Diabetes mellitus (DM), a prevalent metabolic disorder, has significant global health implications. The advent of machine learning (ML) has revolutionized the ability to predict and manage diabetes early, offering new avenues to mitigate its impact. This systematic review examined 53 articles on ML applications for diabetes prediction, focusing on datasets, algorithms, training methods, and evaluation metrics. Various datasets, such as the Singapore National Diabetic Retinopathy Screening Program, REPLACE-BG, National Health and Nutrition Examination Survey (NHANES), and Pima Indians Diabetes Database (PIDD), have been explored, highlighting their unique features and challenges, such as class imbalance. This review assesses the performance of various ML algorithms, such as Convolutional Neural Networks (CNN), Support Vector Machines (SVM), Logistic Regression, and XGBoost, for the prediction of diabetes outcomes from multiple datasets. In addition, it explores explainable AI (XAI) methods such as Grad-CAM, SHAP, and LIME, which improve the transparency and clinical interpretability of AI models in assessing diabetes risk and detecting diabetic retinopathy. Techniques such as cross-validation, data augmentation, and feature selection are discussed in terms of their influence on the versatility and robustness of the model. Some evaluation techniques involving k-fold cross-validation, external validation, and performance indicators such as accuracy, area under curve, sensitivity, and specificity are presented. The findings highlight the usefulness of ML in addressing the challenges of diabetes prediction, the value of sourcing different data types, the need to make models explainable, and the need to keep models clinically relevant. This study highlights significant implications for healthcare professionals, policymakers, technology developers, patients, and researchers, advocating interdisciplinary collaboration and ethical considerations when implementing ML-based diabetes prediction models. By consolidating existing knowledge, this SLR outlines future research directions aimed at improving diagnostic accuracy, patient care, and healthcare efficiency through advanced ML applications. This comprehensive review contributes to the ongoing efforts to utilize artificial intelligence technology for a better prediction of diabetes, ultimately aiming to reduce the global burden of this widespread disease.

Artificial intelligence (AI) has gained significant attention in ophthalmology. This paper reviews, classifies, and summarizes the research literature in this field and aims to provide readers with a detailed understanding of the current status and future directions, laying a solid foundation for further research and decision-making.

Literature was retrieved from the Web of Science database. Bibliometric analysis was performed using VOSviewer, CiteSpace, and the R package Bibliometrix.

The study included 3,377 publications from 4,035 institutions in 98 countries. China and the United States had the most publications. Sun Yat-sen University is a leading institution. Translational Vision Science & Technology"published the most articles, while "Ophthalmology" had the most co-citations. Among 13,145 researchers, Ting DSW had the most publications and citations. Keywords included "Deep learning," "Diabetic retinopathy," "Machine learning," and others.

The study highlights the promising prospects of AI in ophthalmology. Automated eye disease screening, particularly its core technology of retinal image segmentation and recognition, has become a research hotspot. AI is also expanding to complex areas like surgical assistance, predictive models. Multimodal AI, Generative Adversarial Networks, and ChatGPT have driven further technological innovation. However, implementing AI in ophthalmology also faces many challenges, including technical, regulatory, and ethical issues, and others. As these challenges are overcome, we anticipate more innovative applications, paving the way for more effective and safer eye disease treatments.

The field of ophthalmology is rapidly advancing, with technological innovations enhancing the diagnosis and management of eye diseases. Tele-ophthalmology, or the use of telemedicine for ophthalmology, has emerged as a promising solution to improve access to eye care services, particularly for patients in remote or underserved areas. Despite its potential benefits, tele-ophthalmology faces significant challenges, including the need for high volumes of medical images to be analyzed and interpreted by trained clinicians. Artificial intelligence (AI) has emerged as a powerful tool in ophthalmology, capable of assisting clinicians in diagnosing and treating a variety of conditions. Integrating AI models into existing tele-ophthalmology infrastructure has the potential to revolutionize eye care services by reducing costs, improving efficiency, and increasing access to specialized care. By automating the analysis and interpretation of clinical data and medical images, AI models can reduce the burden on human clinicians, allowing them to focus on patient care and disease management. Available literature on the current status of tele-ophthalmology in Canada and successful AI models in ophthalmology was acquired and examined using the Arksey and O'Malley framework. This review covers literature up to 2022 and is split into 3 sections: 1) existing Canadian tele-ophthalmology infrastructure, with its benefits and drawbacks; 2) preeminent AI models in ophthalmology, across a variety of ocular conditions; and 3) bridging the gap between Canadian tele-ophthalmology and AI in a safe and effective manner.

Diabetes mellitus (DM) increases the risk of vascular complications, and retinal vasculature imaging serves as a valuable indicator of both microvascular and macrovascular health. Moreover, artificial intelligence (AI)-enabled systems developed for high-throughput detection of diabetic retinopathy (DR) using digitized retinal images have become clinically adopted. This study reviews AI applications using retinal images for DM-related complications, highlighting advancements beyond DR screening, diagnosis, and prognosis, and addresses implementation challenges, such as ethics, data privacy, equitable access, and explainability.

We conducted a thorough literature search across several databases, including PubMed, Scopus, and Web of Science, focusing on studies involving diabetes, the retina, and artificial intelligence. We reviewed the original research based on their methodology, AI algorithms, data processing techniques, and validation procedures to ensure a detailed analysis of AI applications in diabetic retinal imaging.

Retinal images can be used to diagnose DM complications including DR, neuropathy, nephropathy, and atherosclerotic cardiovascular disease, as well as to predict the risk of cardiovascular events. Beyond DR screening, AI integration also offers significant potential to address the challenges in the comprehensive care of patients with DM.

With the ability to evaluate the patient's health status in relation to DM complications as well as risk prognostication of future cardiovascular complications, AI-assisted retinal image analysis has the potential to become a central tool for modern personalized medicine in patients with DM.

Diabetic retinopathy (DR), the prevalence of which continues to rise, is one of the most common causes of vision loss worldwide. Experimental and clinical research in recent years has contributed to a better understanding of the pathogenesis of DR, which is complex and results from many interrelated processes leading to abnormal permeability and occlusion of the retinal vasculature, with ischemia and subsequent neovascularization. According to the absence or presence of neovascularization, DR is divided into two main forms: nonproliferative and proliferative DR. From nonproliferative to proliferative disease, diabetic macular edema (DME) can develop anywhere along the spectrum. As the majority of diabetics have no ophthalmologic symptoms, screening plays an important role in preventing the development of retinal disease. Specific treatment options beyond metabolic risk factor control, including intravitreal administration of anti-vascular endothelial growth factor (VEGF) agents or corticosteroids, laser photocoagulation, and vitreous surgery, are effective approaches for ocular diabetic complications.

