reviewed, of which 1,409 were excluded after review of the title and abstract because they did not meet the inclusion criteria. Ninety-eight articles were selected for full-text reading, 13 articles could not be accessed, and 23 articles did not meet the inclusion criteria. A total of 62 studies were included in this review. The selection and flow of the included articles are presented via the PRISMA model in \u003cstrong\u003eFigure 1.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSixty-two articles evaluating the use of AI for the de novo diagnosis of DR in adults were included. Among these, reviews (n=5) were identified (one scoping review and four systematic reviews). Most of these studies were experimental studies (n=27) that focused mainly on the development and validation of AI algorithms or models. Cross-sectional observational studies (n=21), including retrospective designs, cohort studies, and diagnostic validations in clinical practice, were also identified. Finally, prospective or methodological studies (n=9) were recorded, with a focus on clinical validations in real-world settings, prospective cohorts, and technical developments with ophthalmic imaging.\u003c/p\u003e\n\u003cp\u003eIn terms of temporal distribution, the studies included were published between 2019 and 2025. In 2019, the evidence was limited (n=3). An increase was subsequently observed in 2020 (n=7), 2021 (n=6), 2022 (n=15), 2023 (n=7), 2024 (n=12), and 2025 (n=12). The geographical distribution of the studies, classified by continent, is illustrated in \u003cstrong\u003eFigure 2.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;With respect to biomedical data, the most widely used tool was retinography (n=37), followed by multimodal tools (n=12). There were also studies based on OCT/OCTA (n=7). Finally, electronic devices (n=6), mainly nonmydriatic portable cameras and systems coupled to smartphones in telemedicine settings, were used. A summary of the quantitative results according to biomedical data is shown in \u003cstrong\u003eTable 2.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTypes of artificial intelligence for detection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAmong the 62 articles reviewed, deep neural networks or DL (n=51) stood out, followed by articles based on machine learning (ML) algorithms (n=4). In addition, some articles used hybrid DL + ML models (n=7).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAmong the subtypes of DL applied were: Convolutional neural networks (CNN) (n=48); Transfer learning (n=18); Ensemble (n=12); Attention mechanisms (n=5); Transformer (n=2); Deep belief nets (n=2) and, finally, Recurrent neural networks (RNN) (n=1). On the other hand, the ML categories implemented were: Vector machine (n=7); Random Forest (n=5); Decision Tree (n=2); Multilayer perceptron (n=2); Naive bayes (n=1); Radial basis function classifier (n=2); and finally, Fuzzy methods (n=2).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eArtificial intelligence applied to retinography\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThirty-seven studies were analysed, which together showed high performance, exceeding the performance of standard screening. DL models and hybrids stand out, maintaining good results even with low-quality images. Among the most relevant models are CNN systems, hybrid approaches, and methods that optimise processing times [12\u0026ndash;14]. Several articles have evaluated validation in real clinical settings, where automated platforms and systems outperform previous algorithms [15\u0026ndash;17]. Recent architectures with results close to those of retina specialists were also identified, as were models designed to improve the discrimination of specific lesions [18\u0026ndash;48].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eArtificial intelligence applied to multimodal models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThree multimodal studies were evaluated, all of which described good diagnostic performance. Notably, an OCTA\u0026ndash;fundus fusion model [49], an OCT-based model using morphology and reflectivity [50], and a comparison between OCT and retinography were used[51]. Nine studies comparing the use of AI for the diagnosis of DR with the criteria of trained ophthalmologists were also reviewed. The analysis of retinography on a telemedicine platform yielded consistent results [52], as did the use of conventional retinography compared with the criteria of ophthalmologists [53\u0026ndash;56]. Finally, the use of portable cameras, compared with standard ophthalmological examinations, has also shown efficacy in both clinical settings and community hospitals [57,58].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eArtificial intelligence applied to optical coherence tomography/optical coherence angiography\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSeven articles evaluated AI for the automatic detection of DR via OCT and OCTA. In a cohort of 664 patients, a DL-based GoogLeNet model proved useful in early stages [59]. A review of 32 studies revealed that CNN and Vision Transformers (ViTs) outperform traditional ML in terms of SE, SP, and area under the curve (AUC) [60]. In contrast, a study using logistic regression and classification trees obtained moderate performance [61], whereas a model using ViTs and data augmentation achieved ACC, SE, and SP values greater than 98% [62]. The integration of three-dimensional features (reflectivity + thickness) improved the results compared with those of SVM, KNN, naive Bayes, decision trees, and classic DL architectures such as GoogLeNet or ResNet-50 [63]. A ResNet101-based CNN externally validated with OCTA showed consistent performance [64], and a hybrid model with multifractal descriptors and multiple layers was able to detect subtle structural irregularities, offering robust and scalable classification in the early stages of DR [65].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eArtificial intelligence applied to electronic devices\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSix studies evaluated the use of AI for DR detection based on images obtained with smartphones. One study used a smartphone with an adapted lens, obtaining 2,130 images and showing high concordance with expert ophthalmologists [66]. Another study compared smartphone images with conventional retinography images and reported comparable diagnostic performance [67]. In addition, the offline MediosAI tool, which operates on smartphones, demonstrated consistent performance when dilated fields of view were used [68] and, when applied to nonmydriatic images, proved useful in low-resource settings [69]. Finally, the use of portable cameras coupled with AI confirmed the feasibility of screening in community settings [70].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBenefits and limitations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe benefits and limitations of AI for the de novo diagnosis of DR extracted from the articles evaluated are presented in \u003cstrong\u003eTable 3.\u003c/strong\u003e\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003eSummary of results\u003c/h2\u003e\u003cp\u003eThe global outlook on the use of AI for the de novo diagnosis of DR has expanded widely. Most of the studies included in this review were experimental or observational studies. There was a relatively high concentration of research in Asia, especially China and India, and low representation in Latin America and Africa. This distribution reflects an equity and access gap, highlighting the need to encourage research in these regions to ensure the validity and global applicability of the evidence. Additionally, there has been a surge in research since 2020, reflecting a sustained increase in interest in exploring this innovative tool. Regarding the type of biomedical data associated with AI, the most reported modality is retinography, followed by multimodal approaches, OCT/OCTA, and electronic devices. The most used type of AI was DL, mainly CNNs.