Insights into Systemic Disease through Retinal Imaging-Based Oculomics

Foundation models in ophthalmology

Foundation models represent a paradigm shift in artificial intelligence (AI), evolving from narrow models designed for specific tasks to versatile, generalisable models adaptable to a myriad of diverse applications. Ophthalmology as a specialty has the potential to act as an exemplar for other medical specialties, offering a blueprint for integrating foundation models broadly into clinical practice. This review hopes to serve as a roadmap for eyecare professionals seeking to better understand foundation models, while equipping readers with the tools to explore the use of foundation models in their own research and practice. We begin by outlining the key concepts and technological advances which have enabled the development of these models, providing an overview of novel training approaches and modern AI architectures. Next, we summarise existing literature on the topic of foundation models in ophthalmology, encompassing progress in vision foundation models, large language models and large multimodal models. Finally, we outline major challenges relating to privacy, bias and clinical validation, and propose key steps forward to maximise the benefit of this powerful technology.

Applicability of Oculomics for Individual Risk Prediction: Repeatability and Robustness of Retinal Fractal Dimension Using DART and AutoMorph

Purpose

To investigate whether fractal dimension (FD)-based oculomics could be used for individual risk prediction by evaluating repeatability and robustness.

Methods

We used two datasets: "Caledonia," healthy adults imaged multiple times in quick succession for research (26 subjects, 39 eyes, 377 color fundus images), and GRAPE, glaucoma patients with baseline and follow-up visits (106 subjects, 196 eyes, 392 images). Mean follow-up time was 18.3 months in GRAPE; thus it provides a pessimistic lower bound because vasculature could change. FD was computed with DART and AutoMorph. Image quality was assessed with QuickQual, but no images were initially excluded. Pearson, Spearman, and intraclass correlation (ICC) were used for population-level repeatability. For individual-level repeatability, we introduce measurement noise parameter λ, which is within-eye standard deviation (SD) of FD measurements in units of between-eyes SD.

Results

In Caledonia, ICC was 0.8153 for DART and 0.5779 for AutoMorph, Pearson/Spearman correlation (first and last image) 0.7857/0.7824 for DART, and 0.3933/0.6253 for AutoMorph. In GRAPE, Pearson/Spearman correlation (first and next visit) was 0.7479/0.7474 for DART, and 0.7109/0.7208 for AutoMorph (all P < 0.0001). Median λ in Caledonia without exclusions was 3.55% for DART and 12.65% for AutoMorph and improved to up to 1.67% and 6.64% with quality-based exclusions, respectively. Quality exclusions primarily mitigated large outliers. Worst quality in an eye correlated strongly with λ (Pearson 0.5350-0.7550, depending on dataset and method, all P < 0.0001).

Conclusions

Repeatability was sufficient for individual-level predictions in heterogeneous populations. DART performed better on all metrics and might be able to detect small, longitudinal changes, highlighting the potential of robust methods.

Oculomics: A Crusade Against the Four Horsemen of Chronic Disease

Chronic, non-communicable diseases present a major barrier to living a long and healthy life. In many cases, early diagnosis can facilitate prevention, monitoring, and treatment efforts, improving patient outcomes. There is therefore a critical need to make screening techniques as accessible, unintimidating, and cost-effective as possible. The association between ocular biomarkers and systemic health and disease (oculomics) presents an attractive opportunity for detection of systemic diseases, as ophthalmic techniques are often relatively low-cost, fast, and non-invasive. In this review, we highlight the key associations between structural biomarkers in the eye and the four globally leading causes of morbidity and mortality: cardiovascular disease, cancer, neurodegenerative disease, and metabolic disease. We observe that neurodegenerative disease is a particularly promising target for oculomics, with biomarkers detected in multiple ocular structures. Cardiovascular disease biomarkers are present in the choroid, retinal vasculature, and retinal nerve fiber layer, and metabolic disease biomarkers are present in the eyelid, tear fluid, lens, and retinal vasculature. In contrast, only the tear fluid emerged as a promising ocular target for the detection of cancer. The retina is a rich source of oculomics data, the analysis of which has been enhanced by artificial intelligence-based tools. Although not all biomarkers are disease-specific, limiting their current diagnostic utility, future oculomics research will likely benefit from combining data from various structures to improve specificity, as well as active design, development, and optimization of instruments that target specific disease signatures, thus facilitating differential diagnoses.

Handgrip strength and risks of diabetic vascular complications: Evidence from Guangzhou Diabetic Eye Study and UK cohorts

PURPOSE

The purpose is to investigate the association between handgrip strength (HGS) and the risk of future diabetic complications in multicountry cohorts.

METHODS

The association between HGS and diabetic complications was evaluated using cox models among 84 453 patients with pre-diabetes and diabetes from the UK Biobank with a 12-year follow-up. The association between HGS and longitudinal microcirculatory damage rates was assessed among 819 patients with diabetes from the Guangzhou Diabetic Eye Study (GDES) with a 3-year follow-up. Participants were divided into three age groups (<56, 56-65 and ≥65 years), and each group was further subdivided into three HGS tertiles.

RESULTS

A 5 kg reduction in HGS was associated with increased risk for all-cause mortality (women, HR=1.10, 95% CI: 1.05 to 1.14; p<0.001; men, HR=1.13, 95% CI: 1.11 to 1.15; p<0.001). Women and men in the lowest HGS group exhibited 1.6-times and 1.3-1.5-times higher risk of myocardial infarction and stroke compared with the highest HGS group. In men, there was a higher risk of developing end-stage renal disease (HR=1.83, 95% CI: 1.30 to 2.57; p=0.001), while this was not observed in women. Both sexes in the lowest HGS group had a 1.3-times higher risk of diabetic retinopathy compared with the highest HGS group. In the GDES group, individuals with the lowest HGS showed accelerated microcirculatory damage in retina (all p<0.05).