A significant area of computer science called artificial intelligence (AI) is successfully applied to the analysis of intricate biological data and the extraction of substantial associations from datasets for a variety of biomedical uses. AI has attracted significant interest in biomedical research due to its features: (i) better patient care through early diagnosis and detection; (ii) enhanced workflow; (iii) lowering medical errors; (v) lowering medical costs; (vi) reducing morbidity and mortality; (vii) enhancing performance; (viii) enhancing precision; and (ix) time efficiency. Quantitative metrics are crucial for evaluating AI implementations, providing insights, enabling informed decisions, and measuring the impact of AI-driven initiatives, thereby enhancing transparency, accountability, and overall impact. The implementation of AI in biomedical fields faces challenges such as ethical and privacy concerns, lack of awareness, technology unreliability, and professional liability. A brief discussion is given of the AI techniques, which include Virtual screening (VS), DL, ML, Hidden Markov models (HMMs), Neural networks (NNs), Generative models (GMs), Molecular dynamics (MD), and Structure-activity relationship (SAR) models. The study explores the application of AI in biomedical fields, highlighting its enhanced predictive accuracy, treatment efficacy, diagnostic efficiency, faster decision-making, personalised treatment strategies, and precise medical interventions.

Our aim is to develop a deep learning-based system for automatically identifying and classifying benign and malignant tumours of the eyelid to improve diagnostic accuracy and efficiency.

The dataset includes photographs of normal eyelids, benign and malignant eyelid tumours and was randomly divided into a training and validation dataset in a ratio of 8:2. We used the training dataset to train eight convolutional neural network models to classify normal eyelids, benign and malignant eyelid tumours. These models included VGG16, ResNet50, Inception-v4, EfficientNet-V2-M and their variants. The validation dataset was used to evaluate and compare the performance of the different deep learning models.

All eight models achieved an average accuracy greater than 0.746 for identifying normal eyelids, benign and malignant eyelid tumours, with an average sensitivity and specificity exceeding 0.790 and 0.866, respectively. The mean area under the receiver operating characteristic curve (AUC) for the eight models was more than 0.904 in correctly identifying normal eyelids, benign and malignant eyelid tumours. The dual-path Inception-v4 network demonstrated the highest performance, with an AUC of 0.930 (95% CI 0.900 to 0.954) and an F1-score of 0.838 (95% CI 0.787 to 0.882).

The deep learning-based system shows significant potential in improving the diagnosis of eyelid tumours, providing a reliable and efficient tool for clinical practice. Future work will validate the model with more extensive and diverse datasets and integrate it into clinical workflows for real-time diagnostic support.

This study validated the artificial intelligence (AI)-based algorithm LuxIA for screening more-than-mild diabetic retinopathy (mtmDR) from a single 45° colour fundus image of patients with diabetes mellitus (DM, type 1 or type 2) in Spain. Secondary objectives included validating LuxIA according to the International Clinical Diabetic Retinopathy (ICDR) classification and comparing its performance between different devices.

In this multicentre, cross-sectional study, retinal colour fundus images of adults (≥18 years) with DM were collected from five hospitals in Spain (December 2021-December 2022). 45° colour fundus photographs were captured using non-mydriatic Topcon and ZEISS cameras. The Discovery platform (RetinAI) was used to collect images. LuxIA output was an ordinal score (1-5), indicating a classification as mtmDR based on an ICDR severity score.

945 patients with DM were included; the mean (SD) age was 64.6 (13.5) years. The LuxIA algorithm detected mtmDR with a sensitivity and specificity of 97.1% and 94.8%, respectively. The area under the receiver-operating characteristic curve was 0.96, indicating a high test accuracy. The 95% CI data for overall accuracy (94.8% to 95.6%), sensitivity (96.8% to 98.2%) and specificity (94.3% to 95.1%) indicated robust estimations by LuxIA, which maintained a concordance of classification (N=829, kappa=0.837, p=0.001) when used to classify Topcon images. LuxIA validation on ZEISS-obtained images demonstrated high accuracy (90.6%), specificity (92.3%) and lower sensitivity (83.3%) as compared with Topcon-obtained images.

AI algorithms such as LuxIA are increasing testing feasibility for healthcare professionals in DR screening. This study validates the real-world utility of LuxIA for mtmDR screening.

As the most common complication of diabetes, diabetic retinopathy (DR) is one of the main causes of irreversible blindness. Automatic DR grading plays a crucial role in early diagnosis and intervention, reducing the risk of vision loss in people with diabetes. In these years, various deep-learning approaches for DR grading have been proposed. Most previous DR grading models are trained using the dataset of single-field fundus images, but the entire retina cannot be fully visualized in a single field of view. There are also problems of scattered location and great differences in the appearance of lesions in fundus images. To address the limitations caused by incomplete fundus features, and the difficulty in obtaining lesion information. This work introduces a novel multi-view DR grading framework, which solves the problem of incomplete fundus features by jointly learning fundus images from multiple fields of view. Furthermore, the proposed model combines multi-view inputs such as fundus images and lesion snapshots. It utilizes heterogeneous convolution blocks (HCB) and scalable self-attention classes (SSAC), which enhance the ability of the model to obtain lesion information. The experimental results show that our proposed method performs better than the benchmark methods on the large-scale dataset.

Artificial intelligence (AI) is rapidly advancing in healthcare, enhancing the efficiency and effectiveness of services across various specialties, including cardiology, ophthalmology, dermatology, emergency medicine, etc. AI applications have significantly improved diagnostic accuracy, treatment personalization, and patient outcome predictions by leveraging technologies such as machine learning, neural networks, and natural language processing. However, these advancements also introduce substantial ethical and fairness challenges, particularly related to biases in data and algorithms. These biases can lead to disparities in healthcare delivery, affecting diagnostic accuracy and treatment outcomes across different demographic groups. This review paper examines the integration of AI in healthcare, highlighting critical challenges related to bias and exploring strategies for mitigation. We emphasize the necessity of diverse datasets, fairness-aware algorithms, and regulatory frameworks to ensure equitable healthcare delivery. The paper concludes with recommendations for future research, advocating for interdisciplinary approaches, transparency in AI decision-making, and the development of innovative and inclusive AI applications.