\u003c/p\u003e\u003cp\u003eThe results confirm the high performance of AI using retinography, with an average ACC of 95.3%, SE of 92%, SP of 94% and an AUC close to 0.95. These results exceed the performance of conventional screening and remain high even in models validated in clinical practice. This finding indicates that retinography was the most robust tool alongside AI in the diagnosis of DR, although its effectiveness will depend on image quality and multicentre validations to ensure its generalisation. On the other hand, multimodal studies achieved SE ranges of 94\u0026ndash;96% and SPs ranges of 94\u0026ndash;99%%, indicating greater reliability when different imaging techniques are combined. Likewise, in direct comparisons with ophthalmologists, SE ranges of 79\u0026ndash;100% and SPs of 86\u0026ndash;100% were obtained, with better performance in retinography and higher-quality cameras. These data indicate high diagnostic performance that can match or exceed the TAs of specialists, although the TP depends on the device and the clinical context, posing the challenge of optimising its use in real-world settings.\u003c/p\u003e\u003cp\u003eStudies based on OCT/OCTA have shown variable AI performance, with ACC ranges of 88\u0026ndash;99%, SE of 63\u0026ndash;99%, SP of 72\u0026ndash;99%, and AUC of 0.74\u0026ndash;0.97. Classic models offered moderate results, whereas more advanced models, such as ViTs and three-dimensional fusions, achieved performance close to 99%. This indicates high potential for early detection, although multicentre validation and image standardisation are required to ensure consistent results outside experimental settings. Finally, studies of electronic devices based on smartphones and portable cameras reported SEs of 83\u0026ndash;100% and SPs of 85\u0026ndash;96%, with AUC close to 0.90\u0026ndash;0.91. Models such as MediosAI showed the best results. This evidence confirms the high potential for community screening given its accessibility and usefulness in low-resource settings.\u003c/p\u003e\u003cp\u003eIn conclusion, evidence shows that AI applied to retinography is the most far-reaching strategy for diagnosing DR, with high ACC values and clinical validity; OCT/OCTA-based models and multimodal approaches reinforce its potential for early detection, although its performance depends on image standardisation and multicentre validations. Similarly, mobile devices confirm the feasibility of implementing screening in communities with limited resources. These findings are consistent with previous systematic reviews, which have documented robust diagnostic performance of AI, with SE and SP comparable to or even superior to those of trained clinicians, reinforcing its potential for integration into large-scale clinical practice [\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e71\u003c/span\u003e]. From this perspective, AI is seen as an effective tool with diagnostic potential in various contexts, but its impact will depend on multicentre validation and an approach that minimises the risks of bias arising from the use of public databases and high-quality images that are not representative of actual clinical practice.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003eExplainability and impact on clinical confidence in artificial intelligence\u003c/h2\u003e\u003cp\u003eIn addition to diagnostic performance, a key aspect identified in the literature is the explainability of AI models. Tools such as gradient-weighted class activation mapping (Grad-CAM) allow heatmaps to be generated on retinography or OCTs, highlighting relevant lesions (microaneurysms, exudates, and haemorrhages) that support the algorithm's decision. Similarly, ViTs, in addition to achieving performance close to 99% in OCT/OCTA [\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e], offer self-attention mechanisms that allow visualisation of which areas of the image influence the classification. However, most of the studies included still report models with limited interpretability, which reduces professional confidence and hinders their validation in real-world scenarios. The evidence highlights that more interpretable and visually understandable models favour both medical acceptance and patient confidence and underscore the need for regulatory frameworks and clinical guidelines that define minimum standards of explainability for safe integration into DR diagnosis.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003eClinical relevance\u003c/h2\u003e\u003cp\u003eAI offers substantial advantages in the de novo diagnosis of DR. Its high SE and SP metrics support its reliability for screening by detecting early microvascular lesions that may go unnoticed in conventional assessments [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e, \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e, \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e66\u003c/span\u003e]. Automation streamlines the analysis of large volumes of data, reduces the workload of ophthalmologists, and facilitates early detection [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e]. In addition, the availability of portable, low-cost systems facilitates screening in primary, rural, or even home settings by nonspecialised personnel, expanding coverage and improving equity in access to care [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e, \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e, \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e, \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e72\u003c/span\u003e]. These characteristics reinforce the relevance of AI in optimising early detection, reducing underdiagnosis, and strengthening the sustainability of visual health programmes [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e, \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e, \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e73\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cb\u003eFuture opportunities.\u003c/b\u003e\u003c/p\u003e\u003cp\u003eFuture opportunities highlight the need for external validation in diverse, multicentre cohorts, which is key to ensuring the generalisation of models [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e, \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e, \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e][\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e, \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e67\u003c/span\u003e]. The Another priority is the optimisation of models to improve robustness against noise and artefacts and adjust diagnostic thresholds [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e, \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e]. The multimodal fusion of fundus images, OCT/OCTA, and clinical data is emerging as a strategy to increase diagnostic ACC [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e, \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e, \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e]. In addition, emphasis is placed on advancing explainability and interpretability through techniques such as heatmaps or SHAP/LIME, which promote clinical confidence and acceptance by patients and professionals [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e], and exploring transfer learning and advanced architectures to strengthen generalisation and standardisation protocols between datasets [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e, \u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e74\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003eEthical considerations\u003c/h2\u003e\u003cp\u003eThis review used only published information, so it did not require ethical approval or informed consent. However, the studies analysed involve the handling of clinical and biomedical data, which require strict confidentiality, data protection, and governance frameworks to regulate ownership, access, and responsible use. According to the Ethics and governance of artificial intelligence for health: WHO guidance [\u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e75\u003c/span\u003e], it is essential to protect privacy through anonymization and transparent governance, with regulatory frameworks that allow auditing and accountability. It is also important to ensure equity in the implementation of these technologies, avoiding the exclusion of vulnerable populations or regions with less digital infrastructure, which can perpetuate health gaps. The WHO highlights the need for transparency and explainability in algorithms so that their decisions are auditable and understandable by health professionals [\u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e76\u003c/span\u003e]. Furthermore, under the principles of nonmaleficence and shared responsibility, maintaining human oversight in clinical interpretation and establishing monitoring mechanisms to identify biases and prevent potential harm are recommended, promoting effective and socially responsible adoption of AI.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003eLimitations of the article\u003c/h2\u003e\u003cp\u003eDespite the promising results of AI for the de novo diagnosis of DR, this study has several limitations. First, much of the evidence collected comes from public databases or specific populations, which leads to heterogeneity and lower external validation. In addition, there was a high dependence on optimal high-quality images obtained in controlled environments, with variability in the devices used, which limits generalisation to real clinical contexts. There is a lack of longitudinal and multicentre studies in diverse populations, which restricts the evaluation of the performance of the models needed in routine practice. Additional methodological limitations, such as publication bias, should also be considered. Finally, the rapid evolution of AI techniques requires periodic updates of the evidence and continuous evaluations to ensure that the models remain valid and relevant in real clinical settings.\u003c/p\u003e\u003c/div\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThe application of AI in the de novo diagnosis of DR is establishing itself as an effective and versatile tool, with retinography standing out as the most robust technique and multimodal approaches, OCT/OCTA, and electronic devices as alternatives with great potential for early detection and screening, even in communities with limited resources. Its benefits include automation, expanded coverage, and a reduced healthcare burden, reinforcing its value as an innovative diagnostic support tool. However, its implementation faces challenges related to image quality, protocol standardisation, the need for multicentre validation, and improved interpretability. In this context, the evidence positions AI as a promising complement to clinical practice, whose integration will depend on overcoming current limitations and consolidating its application in a safe and generalisable manner through multicentre, longitudinal studies with greater methodological rigour.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eDR- Diabetic retinopathy\u003c/p\u003e\n\u003cp\u003eDM- Diabetes mellitus\u003c/p\u003e\n\u003cp\u003eAI - Artificial intelligence\u003c/p\u003e\n\u003cp\u003eDL - Deep learning\u003c/p\u003e\n\u003cp\u003eOCT- Optical coherence tomography\u003c/p\u003e\n\u003cp\u003eOCTA - Optical coherence tomography angiography\u003c/p\u003e\n\u003cp\u003eSE - Sensitivity\u003c/p\u003e\n\u003cp\u003eSP- Specificity\u003c/p\u003e\n\u003cp\u003eML - Machine learning\u003c/p\u003e\n\u003cp\u003eCNN - Convolutional neural network\u003c/p\u003e\n\u003cp\u003eRNN - Recurrent neural networks\u003c/p\u003e\n\u003cp\u003eACC - Accuracy\u003c/p\u003e\n\u003cp\u003eAUC - Area under the curve\u003c/p\u003e\n\u003cp\u003eViTs - Vision Transformers\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAUTHORS\u0026apos; CONTRIBUTIONS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMPTL:\u0026nbsp;\u003c/strong\u003eConceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Writing \u0026ndash; original draft.\u003cstrong\u003e\u0026nbsp;MACG:\u0026nbsp;\u003c/strong\u003eConceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Visualization, Writing \u0026ndash; original draft. \u003cstrong\u003eAVGL:\u0026nbsp;\u003c/strong\u003eConceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Visualization, Writing \u0026ndash; original draft. \u003cstrong\u003eEHHR:\u0026nbsp;\u003c/strong\u003eFunding acquisition, Methodology, Project administration, Supervision, Validation, Writing \u0026ndash; review \u0026amp; editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical approval:\u0026nbsp;\u003c/strong\u003eThis study was approved by the Ethics and Research Committee of the Faculty of Medicine of the University of La Sabana.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability:\u0026nbsp;\u003c/strong\u003eThe datasets analysed during the current study are available in the Zenodo repository [11].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of interest:\u0026nbsp;\u003c/strong\u003eAll authors declare no financial or non-financial competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u0026nbsp;\u003c/strong\u003eResearch derived from the MED-341-2023 project, developed by the Family Medicine and Population Health Research Group at Universidad de La Sabana, Colombia.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements:\u003c/strong\u003e Not applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eShah J, Cheong ZY, Tan B, Wong D, Liu X, Chua J. Dietary Intake and Diabetic Retinopathy: A Systematic Review of the Literature. Nutrients [Internet]. MDPI; 2022 [cited 2025 Sep 7];14:5021. https://doi.org/10.3390/NU14235021/S1\u003c/li\u003e\n\u003cli\u003eSeo H, Park SJ, Song M. Diabetic Retinopathy (DR): Mechanisms, Current Therapies, and Emerging Strategies. Cells [Internet]. Multidisciplinary Digital Publishing Institute (MDPI); 2025 [cited 2025 Sep 7];14. https://doi.org/10.3390/CELLS14050376,\u003c/li\u003e\n\u003cli\u003eBravo Gonz\u0026aacute;lez R, Cuartas Mesa JJ, Garz\u0026oacute;n Duque MO, Bravo Acosta JD, V\u0026aacute;squez MCF, Escobar Howard M, et al. Factores relacionados con la prevalencia de retinopat\u0026iacute;a diab\u0026eacute;tica. Revista Med [Internet]. Universidad Militar Nueva Granada; 2024 [cited 2025 Sep 7];32:73\u0026ndash;91. https://doi.org/10.18359/RMED.7374\u003c/li\u003e\n\u003cli\u003eDiabetic Retinopathy - Latin America - American Academy of Ophthalmology [Internet]. [cited 2025 Sep 7]. https://www.aao.org/education/topic-detail/diabetic-retinopathy-latin-america?utm. Accessed 7 Sep 2025\u003c/li\u003e\n\u003cli\u003eKumari S, Venkatesh P, Tandon N, Chawla R, Takkar B, Kumar A. Selfie fundus imaging for diabetic retinopathy screening. Eye [Internet]. Springer Nature; 2021 [cited 2025 Sep 20];36:1988. https://doi.org/10.1038/S41433-021-01804-7\u003c/li\u003e\n\u003cli\u003eYuribia Karina Mill\u0026aacute;n-G\u0026aacute;mez 2Niels H. Wacher-Rodarte,3Juan Carlos Bravo-Ortiz, 4N\u0026eacute;stor Hugo Garrido-Gaspar,5Rosaura Mendoza-Topete, Adriana Marcela Mart\u0026iacute;nez-Ruiz,Guillermo Miller-Arrevillaga, Alfonso Ram\u0026iacute;rez-N\u0026aacute;jera. Diabetic retinopathy screening in the emerging era of artificial intelligence. Diabetologia [Internet]. Springer Science and Business Media Deutschland GmbH; 2022 [cited 2025 Sep 20];65:1415\u0026ndash;23. https://doi.org/10.1007/S00125-022-05727-0/FIGURES/3\u003c/li\u003e\n\u003cli\u003eGunasekeran D V., Ting DSW, Tan GSW, Wong TY. Artificial intelligence for diabetic retinopathy screening, prediction and management. Curr Opin Ophthalmol [Internet]. Lippincott Williams and Wilkins; 2020 [cited 2025 Sep 7];31:357\u0026ndash;65. https://doi.org/10.1097/ICU.0000000000000693\u003c/li\u003e\n\u003cli\u003eCastrill\u0026oacute;n-Lozano JA, Arango-C\u0026aacute;rdenas D, Marulanda-M\u0026aacute;rquez DE. Inteligencia artificial y tamizaje de la discapacidad visual relacionada con retinopat\u0026iacute;a diab\u0026eacute;tica y edema macular. Gac Med Mex. 