The eye provides novel insights into general health, as well as pathogenesis and development of systemic diseases. In the past decade, growing evidence has demonstrated that the eye's structure and function mirror multiple systemic health conditions, especially in cardiovascular diseases, neurodegenerative disorders, and kidney impairments. This has given rise to the field of oculomics-the application of ophthalmic biomarkers to understand mechanisms, detect and predict disease. The development of this field has been accelerated by three major advances: 1) the availability and widespread clinical adoption of high-resolution and non-invasive ophthalmic imaging ("hardware"); 2) the availability of large studies to interrogate associations ("big data"); 3) the development of novel analytical methods, including artificial intelligence (AI) ("software"). Oculomics offers an opportunity to enhance our understanding of the interplay between the eye and the body, while supporting development of innovative diagnostic, prognostic, and therapeutic tools. These advances have been further accelerated by developments in AI, coupled with large-scale linkage datasets linking ocular imaging data with systemic health data. Oculomics also enables the detection, screening, diagnosis, and monitoring of many systemic health conditions. Furthermore, oculomics with AI allows prediction of the risk of systemic diseases, enabling risk stratification, opening up new avenues for prevention or individualized risk prediction and prevention, facilitating personalized medicine. In this review, we summarise current concepts and evidence in the field of oculomics, highlighting the progress that has been made, remaining challenges, and the opportunities for future research.

This study aimed to develop and evaluate a deep learning model for grading foveal hypoplasia using retinal fundus images.

This retrospective study included patients with foveal developmental disorders, using color fundus images and optical coherence tomography scans taken between January 1, 2001, and August 31, 2021. In total, 605 retinal fundus images were obtained from 303 patients (male, 55.1%; female, 44.9%). After augmentation, the training, validation, and testing data sets comprised 1229, 527, and 179 images, respectively. A deep learning model was developed for binary classification (normal vs. abnormal foveal development) and six-grade classification of foveal hypoplasia. The outcome was compared with those by senior and junior clinicians.

Higher grade of foveal hypoplasia showed worse visual outcomes (P < 0.001). The binary classification achieved a best testing accuracy of 84.36% using the EfficientNet_b1 model, with 84.51% sensitivity and 84.26% specificity. The six-grade classification achieved a best testing accuracy of 78.21% with the model. The model achieved an area under the receiver operating characteristic curve (AUROC) of 0.9441 and an area under the precision-recall curve (AUPRC) of 0.9654 (both P < 0.0001) in the validation set and an AUROC of 0.8777 and an AUPRC of 0.8327 (both P < 0.0001) in the testing set. Compared to junior and senior clinicians, the EfficientNet_b1 model exhibited a superior performance in both binary and six-grade classification (both P < 0.00001).

The deep learning model in this study proved more efficient and accurate than assessments by junior and senior clinicians for identifying foveal developmental diseases in retinal fundus images. With the aid of the model, we were able to accurately assess patients with foveal developmental disorders.

This study strengthened the importance for a pediatric deep learning system to support clinical evaluation, particularly in cases reliant on retinal fundus images.

Existing deep learning (DL) algorithms lack the capability to accurately identify patients in immediate need of treatment for proliferative diabetic retinopathy (PDR). We aimed to develop a DL segmentation model to detect active PDR in six-field retinal images by the annotation of new retinal vessels and preretinal hemorrhages.

We identified six-field retinal images classified at level 4 of the International Clinical Diabetic Retinopathy Disease Severity Scale collected at the Island of Funen from 2009 to 2019 as part of the Danish screening program for diabetic retinopathy (DR). A certified grader (grader 1) manually dichotomized the images into active or inactive PDR, and the images were then reassessed by two independent certified graders. In cases of disagreement, the final classification decision was made in collaboration between grader 1 and one of the secondary graders. Overall, 637 images were classified as active PDR. We then applied our pre-established DL segmentation model to annotate nine lesion types before training the algorithm. The segmentations of new vessels and preretinal hemorrhages were corrected for any inaccuracies before training the DL algorithm. After the classification and pre-segmentation phases the images were divided into training (70%), validation (10%), and testing (20%) datasets. We added 301 images with inactive PDR to the testing dataset.

We included 637 images of active PDR and 301 images of inactive PDR from 199 individuals. The training dataset had 1381 new vessel and preretinal hemorrhage lesions, while the validation dataset had 123 lesions and the testing dataset 374 lesions. The DL system demonstrated a sensitivity of 90% and a specificity of 70% for annotation-assisted classification of active PDR. The negative predictive value was 94%, while the positive predictive value was 57%.

Our DL segmentation model achieved excellent sensitivity and acceptable specificity in distinguishing active from inactive PDR.

While many deep learning systems (DLSs) for diabetic retinopathy (DR) have been developed and validated on cohorts with an average age of 50s or older, fewer studies have examined younger individuals. This study aimed to understand DLS performance for younger individuals, who tend to display anatomic differences, such as prominent retinal sheen. This sheen can be mistaken for exudates or cotton wool spots, and potentially confound DLSs.

This was a prospective cross-sectional cohort study in a "Diabetes of young" clinic in India, enrolling 321 individuals between ages 18 and 45 (98.8% with type 1 diabetes). Participants had fundus photographs taken and the photos were adjudicated by experienced graders to obtain reference DR grades. We defined a younger cohort (age 18-25) and an older cohort (age 26-45) and examined differences in DLS performance between the two cohorts. The main outcome measures were sensitivity and specificity for DR.

Eye-level sensitivity for moderate-or-worse DR was 97.6% [95% confidence interval (CI) 91.2, 98.2] for the younger cohort and 94.0% [88.8, 98.1] for the older cohort (p = 0.418 for difference). The specificity for moderate-or-worse DR significantly differed between the younger and older cohorts, 97.9% [95.9, 99.3] and 92.1% [87.6, 96.0], respectively (p = 0.008). Similar trends were observed for diabetic macular edema (DME); sensitivity was 79.0% [57.9, 93.6] for the younger cohort and 77.5% [60.8, 90.6] for the older cohort (p = 0.893), whereas specificity was 97.0% [94.5, 99.0] and 92.0% [88.2, 95.5] (p = 0.018). Retinal sheen presence (94% of images) was associated with DME presence (p < 0.0001). Image review suggested that sheen presence confounded reference DME status, increasing noise in the labels and depressing measured sensitivity. The gradability rate for both DR and DME was near-perfect (99% for both).

DLS-based DR screening performed well in younger individuals aged 18-25, with comparable sensitivity and higher specificity compared to individuals aged 26-45. Sheen presence in this cohort made identification of DME difficult for graders and depressed measured DLS sensitivity; additional studies incorporating optical coherence tomography may improve accuracy of measuring DLS DME sensitivity.