2024;160:217\u0026ndash;8. https://doi.org/10.24875/GMM.M24000881\u003c/li\u003e\n\u003cli\u003eAromataris E et al. JBI Manual for Evidence Synthesis. Aromataris E, Lockwood C, Porritt K, Pilla B, Jordan Z, editors. JBI; 2024. https://doi.org/10.46658/JBIMES-24-01\u003c/li\u003e\n\u003cli\u003eTricco AC, Lillie E, Zarin W, O\u0026rsquo;Brien KK, Colquhoun H, Levac D, et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med. 2018;169:467\u0026ndash;73. https://doi.org/10.7326/M18-0850\u003c/li\u003e\n\u003cli\u003eTrujillo Lopez MP. Dataset: Artificial Intelligence for the Diagnosis of De Novo Diabetic Retinopathy A Scoping Review of Global Evidence. Zenodo; 2025. https://doi.org/10.5281/zenodo.17600994\u003c/li\u003e\n\u003cli\u003eBaget-Bernaldiz M, Fontoba-Poveda B, Romero-Aroca P, Navarro-Gil R, Hernando-Comerma A, Bautista-Perez A, et al. Artificial Intelligence-Based Screening System for Diabetic Retinopathy in Primary Care. Diagnostics 2024, Vol 14, Page 1992 [Internet]. Multidisciplinary Digital Publishing Institute; 2024 [cited 2025 Oct 4];14:1992. https://doi.org/10.3390/DIAGNOSTICS14171992\u003c/li\u003e\n\u003cli\u003eNaz H, Nijhawan R, Ahuja NJ. An automated unsupervised deep learning\u0026ndash;based approach for diabetic retinopathy detection. Med Biol Eng Comput [Internet]. Springer Science and Business Media Deutschland GmbH; 2022 [cited 2025 Oct 4];60:3635\u0026ndash;54. https://doi.org/10.1007/S11517-022-02688-9/FIGURES/17\u003c/li\u003e\n\u003cli\u003eOmer HK. Diabetic retinopathy detection using Bilayered Neural Network classification model with resubstitution validation. MethodsX [Internet]. Elsevier; 2024 [cited 2025 Oct 4];12:102705. https://doi.org/10.1016/J.MEX.2024.102705\u003c/li\u003e\n\u003cli\u003eAbreu-Gonzalez R, Susanna-Gonz\u0026aacute;lez G, Blair JPM, Vitar RML, Ciller C, Apostolopoulos S, et al. Validation of artificial intelligence algorithm LuxIA for screening of diabetic retinopathy from a single 45\u0026deg; retinal colour fundus images: the CARDS study. BMJ Open Ophthalmol [Internet]. BMJ Publishing Group Ltd; 2025 [cited 2025 Oct 4];10:2109. https://doi.org/10.1136/BMJOPHTH-2024-002109\u003c/li\u003e\n\u003cli\u003eHu CL, Wang YC, Wu WF, Xi Y. Evaluation of AI-enhanced non-mydriatic fundus photography for diabetic retinopathy screening. Photodiagnosis Photodyn Ther [Internet]. Elsevier; 2024 [cited 2025 Oct 4];49:104331. https://doi.org/10.1016/J.PDPDT.2024.104331\u003c/li\u003e\n\u003cli\u003eAntaki F, Hammana I, Tessier M-C, Boucher A, David Jett\u0026eacute; ML, Beauchemin C, et al. Implementation of Artificial Intelligence-Based Diabetic Retinopathy Screening in a Tertiary Care Hospital in Quebec: Prospective Validation Study. JMIR Diabetes [Internet]. JMIR Diabetes; 2024 [cited 2025 Oct 4];9:e59867. https://doi.org/10.2196/59867\u003c/li\u003e\n\u003cli\u003eNaveen K V., Anoop BN, Siju KS, Kar MK, Venugopal V. EffNet-SVM: A Hybrid Model for Diabetic Retinopathy Classification Using Retinal Fundus Images. IEEE Access. Institute of Electrical and Electronics Engineers Inc.; 2025;13:79793\u0026ndash;804. https://doi.org/10.1109/ACCESS.2025.3566073\u003c/li\u003e\n\u003cli\u003eZhang G, Sun B, Chen Z, Gao Y, Zhang Z, Li K, et al. Diabetic Retinopathy Grading by Deep Graph Correlation Network on Retinal Images Without Manual Annotations. Front Med (Lausanne) [Internet]. Frontiers Media S.A.; 2022 [cited 2025 Oct 4];9:872214. https://doi.org/10.3389/FMED.2022.872214/BIBTEX\u003c/li\u003e\n\u003cli\u003eMuthusamy D, Palani P. Deep neural network model for diagnosing diabetic retinopathy detection: An efficient mechanism for diabetic management. Biomed Signal Process Control [Internet]. Elsevier; 2025 [cited 2025 Oct 4];100:107035. https://doi.org/10.1016/J.BSPC.2024.107035\u003c/li\u003e\n\u003cli\u003eButt MM, Iskandar DNFA, Abdelhamid SE, Latif G, Alghazo R. Diabetic Retinopathy Detection from Fundus Images of the Eye Using Hybrid Deep Learning Features. Diagnostics 2022, Vol 12, Page 1607 [Internet]. Multidisciplinary Digital Publishing Institute; 2022 [cited 2025 Oct 4];12:1607. https://doi.org/10.3390/DIAGNOSTICS12071607\u003c/li\u003e\n\u003cli\u003eTavakoli M, Mehdizadeh A, Aghayan A, Shahri RP, Ellis T, Dehmeshki J. Automated Microaneurysms Detection in Retinal Images Using Radon Transform and Supervised Learning: Application to Mass Screening of Diabetic Retinopathy. IEEE Access [Internet]. Institute of Electrical and Electronics Engineers Inc.; 2021 [cited 2025 Oct 4];9:67302\u0026ndash;14. https://doi.org/10.1109/ACCESS.2021.3074458\u003c/li\u003e\n\u003cli\u003eThomas GAS, Robinson YH, Julie EG, Shanmuganathan V, Rho S, Nam Y. Intelligent Prediction Approach for Diabetic Retinopathy Using Deep Learning Based Convolutional Neural Networks Algorithm by Means of Retina Photographs. Computers, Materials \u0026amp; Continua [Internet]. Tech Science Press; 2020 [cited 2025 Oct 4];66:1613\u0026ndash;29. https://doi.org/10.32604/CMC.2020.013443\u003c/li\u003e\n\u003cli\u003eMutawa AM, Hemalakshmi GR, Prakash NB, Murugappan M. Randomization-Driven Hybrid Deep Learning for Diabetic Retinopathy Detection. IEEE Access [Internet]. Institute of Electrical and Electronics Engineers Inc.; 2025 [cited 2025 Oct 4];13:38901\u0026ndash;13. https://doi.org/10.1109/ACCESS.2025.3546359\u003c/li\u003e\n\u003cli\u003eModi P, Kumar Y. An Effective Diabetic Retinopathy Detection System using Deep Belief Nets and Adaptive Learning in Cloud Environment. Scalable Computing: Practice and Experience [Internet]. Scalable Computing: Practice and Experience; 2023 [cited 2025 Oct 4];24:97\u0026ndash;114. https://doi.org/10.12694/SCPE.V24I2.2117\u003c/li\u003e\n\u003cli\u003eTran J, Estevez JJ, Howard NJ, Kumar S. Barriers and Enablers Influencing the Implementation of Artificial Intelligence for Diabetic Retinopathy Screening in Clinical Practice: A Scoping Review. Clin Exp Ophthalmol [Internet]. John Wiley \u0026amp; Sons, Ltd; 2025 [cited 2025 Oct 4];0:1\u0026ndash;12. https://doi.org/10.1111/CEO.14567\u003c/li\u003e\n\u003cli\u003eWang Y, Han X, Li C, Luo L, Yin Q, Zhang J, et al. Impact of Gold-Standard Label Errors on Evaluating Performance of Deep Learning Models in Diabetic Retinopathy Screening: Nationwide Real-World Validation Study. J Med Internet Res [Internet]. JMIR Publications Inc.; 2024 [cited 2025 Oct 4];26:e52506. https://doi.org/10.2196/52506\u003c/li\u003e\n\u003cli\u003eSelvin Charles A, Akila C. Enhanced detection of diabetic retinopathy using machine learning based feature selection and ensemble classifiers. AIP Adv [Internet]. AIP Publishing; 2025 [cited 2025 Oct 4];15:75029. https://doi.org/10.1063/5.0279598\u003c/li\u003e\n\u003cli\u003eRomero-Ora\u0026aacute; R, Garc\u0026iacute;a M, Ora\u0026aacute;-P\u0026eacute;rez J, L\u0026oacute;pez-G\u0026aacute;lvez MI, Hornero R. Effective Fundus Image Decomposition for the Detection of Red Lesions and Hard Exudates to Aid in the Diagnosis of Diabetic Retinopathy. Sensors 2020, Vol 20, Page 6549 [Internet]. Multidisciplinary Digital Publishing Institute; 2020 [cited 2025 Oct 4];20:6549. https://doi.org/10.3390/S20226549\u003c/li\u003e\n\u003cli\u003eKHAN ZAID, GAIDHANE AM, SINGH M, GANESAN S, KAUR M, SHARMA GC, et al. Diagnostic Accuracy of IDX-DR for Detecting Diabetic Retinopathy: A Systematic Review and Meta-Analysis. Am J Ophthalmol [Internet]. Elsevier Inc.; 2025 [cited 2025 Oct 4];273:192\u0026ndash;204. https://doi.org/10.1016/j.ajo.2025.02.022\u003c/li\u003e\n\u003cli\u003eJOSEPH S, SELVARAJ J, MANI I, KUMARAGURUPARI T, SHANG X, MUDGIL P, et al. Diagnostic Accuracy of Artificial Intelligence-Based Automated Diabetic Retinopathy Screening in Real-World Settings: A Systematic Review and Meta-Analysis. Am J Ophthalmol [Internet]. Elsevier Inc.; 2024 [cited 2025 Oct 4];263:214\u0026ndash;30. https://doi.org/10.1016/j.ajo.2024.02.012\u003c/li\u003e\n\u003cli\u003eMohammedhasan M, Uğuz H. A New Early Stage Diabetic Retinopathy Diagnosis Model Using Deep Convolutional Neural Networks and Principal Component Analysis. [cited 2025 Oct 4]; https://doi.org/10.18280/ts.370503\u003c/li\u003e\n\u003cli\u003eTan-Torres A, Praveen