To evaluate the acceptability, applicability, and cost-utility of AI-powered portable fundus cameras for diabetic retinopathy (DR) screening in Hong Kong, providing a viable alternative screening solution for resource-limited areas.

This pragmatic trial conducted in an optometric clinic and two optical shops. A self-testing system was used, integrating a portable fundus camera and AI software that automatically identified DR. Three months following the screening, selected participants were invited to complete an open-ended questionnaire.

A total of 316 subjects participated, with age of 60.80 ± 8.30 years. The success rate of the self-testing system without active assistance was 89 %. Among 61 subjects who completed follow-up interview, a majority agreed that the system and report were easy to follow and understand (85.3 % and 75.4 %). The satisfaction rate was 64 %, and the willingness to use again was 80 %. The AI screening showed a cost saving of 6312.92 USD per QALY, while the adjusted AI model saved 18639. AI screening and adjusted model outperformed traditional screening (Net Monetary Benefit 367,863.31 and 354,904.76 vs 339,919.83 USD).

The AI-powered portable fundus camera demonstrated high acceptability and applicability in real-world settings, suggesting that AI screening could be a viable alternative in resource-limited settings.

To review and evaluate the current literature on the application and impact of large language models (LLMs) in the field of ophthalmology, focusing on studies published in high-ranking ophthalmology journals.

This is a retrospective review of published articles.

This study did not involve human participation.

Articles published in the first quartile (Q1) of ophthalmology journals on Scimago Journal & Country Rank discussing different LLMs up to June 7, 2024, were reviewed, parsed, and analyzed.

All available articles were parsed and analyzed, which included the article and author characteristics and data regarding the LLM used and its applications, focusing on its use in medical education, clinical assistance, research, and patient education.

There were 35 Q1-ranked journals identified, 19 of which contained articles discussing LLMs, with 101 articles eligible for review. One-third were original investigations (32%; 32/101), with an average of 5.3 authors per article. The United States (50.4%; 51/101) was the most represented country, followed by the United Kingdom (25.7%; 26/101) and Canada (16.8%; 17/101). ChatGPT was the most used LLM among the studies, with different versions discussed and compared. Large language model applications were discussed relevant to their implications in medical education, clinical assistance, research, and patient education.

The numerous publications on the use of LLM in ophthalmology can provide valuable insights for stakeholders and consumers of these applications. Large language models present significant opportunities for advancement in ophthalmology, particularly in team science, education, clinical assistance, and research. Although LLMs show promise, they also show challenges such as performance inconsistencies, bias, and ethical concerns. The study emphasizes the need for ongoing artificial intelligence improvement, ethical guidelines, and multidisciplinary collaboration.

The author(s) have no proprietary or commercial interest in any materials discussed in this article.

The sensitivity of reverse-transcription polymerase chain reaction (RT-PCR) is limited for diagnosis of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Chest computed tomography (CT) is reported to have high sensitivity; however, given the limited availability of chest CT during a pandemic, the assessment of more readily available imaging, such as chest radiographs, augmented by artificial intelligence may substitute for the detection of the features of coronavirus disease 2019 (COVID-19) pneumonia.

We trained a deep convolutional neural network to detect SARS-CoV-2 pneumonia using publicly available chest radiography imaging data including 8,851 normal, 6,045 pneumonia, and 200 COVID-19 pneumonia radiographs. The entire cohort was divided into training (n = 13,586) and test groups (n = 1510). We assessed the accuracy of prediction with independent external data.

The sensitivity and positive predictive values of the assessment by artificial intelligence were 96.8% and 90.9%, respectively. In the first external validation of 204 chest radiographs among 107 patients with confirmed COVID-19, the artificial intelligence algorithm correctly identified 174 (85%) chest radiographs as COVID-19 pneumonia among 97 (91%) patients. In the second external validation with 50 immunocompromised patients with leukemia, the higher probability of the artificial intelligence assessment for COVID-19 was correlated with suggestive features of COVID-19, while the normal chest radiographs were closely correlated with the likelihood of normal chest radiographs by the artificial intelligence prediction.

The assessment method by artificial intelligence identified suspicious lung lesions on chest radiographs. This novel approach can identify patients for confirmatory chest CT before the progression of COVID-19 pneumonia.

While the availability of public internet-scale datasets of images and language has catalyzed remarkable progress in machine learning, medical datasets are constrained by regulations protecting patient privacy and the time and cost required for curation and labeling. Self-supervised learning or pretraining has demonstrated great success in learning meaningful representations from large unlabeled datasets to enable efficient learning on downstream tasks. In ophthalmology, the RETFound model, a large vision transformer (ViT-L) model trained by masked autoencoding on 1.6 million color fundus photos and OCT B-scans, is the first model pretrained at such scale for ophthalmology, demonstrating strong performance on downstream tasks from diabetic retinopathy grading to stroke detection. Here, we measure the label efficiency of the RETFound model in learning to identify normal vs. abnormal OCT B-scans obtained as part of a pilot study for primary care-based diabetic retinopathy screening in North Carolina.

The 1150 TopCon Maestro OCT central B-scans (981 normal and 169 abnormal) were randomly split 80/10/10 into training, validation, and test datasets. Model training and hyperparameter tuning were performed on the training set guided by validation set performance. The best performing models were then evaluated on the final test set.

Six hundred forty-seven patients with diabetes in the Duke Health System participating in primary care diabetic retinopathy screening contributed 1150 TopCon Maestro OCT central B-scans.

Three models (ResNet-50, ViT-L, and RETFound) were fine-tuned on the full training dataset of 915 OCT B-scans and on smaller training data subsets of 500, 250, 100, and 50 OCT B-scans, respectively, across 3 random seeds.

Mean accuracy, area under the receiver operator curve (AUROC), area under the precision recall curve (AUPRC), F1 score, precision, and recall on the final held-out test set were reported for each model.

Across 3 random seeds and all training dataset sizes, RETFound outperformed both ResNet-50 and ViT-L on all evaluation metrics on the final held-out test dataset. Large vision transformer and ResNet-50 performed comparably at the largest training dataset sizes of 915 and 500 OCT B-scans; however, ResNet-50 suffered more pronounced performance degradation at the smallest dataset sizes of 100 and 50 OCT B-scans.

Our findings validate the benefits of RETFound's additional retina-specific pretraining. Further research is needed to establish best practices for fine-tuning RETFound to downstream tasks.

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

The Mary Tyler Moore Vision Initiative (MTM Vision) honors Mary Tyler Moore's commitment to ending vision loss from diabetes. Founded by Moore's husband, Dr. S. Robert Levine, MTM Vision aims to accelerate breakthroughs in diabetic retinal disease (DRD). At the MTM Vision Symposium 2024 on Curing Vision Loss from Diabetes, experts highlighted the urgent need for updated DRD staging systems, clinically relevant endpoints, and novel biomarkers to detect early disease changes. MTM Vision is advancing two clinical trials in collaboration with the DRCR Retina Network, launching a public awareness campaign, and welcoming Boehringer Ingelheim as the first founding industry member of its pre-competitive Consortium. Speakers emphasized big-data strategies and artificial intelligence (AI)-driven tools to improve DRD diagnosis, risk prediction, and personalized treatment. They also showcased new efforts to bridge academic discoveries with industry expertise, illustrating promising work on vascular regeneration and cellular senescence that may yield future therapies. The MTM Vision Biorepository and Resource Center is expanding tissue collections, enabling multi-omics analyses to study DRD mechanisms. Patient voices were central to the discussion, with calls for enhanced patient-reported outcomes, caregiver support, and broader education on DRD's risks. The symposium also underscored the importance of integrating mental health, quality of life measures, and ongoing patient input to guide clinical research.

Improving the accessibility of screening diabetic kidney disease (DKD) and differentiating isolated diabetic nephropathy from non-diabetic kidney disease (NDKD) are two major challenges in the field of diabetes care. We aimed to develop and validate an artificial intelligence (AI) deep learning system to detect DKD and isolated diabetic nephropathy from retinal fundus images.

In this population-based study, we developed a retinal image-based AI-deep learning system, DeepDKD, pretrained using 734 084 retinal fundus images. First, for DKD detection, we used 486 312 retinal images from 121 578 participants in the Shanghai Integrated Diabetes Prevention and Care System for development and internal validation, and ten multi-ethnic datasets from China, Singapore, Malaysia, Australia, and the UK (65 406 participants) for external validation. Second, to differentiate isolated diabetic nephropathy from NDKD, we used 1068 retinal images from 267 participants for development and internal validation, and three multi-ethnic datasets from China, Malaysia, and the UK (244 participants) for external validation. Finally, we conducted two proof-of-concept studies: a prospective real-world study with 3 months' follow-up to evaluate the effectiveness of DeepDKD in screening DKD; and a longitudinal analysis of the effectiveness of DeepDKD in differentiating isolated diabetic nephropathy from NDKD on renal function changes with 4·6 years' follow-up.

For detecting DKD, DeepDKD achieved an area under the receiver operating characteristic curve (AUC) of 0·842 (95% CI 0·838-0·846) on the internal validation dataset and AUCs of 0·791-0·826 across external validation datasets. For differentiating isolated diabetic nephropathy from NDKD, DeepDKD achieved an AUC of 0·906 (0·825-0·966) on the internal validation dataset and AUCs of 0·733-0·844 across external validation datasets. In the prospective study, compared with the metadata model, DeepDKD could detect DKD with higher sensitivity (89·8% vs 66·3%, p<0·0001). In the longitudinal study, participants with isolated diabetic nephropathy and participants with NDKD identified by DeepDKD had a significant difference in renal function outcomes (proportion of estimated glomerular filtration rate decline: 27·45% vs 52·56%, p=0·0010).

Among diverse multi-ethnic populations with diabetes, a retinal image-based AI-deep learning system showed its potential for detecting DKD and differentiating isolated diabetic nephropathy from NDKD in clinical practice.

National Key R & D Program of China, National Natural Science Foundation of China, Beijing Natural Science Foundation, Shanghai Municipal Key Clinical Specialty, Shanghai Research Centre for Endocrine and Metabolic Diseases, Innovative research team of high-level local universities in Shanghai, Noncommunicable Chronic Diseases-National Science and Technology Major Project, Clinical Special Program of Shanghai Municipal Health Commission, and the three-year action plan to strengthen the construction of public health system in Shanghai.

To evaluate the capability of artificial intelligence (AI) in screening for diabetic retinopathy (DR) utilizing digital retinography captured by non-mydriatic (NM) ≥45° cameras, focusing on diagnosis accuracy, effectiveness, and clinical safety.

We performed an overview of systematic reviews (SRs) up to May 2023 in Medline, Embase, CINAHL, and Web of Science. We used AMSTAR-2 tool to assess the reliability of each SR. We reported meta-analysis estimates or ranges of diagnostic performance figures.

Out of 1336 records, ten SRs were selected, most deemed low or critically low quality. Eight primary studies were included in at least five of the ten SRs and 125 in less than five SRs. No SR reported efficacy, effectiveness, or safety outcomes. The sensitivity and specificity for referable DR were 68-100% and 20-100%, respectively, with an AUROC range of 88 to 99%. For detecting DR at any stage, sensitivity was 79-100%, and specificity was 50-100%, with an AUROC range of 93 to 98%.

AI demonstrates strong diagnostic potential for DR screening using NM cameras, with adequate sensitivity but variable specificity. While AI is increasingly integrated into routine practice, this overview highlights significant heterogeneity in AI models and the cameras used. Additionally, our study enlightens the low quality of existing systematic reviews and the significant challenge of integrating the rapidly growing volume of emerging evidence in this field. Policymakers should carefully evaluate AI tools in specific contexts, and future research must generate updated high-quality evidence to optimize their application and improve patient outcomes.

Visual impairment, stemming from genetic, degenerative, and traumatic causes, affects millions globally. Recent advancements in ophthalmology present novel strategies for managing and potentially reversing these conditions. Here, we explore 10 emerging avenues-including gene therapy, stem cell therapy, advanced imaging, novel therapeutics, nanotechnology, artificial intelligence (AI) and machine learning, teleophthalmology, optogenetics, bionics, and neuro-ophthalmology-all making strides to improve diagnosis, treatment, and vision restoration. Among these, gene therapy and stem cell therapy are revolutionizing the treatment of retinal degenerative diseases, while advanced imaging technologies enable early detection and personalized care. Therapeutic advancements like anti-vascular endothelial growth factor therapies and neuroprotective agents, along with nanotechnology, have improved clinical outcomes for multiple ocular conditions. AI, especially machine learning, is enhancing diagnostic accuracy, facilitating early detection, and personalized treatment strategies, particularly when integrated with advanced imaging technologies. Teleophthalmology, further strengthened by AI, is expanding access to care, particularly in underserved regions, whereas emerging technologies like optogenetics, bionics, and neuro-ophthalmology offer new hope for patients with severe vision impairment. In light of ongoing research, we summarize the current clinical landscape and the potential advantages of these innovations to revolutionize the management of visual impairments. Additionally, we address the challenges and limitations associated with these emerging avenues in ophthalmology, providing insights into their future trajectories in clinical practice. Continued advancements in these fields promise to reshape the landscape of ophthalmic care, ultimately improving the quality of life for individuals with visual impairments.