PA, Jeji D, Brant A, Yin X, Yang L, et al. Validation of a Deep Learning Model for Diabetic Retinopathy on Patients with Young-Onset Diabetes. Ophthalmol Ther [Internet]. Adis; 2025 [cited 2025 Oct 4];14:1147\u0026ndash;55. https://doi.org/10.1007/S40123-025-01116-Z/FIGURES/1\u003c/li\u003e\n\u003cli\u003eZhang G-H, Zhuo G-P, Zhang Z-X, Sun B, Yang W-H, Zhang S-C, et al. Diabetic retinopathy identification based on multi-source-free domain adaptation. Int J Ophthalmol [Internet]. International Journal of Ophthalmology (c/o Editorial Office); 2024 [cited 2025 Oct 4];17:1193\u0026ndash;204. https://doi.org/10.18240/IJO.2024.07.03\u003c/li\u003e\n\u003cli\u003eHao S, Liu C, Li N, Wu Y, Li D, Gao Q, et al. Clinical evaluation of AI-assisted screening for diabetic retinopathy in rural areas of midwest China. PLoS One [Internet]. Public Library of Science; 2022 [cited 2025 Oct 4];17:e0275983. https://doi.org/10.1371/JOURNAL.PONE.0275983\u003c/li\u003e\n\u003cli\u003eZhang Y, Shi J, Peng Y, Zhao Z, Zheng Q, Wang Z, et al. Artificial intelligence-enabled screening for diabetic retinopathy: a real-world, multicenter and prospective study. BMJ Open Diabetes Res Care [Internet]. American Diabetes Association; 2020 [cited 2025 Oct 4];8:1596. https://doi.org/10.1136/BMJDRC-2020-001596\u003c/li\u003e\n\u003cli\u003eBellemo V, Lim ZW, Lim G, Nguyen QD, Xie Y, Yip MYT, et al. Artificial intelligence using deep learning to screen for referable and vision-threatening diabetic retinopathy in Africa: a clinical validation study. Lancet Digit Health [Internet]. Elsevier Ltd; 2019 [cited 2025 Oct 4];1:e35\u0026ndash;44. https://doi.org/10.1016/S2589-7500(19)30004-4\u003c/li\u003e\n\u003cli\u003eKarakaya M, Aygun RS, Sallam AB. Collaborative Deep Learning for Privacy Preserving Diabetic Retinopathy Detection. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS. Institute of Electrical and Electronics Engineers Inc.; 2022;2022-July:2181\u0026ndash;4. https://doi.org/10.1109/EMBC48229.2022.9871617\u003c/li\u003e\n\u003cli\u003eAquino-Br\u0026iacute;tez D, G\u0026oacute;mez JA, Noguera JLV, Garc\u0026iacute;a-Torres M, Rom\u0026aacute;n JCM, Gardel-Sotomayor PE, et al. Automatic Diagnosis of Diabetic Retinopathy from Fundus Images Using Neuro-Evolutionary Algorithms. Stud Health Technol Inform [Internet]. IOS Press; 2022 [cited 2025 Oct 4];290:689\u0026ndash;93. https://doi.org/10.3233/SHTI220166\u003c/li\u003e\n\u003cli\u003eRuamviboonsuk P, Tiwari R, Sayres R, Nganthavee V, Hemarat K, Kongprayoon A, et al. Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study. Lancet Digit Health [Internet]. Elsevier Ltd; 2022 [cited 2025 Oct 4];4:e235\u0026ndash;44. https://doi.org/10.1016/S2589-7500(22)00017-6\u003c/li\u003e\n\u003cli\u003eWang J, Jiao W, Zheng Y, Ullah MZ, Song J. Multicolor image classification using the multimodal information bottleneck network (MMIB-Net) for detecting diabetic retinopathy. Optics Express, Vol 29, Issue 14, pp 22732-22748 [Internet]. Optica Publishing Group; 2021 [cited 2025 Oct 4];29:22732\u0026ndash;48. https://doi.org/10.1364/OE.430508\u003c/li\u003e\n\u003cli\u003eVives-Boix V, Ruiz-Fern\u0026aacute;ndez D. Diabetic retinopathy detection through convolutional neural networks with synaptic metaplasticity. Comput Methods Programs Biomed [Internet]. Elsevier; 2021 [cited 2025 Oct 4];206:106094. https://doi.org/10.1016/J.CMPB.2021.106094\u003c/li\u003e\n\u003cli\u003ePalaniswamy T, Vellingiri M. Internet of Things and Deep Learning Enabled Diabetic Retinopathy Diagnosis Using Retinal Fundus Images. IEEE Access. Institute of Electrical and Electronics Engineers Inc.; 2023;11:27590\u0026ndash;601. https://doi.org/10.1109/ACCESS.2023.3257988\u003c/li\u003e\n\u003cli\u003eMazar Pasha LT, Rajashekar JS. Image Based Edge Weighted Linked Segmentation Model Using Deep Learning for Detection of Diabetic Retinopathy. Traitement du Signal [Internet]. International Information and Engineering Technology Association; 2022 [cited 2025 Oct 4];39:165\u0026ndash;72. https://doi.org/10.18280/TS.390116\u003c/li\u003e\n\u003cli\u003ePareja-R\u0026iacute;os A, Ceruso S, Romero-Aroca P, Bonaque-Gonz\u0026aacute;lez S. A New Deep Learning Algorithm with Activation Mapping for Diabetic Retinopathy: Backtesting after 10 Years of Tele-Ophthalmology. Journal of Clinical Medicine 2022, Vol 11, Page 4945 [Internet]. Multidisciplinary Digital Publishing Institute; 2022 [cited 2025 Oct 4];11:4945. https://doi.org/10.3390/JCM11174945\u003c/li\u003e\n\u003cli\u003eJabbar A, Liaqat H Bin, Akram A, Sana MU, Azpiroz ID, Diez IDLT, et al. A Lesion-Based Diabetic Retinopathy Detection Through Hybrid Deep Learning Model. IEEE Access. Institute of Electrical and Electronics Engineers Inc.; 2024;12:40019\u0026ndash;36. https://doi.org/10.1109/ACCESS.2024.3373467\u003c/li\u003e\n\u003cli\u003eLuo X, Pu Z, Xu Y, Wong WK, Su J, Dou X, et al. MVDRNet: Multi-view diabetic retinopathy detection by combining DCNNs and attention mechanisms. Pattern Recognit [Internet]. Pergamon; 2021 [cited 2025 Oct 4];120:108104. https://doi.org/10.1016/J.PATCOG.2021.108104\u003c/li\u003e\n\u003cli\u003eButt MM, Iskandar DNFA, Abdelhamid SE, Latif G, Alghazo R. Diabetic Retinopathy Detection from Fundus Images of the Eye Using Hybrid Deep Learning Features. Diagnostics 2022, Vol 12, Page 1607 [Internet]. Multidisciplinary Digital Publishing Institute; 2022 [cited 2025 Oct 4];12:1607. https://doi.org/10.3390/DIAGNOSTICS12071607\u003c/li\u003e\n\u003cli\u003eBidwai P, Gite S, Pahuja N, Pahuja K, Kotecha K, Jain N, et al. Multimodal image fusion for the detection of diabetic retinopathy using optimized explainable AI-based Light GBM classifier. Information Fusion [Internet]. Elsevier B.V.; 2024 [cited 2025 Oct 4];111. https://doi.org/10.1016/j.inffus.2024.102526\u003c/li\u003e\n\u003cli\u003eSharafeldeen A, Elsharkawy M, Khalifa F, Soliman A, Ghazal M, AlHalabi M, et al. Precise higher-order reflectivity and morphology models for early diagnosis of diabetic retinopathy using OCT images. Sci Rep [Internet]. Nature Research; 2021 [cited 2025 Oct 4];11:1\u0026ndash;16. https://doi.org/10.1038/S41598-021-83735-7;SUBJMETA\u003c/li\u003e\n\u003cli\u003eBi Z, Li J, Liu Q, Fang Z. Deep learning-based optical coherence tomography and retinal images for detection of diabetic retinopathy: a systematic and meta analysis. Front Endocrinol (Lausanne) [Internet]. Frontiers Media SA; 2025 [cited 2025 Oct 4];16:1485311. https://doi.org/10.3389/FENDO.2025.1485311/BIBTEX\u003c/li\u003e\n\u003cli\u003eBoyle J, Vignarajan J, Saha S. Automated Diabetic Retinopathy Diagnosis for Improved Clinical Decision Support. Stud Health Technol Inform [Internet]. IOS Press; 2024 [cited 2025 Oct 4];310:1490\u0026ndash;1. https://doi.org/10.3233/SHTI231259\u003c/li\u003e\n\u003cli\u003eIba\u0026ntilde;ez-Bruron MC, Cruzat A, Ordenes-Caviere G, Coria M, Iba\u0026ntilde;ez-Bruron MC, Cruzat A, et al. Use of Artificial Intelligence in Diabetic Retinopathy Screening: Experience in a Health Service in Santiago, Chile. Rev Med Chil [Internet]. Sociedad M\u0026eacute;dica de Santiago; 2024 [cited 2025 Oct 4];152:1148\u0026ndash;53. https://doi.org/10.4067/S0034-98872024001101148\u003c/li\u003e\n\u003cli\u003eRiotto E, Gasser S, Potic J, Sherif M, Stappler T, Schlingemann R, et al. Accuracy of Autonomous Artificial Intelligence-Based Diabetic Retinopathy Screening in Real-Life Clinical Practice. Journal of Clinical Medicine 2024, Vol 13, Page 4776 [Internet]. Multidisciplinary Digital Publishing Institute; 2024 [cited 2025 Oct 4];13:4776. https://doi.org/10.3390/JCM13164776\u003c/li\u003e\n\u003cli\u003eHe J, Cao T, Xu F, Wang S, Tao H, Wu T, et al. Artificial intelligence-based screening for diabetic retinopathy at community hospital. Eye (Basingstoke) [Internet]. Springer Nature; 2020 [cited 2025 Oct 4];34:572\u0026ndash;6. https://doi.org/10.1038/S41433-019-0562-4;SUBJMETA\u003c/li\u003e\n\u003cli\u003eLim JI, Regillo CD, Sadda SVR, Ipp E, Bhaskaranand M, Ramachandra C, et al. Artificial Intelligence Detection of Diabetic Retinopathy: Subgroup Comparison of the EyeArt System with