Machine learning technology has demonstrated significant potential in glaucoma research, particularly in early diagnosis, predicting disease progression, evaluating treatment responses, and developing personalized treatment strategies. The application of machine learning not only enhances the understanding of the pathological mechanism of glaucoma and optimizes the diagnostic process but also provides patients with accurate medical services.

This study aimed to describe the current state of research, highlight directions for further development, and identify potential trends for improvement. This review was conducted following the scoping review of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) extension to showcase advancements in the application of machine learning in glaucoma research and treatment.

We employed a comprehensive search strategy to retrieve literature from the Web of Science Core Collection database, ultimately including 3,581 articles in the analysis. Through data analysis, we identified current research hotspots, noted differences in researchers' attitudes and opinions, and predicted potential future development trends.

We divided the research topics into six categories, clearly identifying "eye diseases", "retinal fundus imaging" and "risk factors" as the key terms for the development of this field. These findings signify the promising prospects of machine learning, particularly when integrated with multimodal technologies and large language models, to enhance the diagnosis and treatment of glaucoma.

This study aimed to investigate diabetic patients' acceptance of artificial intelligence (AI) devices for diabetic retinopathy screening and the related influencing factors.

An integrated model was proposed, and structural equation modeling was used to evaluate items and construct reliability and validity via confirmatory factor analysis. The model's path effects, significance, goodness of fit, and mediation and moderation effects were analyzed.

Intention to Use (IU) is significantly affected by Subjective Norms (SN), Resistance Bias (RB), and Uniqueness Neglect (UN). Perceived Usefulness (PU) and Perceived Ease of Use (PEOU) were significant mediators between IU and other variables. The moderating effect of trust (TR) is non-significant on the path of PU to IU.

The significant positive impact of SN may be caused by China's collectivist and authoritarian cultures. Both PU and PEOU had a significant mediation effect, which suggests that impressions influence acceptance. Although the moderating effect of TR was not significant, the unstandardized factor loading remained positive in this study. We presume that this may be due to an insufficient sample size, and the public was unfamiliar with AI medical devices.

To develop artificial intelligence (AI) models for automated detection of center-involved diabetic macular edema (CI-DME) with visual impairment using color fundus photographs (CFP) and optical coherence tomography (OCT) scans.

AI effort using pooled data from multi-center studies.

Datasets consisted of diabetic participants with or without CI-DME, who had CFP, OCT, and best corrected visual acuity (BCVA) obtained after manifest refraction. The development dataset was from DRCR Retina Network clinical trials, external testing dataset 1 was from the Singapore National Eye Centre, Singapore, and external testing dataset 2 was from the Eye Clinic, IRCCS MultiMedica, Milan, Italy.

AI models were trained to detect CI-DME, visual impairment (BCVA 20/32 or worse), and CI-DME with visual impairment, using CFPs alone, OCTs alone, and both CFPs and OCTs together (multimodal). Data from 1,007 eyes were used to train and validate the algorithms, and data from 448 eyes were used for testing.

Area under the receiver operating characteristic curve (AUC) values.

In the primary testing set, the CFP model, OCT model, and multimodal model had AUCs of 0.848 (95% CI 0.787-0.900), 0.913 (95% CI 0.870-0.947), and 0.939 (95% CI 0.906-0.964), respectively, for detection of CI-DME with visual impairment. In external testing dataset 1, the CFP, OCT, and multimodal models had AUCs of 0.756 (95% CI 0.624-0.870), 0.949 (95% CI 0.889-0.989), and 0.917 (95% CI 0.837-0.979), respectively, for detection of CI-DME with visual impairment. In external testing dataset 2, the CFP, OCT, and multimodal models had AUCs of 0.881 (95% CI 0.822-0.940), 0.828 (95% CI 0.749-0.905), and 0.907 (95% CI 0.852-0.952), respectively, for detection of CI-DME with visual impairment.

The AI models showed good diagnostic performance for detection of CI-DME with visual impairment. The multimodal (CFP and OCT) model did not offer additional benefit over the OCT model alone. If validated in prospective studies, these AI models could potentially help to improve triage and detection of patients who require prompt treatment.

To examine the influence of artificial intelligence (AI) on physicians' judgments regarding the presence and severity of glaucoma on fundus photographs in an online simulation system.

Forty-five trainee and expert ophthalmologists independently evaluated 120 fundus photographs, including 30 photographs each from patients with no glaucoma, mild glaucoma, moderate glaucoma, and severe glaucoma. A second trial was conducted at least one week after the initial trial in which photograph presentation order was randomized. During the second trial, 30% of the glaucoma judgments made by the AI system were intentionally incorrect. The evaluators were asked about their thoughts on AI in ophthalmology via a 3-item questionnaire.

The percentage of correct responses for all images significantly improved (P < 0.001) from 48.4 ± 24.8% in the initial trial to 59.6 ± 20.3% in the second trial. The improvement in the correct response rate was significantly greater for trainees (14.2 ± 19.0%) than for experts (8.6 ± 11.4%) (P = 0.04). The correct response rate was 63.9 ± 20.6% when the AI response was correct, significantly greater than the 47.9 ± 26.6% when the AI response was incorrect (P < 0.0001). For trainees, the correct response rate was significantly greater when the AI's response was correct than when it was incorrect. However, for experts, the effect was less pronounced. The decision time was significantly longer when the AI response was incorrect than when it was correct (P = 0.003).

In fundus photography-based glaucoma detection, the results of AI systems can influence physicians' judgments, particularly those of physicians with less experience.

This paper aims to assess ChatGPT's performance in answering European Board of Ophthalmology Diploma (EBOD) examination papers and to compare these results to pass benchmarks and candidate results.

This cross-sectional study used a sample of past exam papers from 2012, 2013, 2020-2023 EBOD examinations. This study analyzed ChatGPT's responses to 440 multiple choice questions (MCQs), each containing five true/false statements (2200 statements in total) and 48 single best answer (SBA) questions.

ChatGPT, for MCQs, scored on average 64.39%. ChatGPT's strongest metric performance for MCQs was precision (68.76%). ChatGPT performed best at answering pathology MCQs (Grubbs test p < 0.05). Optics and refraction had the lowest-scoring MCQ performance across all metrics. ChatGPT-3.5 Turbo performed worse than human candidates and ChatGPT-4o on easy questions (75% vs. 100% accuracy) but outperformed humans and ChatGPT-4o on challenging questions (50% vs. 28% accuracy). ChatGPT's SBA performance averaged 28.43%, with the highest score and strongest performance in precision (29.36%). Pathology SBA questions were consistently the lowest-scoring topic across most metrics. ChatGPT demonstrated a nonsignificant tendency to select option 1 more frequently (p = 0.19). When answering SBAs, human candidates scored higher than ChatGPT in all metric areas measured.

ChatGPT performed stronger for true/false questions, scoring a pass mark in most instances. Performance was poorer for SBA questions, suggesting that ChatGPT's ability in information retrieval is better than that in knowledge integration. ChatGPT could become a valuable tool in ophthalmic education, allowing exam boards to test their exam papers to ensure they are pitched at the right level, marking open-ended questions and providing detailed feedback.

Fundus disease is a complex and universal disease involving a variety of pathologies. Its early diagnosis using fundus images can effectively prevent further diseases and provide targeted treatment plans for patients. Recent deep learning models for classification of this disease are gradually emerging as a critical research field, which is attracting widespread attention. However, in practice, most of the existing methods only focus on local visual cues of a single image, and ignore the underlying explicit interaction similarity between subjects and correlation information among pathologies in fundus diseases. In this paper, we propose a novel label-aware dual graph neural networks for multi-label fundus image classification that consists of population-based graph representation learning and pathology-based graph representation learning modules. Specifically, we first construct a population-based graph by integrating image features and non-image information to learn patient's representations by incorporating associations between subjects. Then, we represent pathologies as a sparse graph where its nodes are associated with pathology-based feature vectors and the edges correspond to probability of the co-occurrence of labels to generate a set of classifier scores by the propagation of multi-layer graph information. Finally, our model can adaptively recalibrate multi-label outputs. Detailed experiments and analysis of our results show the effectiveness of our method compared with state-of-the-art multi-label fundus image classification methods.

In this study, artificial intelligence (AI) was used to deeply learn the classification of the anterior segment-Optical Coherence Tomography (AS-OCT) images. This AI systems automatically analyzed the angular structure of the AS-OCT images and automatically classified anterior chamber angle. It would improve the efficiency of AS-OCT image analysis.

The subjects were from the glaucoma disease screening and prevention project for elderly people in Shanghai community. Each scan contained 72 cross-sectional AS-OCT frames. We developed a deep learning-based AS-OCT image automatic anterior chamber angle analysis software. Classifier performance was evaluated against glaucoma experts' grading of AS-OCT images as standard. Outcome evaluation included accuracy (ACC) and area under the receiver operator curve (AUC).

94895 AS-OCT images were collected from 687 participants, in which 69,243 images were annotated as open, 16,433 images were annotated as closed, and 9219 images were annotated as non-gradable. The class-balanced train data were formed from randomly extracting the same number of open angle images as the closed angle images, which contained 22,393 images (11127 open, 11256 closed). The best-performing classifier was developed by applying transfer learning to the ResNet-50 architecture. against experts' grading, this classifier achieved an AUC of 0.9635.

Deep learning classifiers effectively detect angle closure based on automated analysis of AS-OCT images. This system could be used to automate clinical evaluations of the anterior chamber angle and improve efficiency of interpreting AS-OCT images. The results demonstrated the potential of the deep learning system for rapid recognition of high-risk populations of PACD.

The burst of modern information has significantly promoted the development of global medicine into a new era of big data healthcare. Ophthalmology is one of the most prominent medical specialties driven by big data analytics. This study aimed to describe the development status and research hotspots of big data in ophthalmology.

English articles and reviews related to big data in ophthalmology published from January 1, 1999, to April 30, 2024, were retrieved from the Web of Science Core Collection. The relevant information was analyzed and visualized using VOSviewer and CiteSpace software.

A total of 406 qualified documents were included in the analysis. The annual number of publications on big data in ophthalmology reached a rapidly increasing stage since 2019. The United States (n = 147) led in the number of publications, followed by India (n = 77) and China (n = 69). The L.V. Prasad Eye Institute in India was the most productive institution (n = 50), and Anthony Vipin Das was the most influential author with the most relevant literature (n = 45). The electronic medical records were the primary source of ophthalmic big data, and artificial intelligence served as the principal analytics tool. Diabetic retinopathy, glaucoma, and myopia are currently the main topics of interest in this field.

The application of big data in ophthalmology has experienced rapid growth in recent years. Big data is expected to play an increasingly significant role in shaping the future of research and clinical practice in ophthalmology.

To assess the diagnostic capabilities of the most recent chatbots releases, GPT-4o and Gemini Advanced, facing different retinal diseases.

Exploratory analysis on 50 cases with different surgical (n = 27) and medical (n = 23) retinal pathologies, whose optical coherence tomography/angiography scans were dragged into ChatGPT and Gemini's interfaces. Then, the authors asked "Please describe this image" and classified the diagnosis as: 1) Correct; 2) Partially correct; 3) Wrong; 4) Unable to assess exam type; and 5) Diagnosis not given.

ChatGPT indicated the correct diagnosis in 31 of 50 cases (62%), significantly higher than Gemini Advanced in 16 of 50 cases ( P = 0.0048). In 24% of cases, Gemini Advanced was not able to produce any answer, stating "That's not something I'm able to do yet." For both, primary misdiagnosis was macular edema, given erroneously in 16% and 14% of cases, respectively. ChatGPT-4o showed higher rates of correct diagnoses either in surgical (52% vs. 30%) or in medical retina (78% vs. 43%). Notably, when presented without the corresponding structural image, in any case Gemini was able to recognize optical coherence tomography angiography scans, confusing images for artworks.

ChatGPT-4o outperformed Gemini Advanced in diagnostic accuracy facing optical coherence tomography/angiography images, even if the range of diagnoses is still limited.