Ophthalmologists\u0026rsquo; Dilated Examinations. Ophthalmology Science [Internet]. Elsevier Inc.; 2023 [cited 2025 Oct 4];3. https://doi.org/10.1016/j.xops.2022.100228\u003c/li\u003e\n\u003cli\u003eDong X, Du S, Zheng W, Cai C, Liu H, Zou J. Evaluation of an Artificial Intelligence System for the Detection of Diabetic Retinopathy in Chinese Community Healthcare Centers. Front Med (Lausanne) [Internet]. Frontiers Media S.A.; 2022 [cited 2025 Oct 4];9:883462. https://doi.org/10.3389/FMED.2022.883462/BIBTEX\u003c/li\u003e\n\u003cli\u003eTomić M, Vrabec R, Hendelja Đ, Kolarić V, Bulum T, Rahelić D. Diagnostic Accuracy of Hand-Held Fundus Camera and Artificial Intelligence in Diabetic Retinopathy Screening. Biomedicines 2024, Vol 12, Page 34 [Internet]. Multidisciplinary Digital Publishing Institute; 2023 [cited 2025 Oct 4];12:34. https://doi.org/10.3390/BIOMEDICINES12010034\u003c/li\u003e\n\u003cli\u003eSingh R, Singuri S, Batoki J, Lin K, Luo S, Hatipoglu D, et al. Deep Learning Algorithm Detects Presence of Disorganization of Retinal Inner Layers (DRIL)\u0026ndash;An Early Imaging Biomarker in Diabetic Retinopathy. Transl Vis Sci Technol. Association for Research in Vision and Ophthalmology Inc.; 2023;12. https://doi.org/10.1167/tvst.12.7.6\u003c/li\u003e\n\u003cli\u003eHayati A, Abdol Homayuni MR, Sadeghi R, Asadigandomani H, Dashtkoohi M, Eslami S, et al. Advancing Diabetic Retinopathy Screening: A Systematic Review of Artificial Intelligence and Optical Coherence Tomography Angiography Innovations. Diagnostics [Internet]. Multidisciplinary Digital Publishing Institute (MDPI); 2025 [cited 2025 Oct 4];15:737. https://doi.org/10.3390/DIAGNOSTICS15060737/S1\u003c/li\u003e\n\u003cli\u003eYao H, Wu S, Zhan Z, Li Z. A Classification Tree Model with Optical Coherence Tomography Angiography Variables to Screen Early-Stage Diabetic Retinopathy in Diabetic Patients. J Ophthalmol [Internet]. John Wiley \u0026amp; Sons, Ltd; 2022 [cited 2025 Oct 4];2022:9681034. https://doi.org/10.1155/2022/9681034\u003c/li\u003e\n\u003cli\u003eZhou Z, Yu H, Zhao J, Wang X, Wu Q, Dai C. Automatic diagnosis of diabetic retinopathy using vision transformer based on wide-field optical coherence tomography angiography. https://doi.org/101142/S1793545823500190 [Internet]. World Scientific Publishing Company ; 2023 [cited 2025 Oct 4];17:2350019. https://doi.org/10.1142/S1793545823500190\u003c/li\u003e\n\u003cli\u003eElgafi M, Sharafeldeen A, Elnakib A, Elgarayhi A, Alghamdi NS, Sallah M, et al. Detection of Diabetic Retinopathy Using Extracted 3D Features from OCT Images. Sensors 2022, Vol 22, Page 7833 [Internet]. Multidisciplinary Digital Publishing Institute; 2022 [cited 2025 Oct 4];22:7833. https://doi.org/10.3390/S22207833\u003c/li\u003e\n\u003cli\u003eRyu G, Lee K, Park D, Park SH, Sagong M. A deep learning model for identifying diabetic retinopathy using optical coherence tomography angiography. Sci Rep [Internet]. Nature Research; 2021 [cited 2025 Oct 4];11:1\u0026ndash;9. https://doi.org/10.1038/S41598-021-02479-6;SUBJMETA\u003c/li\u003e\n\u003cli\u003eAziz A, Tezel NS, Ka\u0026ccedil;maz S, Attallah Y. Early Diabetic Retinopathy Detection from OCT Images Using Multifractal Analysis and Multi-Layer Perceptron Classification. Diagnostics 2025, Vol 15, Page 1616 [Internet]. Multidisciplinary Digital Publishing Institute; 2025 [cited 2025 Oct 4];15:1616. https://doi.org/10.3390/DIAGNOSTICS15131616\u003c/li\u003e\n\u003cli\u003eWroblewski JJ, Sanchez-Buenfil E, Inciarte M, Berdia J, Blake L, Wroblewski S, et al. Diabetic Retinopathy Screening Using Smartphone-Based Fundus Photography and Deep-Learning Artificial Intelligence in the Yucatan Peninsula: A Field Study. J Diabetes Sci Technol [Internet]. SAGE PublicationsSage CA: Los Angeles, CA; 2025 [cited 2025 Oct 4];19:370\u0026ndash;6. https://doi.org/10.1177/19322968231194644\u003c/li\u003e\n\u003cli\u003eKarakaya M, Hacisoftaoglu RE. Comparison of smartphone-based retinal imaging systems for diabetic retinopathy detection using deep learning. BMC Bioinformatics [Internet]. BioMed Central; 2020 [cited 2025 Oct 4];21:1\u0026ndash;18. https://doi.org/10.1186/S12859-020-03587-2/TABLES/3\u003c/li\u003e\n\u003cli\u003eSosale B, Sosale A, Murthy H, Sengupta S, Naveenam M. Medios-An offline, smartphone-based artificial intelligence algorithm for the diagnosis of diabetic retinopathy. Indian J Ophthalmol [Internet]. Wolters Kluwer Medknow Publications; 2020 [cited 2025 Oct 4];68:391\u0026ndash;5. https://doi.org/10.4103/IJO.IJO_1203_19\u003c/li\u003e\n\u003cli\u003eSosale B, Aravind SR, Murthy H, Narayana S, Sharma U, Gowda SGV, et al. Simple, Mobile-based Artificial Intelligence Algorithm in the detection of Diabetic Retinopathy (SMART) study. BMJ Open Diabetes Res Care [Internet]. American Diabetes Association; 2020 [cited 2025 Oct 4];8:892. https://doi.org/10.1136/BMJDRC-2019-000892\u003c/li\u003e\n\u003cli\u003eMalerbi FK, Andrade RE, Morales PH, Stuchi JA, Lencione D, de Paulo JV, et al. Diabetic Retinopathy Screening Using Artificial Intelligence and Handheld Smartphone-Based Retinal Camera. J Diabetes Sci Technol [Internet]. SAGE PublicationsSage CA: Los Angeles, CA; 2022 [cited 2025 Oct 4];16:716\u0026ndash;23. https://doi.org/10.1177/1932296820985567\u003c/li\u003e\n\u003cli\u003eAlqahtani AS, Alshareef WM, Aljadani HT, Hawsawi WO, Shaheen MH. The efficacy of artificial intelligence in diabetic retinopathy screening: a systematic review and meta-analysis. Int J Retina Vitreous [Internet]. BioMed Central Ltd; 2025 [cited 2025 Oct 4];11:1\u0026ndash;12. https://doi.org/10.1186/S40942-025-00670-9/FIGURES/8\u003c/li\u003e\n\u003cli\u003eNatarajan S, Jain A, Krishnan R, Rogye A, Sivaprasad S. Diagnostic Accuracy of Community-Based Diabetic Retinopathy Screening With an Offline Artificial Intelligence System on a Smartphone. JAMA Ophthalmol [Internet]. American Medical Association; 2019 [cited 2025 Oct 4];137:1182\u0026ndash;8. https://doi.org/10.1001/JAMAOPHTHALMOL.2019.2923\u003c/li\u003e\n\u003cli\u003eDow ER, Khan NC, Chen KM, Mishra K, Perera C, Narala R, et al. AI-Human Hybrid Workflow Enhances Teleophthalmology for the Detection of Diabetic Retinopathy. Ophthalmology Science [Internet]. Elsevier Inc.; 2023 [cited 2025 Oct 4];3. https://doi.org/10.1016/j.xops.2023.100330\u003c/li\u003e\n\u003cli\u003eZhang G, Lin JW, Wang J, Ji J, Cen LP, Chen W, et al. Automated multidimensional deep learning platform for referable diabetic retinopathy detection: a multicentre, retrospective study. BMJ Open [Internet]. British Medical Journal Publishing Group; 2022 [cited 2025 Oct 4];12:e060155. https://doi.org/10.1136/BMJOPEN-2021-060155\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance: executive summary. Https://IrisWhoInt/Bitstream/Handle/10665/350567/9789240037403-EngPdf. 2021;1\u0026ndash;148. \u003c/li\u003e\n\u003cli\u003eArtificial Intelligence for Health [Internet]. [cited 2025 Oct 5]. https://www.who.int/publications/m/item/artificial-intelligence-for-health. Accessed 5 Oct 2025\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u0026nbsp;\u003c/strong\u003eSearch strategies and inclusion and exclusion criteria\u0026nbsp;\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInclusion criteria\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 613px;\"\u003e\n \u003cul type=\"disc\"\u003e\n \u003cli\u003eStudies analyzing the use of artificial intelligence for automatic detection of new-onset diabetic retinopathy in adults worldwide\u003c/li\u003e\n \u003cli\u003ePublications between January 2 019 and September 2 025.\u003c/li\u003e\n \u003cli\u003eArticles in English, Spanish, or Portuguese.\u003c/li\u003e\n \u003cli\u003eScientific articles, systematic or scoping reviews, qualitative, quantitative, or mixed studies, and technical reports.\u003c/li\u003e\n \u003cli\u003eAdult patients over 18 years old.\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExclusion criteria\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 613px;\"\u003e\n \u003cul type=\"disc\"\u003e\n \u003cli\u003eEditorials, letters to the editor, opinions without empirical support, narrative reviews, case reports.\u003c/li\u003e\n \u003cli\u003eStudies published outside the established time frame.\u003c/li\u003e\n \u003cli\u003eExclusively clinical or technological studies that do not address information about the central topic.\u003c/li\u003e\n \u003cli\u003ePublications focused solely on clinical or surgical interventions, unrelated to the use of artificial intelligence.\u003c/li\u003e\n \u003cli\u003ePublications focused solely on classification, stratification, treatment, prognosis, and complications associated with diabetic retinopathy.\u003c/li\u003e\n \u003cli\u003ePatients with other conditions or comorbidities.\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"4\" style=\"width: 81px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSearch strategies\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePubMed\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 512px;\"\u003e\n \u003cp\u003e(\u0026quot;Diabetic Retinopathy\u0026quot;[Mesh] OR \u0026quot;diabetic retinopathy\u0026quot;[tiab] OR \u0026quot;retinal diabetic\u0026quot;[tiab] OR \u0026quot;retinal microaneurysm*\u0026quot;[tiab]) AND (\u0026quot;Artificial Intelligence\u0026quot;[Mesh] OR \u0026quot;Machine Learning\u0026quot;[Mesh] OR \u0026quot;Deep Learning\u0026quot;[Mesh] OR \u0026quot;Neural Networks (Computer)\u0026quot;[Mesh] OR \u0026quot;Algorithms\u0026quot;[Mesh] OR \u0026quot;artificial intelligence\u0026quot;[tiab] OR \u0026quot;machine learning\u0026quot;[tiab] OR \u0026quot;deep learning\u0026quot;[tiab] OR \u0026quot;neural network*\u0026quot;[tiab] OR \u0026quot;convolutional neural network*\u0026quot;[tiab] OR CNN[tiab] OR \u0026quot;support vector machine*\u0026quot;[tiab] OR SVM[tiab] OR \u0026quot;transfer learning\u0026quot;[tiab] OR \u0026quot;automated detection\u0026quot;[tiab] OR \u0026quot;automatic diagnos*\u0026quot;[tiab] OR \u0026quot;computer-aided diagnos*\u0026quot;[tiab] OR CAD[tiab]) AND (diagnos*[tiab] OR screen*[tiab] OR detect*[tiab] OR \u0026quot;early detection\u0026quot;[tiab]) NOT (pediatr*[tiab] OR child*[tiab] OR infant*[tiab]) AND (\u0026quot;2019/01/01\u0026quot;[Date - Publication] : \u0026quot;2025/09/30\u0026quot;[Date - Publication])\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eScopus\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 512px;\"\u003e\n \u003cp\u003eTITLE-ABS-KEY(\u0026quot;diabetic retinopathy\u0026quot; OR \u0026quot;retinopat\u0026iacute;a diab\u0026eacute;tica\u0026quot; OR \u0026quot;retinopatia diab\u0026eacute;tica\u0026quot;) AND TITLE-ABS-KEY(\u0026quot;artificial intelligence\u0026quot; OR \u0026quot;machine learning\u0026quot; OR \u0026quot;deep learning\u0026quot; OR \u0026quot;neural network*\u0026quot; OR \u0026quot;computer-assisted diagnosis\u0026quot; OR \u0026quot;computer-aided diagnosis\u0026quot; OR \u0026quot;automated detection\u0026quot; OR \u0026quot;automatic detection\u0026quot; OR \u0026quot;AI\u0026quot; OR \u0026quot;IA\u0026quot;) AND TITLE-ABS-KEY(adult OR adults OR \u0026quot;\u0026ge;18 years\u0026quot;) AND NOT TITLE-ABS-KEY(pediatr* OR child* OR infant*) AND TITLE-ABS-KEY(ophthalmology OR oftalmolog\u0026iacute;a OR \u0026quot;eye\u0026quot; OR \u0026quot;ocular\u0026quot; OR \u0026quot;retina\u0026quot;) AND NOT TITLE-ABS-KEY(treatment OR therapy OR prognosis OR pron\u0026oacute;stico OR complication* OR complicaci\u0026oacute;n*) AND PUBYEAR \u0026gt; 2018 AND PUBYEAR \u0026lt; 2026 AND (LIMIT-TO (DOCTYPE,\u0026quot;ar\u0026quot;) OR LIMIT-TO (DOCTYPE,\u0026quot;re\u0026quot;)) AND (LIMIT-TO (LANGUAGE,\u0026quot;English\u0026quot;) OR LIMIT-TO (LANGUAGE,\u0026quot;Spanish\u0026quot;) OR LIMIT-TO (LANGUAGE,\u0026quot;Portuguese\u0026quot;))\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWeb of Science\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 512px;\"\u003e\n \u003cp\u003e(diabetic retinopathy OR retinopat\u0026iacute;a diab\u0026eacute;tica OR retinopathia diab\u0026eacute;tica) (All Fields) and (artificial intelligence OR machine learning OR deep learning OR neural network* OR computer-aided diagnosis OR computer assisted diagnosis OR AI OR data mining) (All Fields) and (automatic detection OR detection OR diagnos* OR screening OR identificat*) (All Fields) not (pediatr* OR child* OR infant* OR ni\u0026ntilde;o* OR infantil) (All Fields) and (ophthalmology OR oftalmolog\u0026iacute;a OR ocular OR eye OR retina) (All Fields) not (treatment OR therapy OR tratamiento OR pron\u0026oacute;stico OR prognosis OR complicaci\u0026oacute;n* OR complication*) (All Fields) and 2025 or 2024 or 2023 or 2022 or 2021 or 2020 (Publication Years) and Article or Review Article or Early Access (Document Types)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBIREME (LILACS + SciELO)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 512px;\"\u003e\n \u003cp\u003e(((tw:(\u0026quot;retinopat\u0026iacute;a diab\u0026eacute;tica\u0026quot;) OR mh:\u0026quot;Retinopat\u0026iacute;a Diab\u0026eacute;tica\u0026quot; OR tw:\u0026quot;diabetic retinopathy\u0026quot;)) AND ((tw:\u0026quot;inteligencia artificial\u0026quot; OR mh:\u0026quot;Inteligencia Artificial\u0026quot; OR tw:\u0026quot;artificial intelligence\u0026quot; OR tw:\u0026quot;machine learning\u0026quot; OR tw:\u0026quot;aprendizaje autom\u0026aacute;tico\u0026quot; OR tw:\u0026quot;deep learning\u0026quot; OR tw:\u0026quot;red neuronal\u0026quot; OR tw:\u0026quot;neural network*\u0026quot; OR tw:\u0026quot;support vector machine*\u0026quot; OR tw:\u0026quot;SVM\u0026quot; OR tw:\u0026quot;convolutional neural network*\u0026quot; OR tw:\u0026quot;CNN\u0026quot;)) AND ((tw:\u0026quot;detecci\u0026oacute;n\u0026quot; OR tw:\u0026quot;diagn\u0026oacute;stico\u0026quot; OR tw:\u0026quot;screening\u0026quot; OR tw:\u0026quot;automatic detection\u0026quot; OR tw:\u0026quot;computer-aided diagnosis\u0026quot; OR tw:\u0026quot;CAD\u0026quot;)) AND NOT (tw:pediatr* OR tw:ni\u0026ntilde;o* OR tw:infantil OR tw:child* OR tw:infant*) AND (tw:oftalmolog\u0026iacute;a OR mh:\u0026quot;Oftalmolog\u0026iacute;a\u0026quot; OR tw:ophthalmology OR tw:ocular OR tw:retina) AND NOT (tw:tratamiento OR tw:therapy OR tw:pron\u0026oacute;stico OR tw:prognosis OR tw:complicaci\u0026oacute;n* OR tw:complication*)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u0026nbsp;\u003c/strong\u003eSynthesis of the results according to the biomedical data\u0026nbsp;\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 122px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModality/Signal\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e# Articles\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAverage SE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAverage \u0026nbsp;SP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAverage ACC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAverage AUC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 117px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOther Metrics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eReferences\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 122px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRetinography\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e92.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e93.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 75px;\"\u003e\n \u003cp\u003e93.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e95%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 117px;\"\u003e\n \u003cp\u003eF1 Score: 95%\u0026nbsp;\u003cbr\u003e\u0026nbsp;Kappa: 70%\u0026nbsp;\u003cbr\u003e\u0026nbsp;Precision: 90%\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e[11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 122px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMultimodal models + comparison with ophthalmologists\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e90%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e91%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 75px;\"\u003e\n \u003cp\u003e90%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e95%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 117px;\"\u003e\n \u003cp\u003eF1 Score: 92%\u0026nbsp;\u003cbr\u003e\u0026nbsp;PPV: 39%\u0026nbsp;\u003cbr\u003e\u0026nbsp;NPV: 98%\u003c/p\u003e\n \u003cp\u003eKappa: 86%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e[48, 49, 50, 51 52, 53, 54, 55, 56, 57, 72, 73]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 122px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOCT y OCTA\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e88.