Diabetic retinopathy (DR) is a leading cause of preventable blindness and has emerged as a global health challenge, necessitating the development of robust management strategies. As DR prevalence continues to rise, advancements in screening methods have become increasingly critical for timely detection and intervention. This review examines three key advancements in DR screening: a shift from specialist to generalist approach, the adoption of telemedicine strategies for expanded access and enhanced efficiency, and the integration of artificial intelligence (AI). In particular, AI offers unprecedented benefits in the form of sustainability and scalability for not only DR screening but other aspects of eye health and the medical field as a whole. Though there remain barriers to address, AI holds vast potential for reshaping DR screening and significantly improving patient outcomes globally.

The early diagnosis of retinal disorders is essential in preventing permanent or partial blindness. Identifying these conditions promptly guarantees early treatment and prevents blindness. However, the challenge lies in accurately diagnosing these conditions, especially with limited labeled data. This study aims to enhance the diagnostic accuracy of retinal disorders using a novel Dual-Branch Semi-Supervised Learning (DB-SSL) approach that leverages both labeled and unlabeled data for multi-class classification of eye diseases. Employing Color Fundus Photography (CFP), our research integrates a Convolutional Neural Network (CNN) that integrates features from two parallel branches. This framework effectively handles the complexity of ocular imaging by utilizing self-training-based semi-supervised learning to explore relationships within unlabeled data. We propose and evaluate six CNN models: ResNet50, DenseNet121, MobileNetV2, EfficientNetB0, SqueezeNet1_0, and a hybrid of ResNet50 and MobileNetV2 on their ability to classify four key eye conditions: cataract, diabetic retinopathy, glaucoma, and normal, using a large, diverse OIH dataset containing 4217 fundus images. Among the evaluated models, ResNet50 emerged as the most accurate, achieving 93.14 % accuracy on unseen data. The model demonstrates robustness with a sensitivity of 93 % and specificity of 98.37 %, along with a precision and F1 Score of 93 % each, and a Cohen's Kappa of 90.85 %. Additionally, it exhibits an AUC score of 97.75 % nearing perfection. Systematically removing certain components from the ResNet50 model further validates its efficacy. Our findings underscore the potential of advanced CNN architectures combined with semi-supervised learning in enhancing the accuracy of eye disease classification systems, particularly in resource-constrained environments where the procurement of large labeled datasets is challenging and expensive. This approach is well-suited for integration into Clinical Decision Support Systems (CDSS), providing valuable diagnostic assistance in real-world clinical settings.

To evaluate the use of an autonomous artificial intelligence (AI)-based device to screen for diabetic retinopathy (DR) and to evaluate the frequency of diabetes mellitus (DM) and DR in an under-resourced population served by the Stanford Belize Vision Clinic (SBVC).

The records of all patients from 2017 to 2024 were collected and analyzed, dividing the study into two time periods: Pre-AI (before June 2022, prior to the implementation of the LumineticsCore® device at SBVC) and Post-AI (from June 2022 to the present) and subdivided into post-COVID19 and pre-COVID19 periods. Patients were categorized based on self-reported past medical history (PMH) as DM positive (diagnosed DM) and DM negative (no PMH of DM). AI camera outcomes included: negative for more than mild DR (MTMDR), positive for MTMDR, and insufficient exam quality.

A total of 1897 patients with a mean age of 47.6 years were included. The gradability of encounters by the AI device was 89.1%. The frequency of DR detection increased significantly in the Post-AI period (55/639) compared to the Pre-AI period (38/1258), including during the COVID-19 pandemic. The mean age of DR diagnosis was significantly lower in the Post-AI period (44.1 years) compared to Pre-AI period (60.7 years) among DM negative patients. There was a significant association between having DR and hypertension. Additionally, the detection rate of DM increased in the Post-AI period compared to Pre-AI period.

Autonomous AI-based screening significantly improves the detection of patients with DR in areas with limited healthcare resources by reducing dependence on on-field ophthalmologists. This innovative approach can be seamlessly integrated into primary care settings, with technicians capturing images quickly and efficiently within just a few minutes. This study demonstrates the effectiveness of autonomous AI in identifying patients with both DR and DM, as well as associated high-burden diseases such as hypertension, across various age ranges.

This study aimed to identify risk factors for diabetic retinopathy (DR) and develop machine learning (ML)-based predictive models using routine laboratory data in patients with type 2 diabetes mellitus (T2DM).

Clinical data from 4259 T2DM inpatients at Beijing Tongren Hospital were analyzed, divided into a model construction data set (N = 3936) and an external validation data set (N = 323). Using 39 optimal variables, a prediction model was constructed using the eXtreme Gradient Boosting (XGBoost) algorithm and compared with four other algorithms: support vector machine (SVM), gradient boosting decision tree (GBDT), neural network (NN), and logistic regression (LR). The Shapley Additive exPlanation (SHAP) method was employed to interpret the XGBoost model. External validation was performed to assess model performance.

DR was present in 47.69% (N = 1877) of T2DM patients in the model construction data set. Among the models tested, the XGBoost model performed best with an AUC of 0.831, accuracy of 0.757, sensitivity of 0.754, specificity of 0.759, and F1-score of 0.752. SHAP explained feature importance for XGBoost model and identified key risk factors for DR. External validation yielded an accuracy of 0.650 for the XGBoost model.

The XGBoost-based prediction model effectively assesses DR risk in T2DM patients using routine laboratory data, aiding clinicians in identifying high-risk individuals and guiding personalized management strategies, especially in medically underserved areas.

In recent years, artificial intelligence and machine learning algorithms have been used more extensively to diagnose diabetic retinopathy and other diseases. Still, the effectiveness of these methods has not been thoroughly investigated. This study aimed to evaluate the performance and limitations of machine learning and deep learning algorithms in detecting diabetic retinopathy.

This study was conducted based on the PRISMA checklist. We searched online databases, including PubMed, Scopus, and Google Scholar, for relevant articles up to September 30, 2023. After the title, abstract, and full-text screening, data extraction and quality assessment were done for the included studies. Finally, a meta-analysis was performed.

We included 76 studies with a total of 1,371,517 retinal images, of which 51 were used for meta-analysis. Our meta-analysis showed a significant sensitivity and specificity with a percentage of 90.54 (95%CI [90.42, 90.66], P < 0.001) and 78.33% (95%CI [78.21, 78.45], P < 0.001). However, the AUC (area under curvature) did not statistically differ across studies, but had a significant figure of 0.94 (95% CI [- 46.71, 48.60], P = 1).

Although machine learning and deep learning algorithms can properly diagnose diabetic retinopathy, their discriminating capacity is limited. However, they could simplify the diagnosing process. Further studies are required to improve algorithms.