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e92.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 75px;\"\u003e\n \u003cp\u003e91.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e88.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 117px;\"\u003e\n \u003cp\u003ePPV: 88.3%\u003c/p\u003e\n \u003cp\u003eF1 score: 94.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e[58, 59, 60, 61, 63, 64, 71]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 122px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eElectronic devices\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e90.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e87.4 %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 75px;\"\u003e\n \u003cp\u003e70.6 %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e86.6 %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 117px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003ePPV: 82.6 %\u003c/p\u003e\n \u003cp\u003eNPV: 93.9 %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 101px;\"\u003e\n \u003cp\u003e[65, 66, 67, 68, 69, 71].\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003csup\u003eSE: sensitivity; SP: specificity; ACC: accuracy; AUC: area under the curve; PPV: positive predictive value; NPV: negative predictive value\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u0026nbsp;\u003c/strong\u003eBenefits and limitations of using artificial intelligence in the diagnosis of diabetic retinopathy\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 108px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCategory\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDescription\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eReferences\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"4\" style=\"width: 108px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBenefits\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003eSuperior SE and SP metrics support AI as a reliable tool for screening and clinical support.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[11, 19, 21, 26, 34, 39, 54, 59]\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003eAutomating image analysis allows large volumes to be processed in less time, reducing the ophthalmologist\u0026apos;s workload and promoting early detection programs.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[20, 22, 25, 30, 55, 56, 72]\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003ePortable, low-cost systems facilitate mass screening in rural and primary care settings, expanding coverage in populations with limited resources.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[24, 51, 57, 68, 69, 71] \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003eModels using techniques such as Grad-CAM allow visualization of the regions of the image used in the prediction, strengthening clinical confidence.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[16, 43, 44, 58]\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"5\" style=\"width: 108px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLimitations\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003eAlgorithms require high-quality images; variability between devices and capture conditions affects SE and SP in real-world practice.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[17, 21, 22, 28, 54, 59, 72]\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003eSome studies are validated in public databases, with limited multicenter evaluation or in diverse populations.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[11, 23, 42, 45, 55, 58, 63]\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003eHigh rates of false positives are reported, leading to unnecessary referrals and reducing SP, especially in mild cases or cases with other eye conditions.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[51, 52, 53, 55, 69, 72]\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003eThere are no uniform acquisition protocols; furthermore, it is difficult to detect early lesions or macular edema with a single image.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[15, 32, 34, 48, 68, 73]\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 390px;\"\u003e\n \u003cp\u003eTransparency and clinical validation of algorithms must be strengthened to ensure safe integration into healthcare practice.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 185px;\"\u003e\n \u003cp\u003e[18, 50, 59]\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003csup\u003eSE: sensitivity; SP: specificity; AI: artificial intelligence\u003c/sup\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Diabetic retinopathy, artificial intelligence, automation, early detection, imaging devices","lastPublishedDoi":"10.21203/rs.3.rs-8071096/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8071096/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eIntroduction:\u003c/h2\u003e\u003cp\u003eDiabetic retinopathy (DR) requires early diagnosis, as it is a common complication of diabetes mellitus (DM) and a leading cause of blindness worldwide. Artificial intelligence (AI) is emerging as a promising tool for detecting DR, as it can analyse large volumes of data with high accuracy (ACC).\u003c/p\u003e\u003ch2\u003eMethodology:\u003c/h2\u003e\u003cp\u003eA scoping review of the literature from 2019\u0026ndash;2025 was conducted in four databases, and inclusion and exclusion criteria were applied for the selection of articles. Data extraction was performed, and a synthesis of the results was generated.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eSixty-two articles were reviewed, mostly experimental studies (43%) and deep learning (DL) models (82%). AI was mainly applied to retinography (59%) and other images, such as optical coherence tomography (OCT) and optical coherence tomography angiography (OCTA) (11.3%). It demonstrated high sensitivity (SE) and specificity (SP) and highlighted benefits such as mass screening capacity. Limitations, such as image quality and variability between devices, have also been identified, but there are future opportunities to revolutionise the diagnosis of DR.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eAI is emerging as an effective tool for the de novo diagnosis of DR, with retinography as the main technique and other methods as options with great potential. Its benefits include automation, greater coverage, and a reduced healthcare burden, reinforcing its diagnostic value. However, some technical and validation challenges need to be overcome before these methods can be implemented in clinical practice.\u003c/p\u003e","manuscriptTitle":"Artificial Intelligence for the Diagnosis of De Novo Diabetic Retinopathy: A Scoping Review of Global Evidence","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-11 00:43:34","doi":"10.21203/rs.3.rs-8071096/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4dd3f647-aca2-455f-960c-020fc9b18711","owner":[],"postedDate":"December 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":59352950,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":59352951,"name":"Health sciences/Diseases"},{"id":59352952,"name":"Health sciences/Health care"},{"id":59352953,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2025-12-20T17:08:43+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-11 00:43:34","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8071096","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8071096","identity":"rs-8071096","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}