CONCLUSIONS

Reduced HGS is significantly associated with a higher risk of diabetic complications and accelerated microvascular damage. HGS could serve as a practical indicator of vascular health in patients with pre-diabetes and diabetes.

Non-Invasive Retinal Vessel Analysis as a Predictor for Cardiovascular Disease

Cardiovascular disease (CVD) is the most frequent cause of death worldwide. The alterations in the microcirculation may predict the cardiovascular mortality. The retinal vasculature can be used as a model to study vascular alterations associated with cardiovascular disease. In order to quantify microvascular changes in a non-invasive way, fundus images can be taken and analysed. The central retinal arteriolar (CRAE), the venular (CRVE) diameter and the arteriolar-to-venular diameter ratio (AVR) can be used as biomarkers to predict the cardiovascular mortality. A narrower CRAE, wider CRVE and a lower AVR have been associated with increased cardiovascular events. Dynamic retinal vessel analysis (DRVA) allows the quantification of retinal changes using digital image sequences in response to visual stimulation with flicker light. This article is not just a review of the current literature, it also aims to discuss the methodological benefits and to identify research gaps. It highlights the potential use of microvascular biomarkers for screening and treatment monitoring of cardiovascular disease. Artificial intelligence (AI), such as Quantitative Analysis of Retinal vessel Topology and size (QUARTZ), and SIVA-deep learning system (SIVA-DLS), seems efficient in extracting information from fundus photographs and has the advantage of increasing diagnosis accuracy and improving patient care by complementing the role of physicians. Retinal vascular imaging using AI may help identify the cardiovascular risk, and is an important tool in primary cardiovascular disease prevention. Further research should explore the potential clinical application of retinal microvascular biomarkers, in order to assess systemic vascular health status, and to predict cardiovascular events.

Prognostic potentials of AI in ophthalmology: systemic disease forecasting via retinal imaging

BACKGROUND

Artificial intelligence (AI) that utilizes deep learning (DL) has potential for systemic disease prediction using retinal imaging. The retina's unique features enable non-invasive visualization of the central nervous system and microvascular circulation, aiding early detection and personalized treatment plans for personalized care. This review explores the value of retinal assessment, AI-based retinal biomarkers, and the importance of longitudinal prediction models in personalized care.

MAIN TEXT

This narrative review extensively surveys the literature for relevant studies in PubMed and Google Scholar, investigating the application of AI-based retina biomarkers in predicting systemic diseases using retinal fundus photography. The study settings, sample sizes, utilized AI models and corresponding results were extracted and analysed. This review highlights the substantial potential of AI-based retinal biomarkers in predicting neurodegenerative, cardiovascular, and chronic kidney diseases. Notably, DL algorithms have demonstrated effectiveness in identifying retinal image features associated with cognitive decline, dementia, Parkinson's disease, and cardiovascular risk factors. Furthermore, longitudinal prediction models leveraging retinal images have shown potential in continuous disease risk assessment and early detection. AI-based retinal biomarkers are non-invasive, accurate, and efficient for disease forecasting and personalized care.

CONCLUSION

AI-based retinal imaging hold promise in transforming primary care and systemic disease management. Together, the retina's unique features and the power of AI enable early detection, risk stratification, and help revolutionizing disease management plans. However, to fully realize the potential of AI in this domain, further research and validation in real-world settings are essential.

Machine-learning-based models to predict cardiovascular risk using oculomics and clinic variables in KNHANES

BACKGROUND

Recent researches have found a strong correlation between the triglyceride-glucose (TyG) index or the atherogenic index of plasma (AIP) and cardiovascular disease (CVD) risk. However, there is a lack of research on non-invasive and rapid prediction of cardiovascular risk. We aimed to develop and validate a machine-learning model for predicting cardiovascular risk based on variables encompassing clinical questionnaires and oculomics.

METHODS

We collected data from the Korean National Health and Nutrition Examination Survey (KNHANES). The training dataset (80% from the year 2008 to 2011 KNHANES) was used for machine learning model development, with internal validation using the remaining 20%. An external validation dataset from the year 2012 assessed the model's predictive capacity for TyG-index or AIP in new cases. We included 32122 participants in the final dataset. Machine learning models used 25 algorithms were trained on oculomics measurements and clinical questionnaires to predict the range of TyG-index and AIP. The area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1 score were used to evaluate the performance of our machine learning models.

RESULTS

Based on large-scale cohort studies, we determined TyG-index cut-off points at 8.0, 8.75 (upper one-third values), 8.93 (upper one-fourth values), and AIP cut-offs at 0.318, 0.34. Values surpassing these thresholds indicated elevated cardiovascular risk. The best-performing algorithm revealed TyG-index cut-offs at 8.0, 8.75, and 8.93 with internal validation AUCs of 0.812, 0.873, and 0.911, respectively. External validation AUCs were 0.809, 0.863, and 0.901. For AIP at 0.34, internal and external validation achieved similar AUCs of 0.849 and 0.842. Slightly lower performance was seen for the 0.318 cut-off, with AUCs of 0.844 and 0.836. Significant gender-based variations were noted for TyG-index at 8 (male AUC=0.832, female AUC=0.790) and 8.75 (male AUC=0.874, female AUC=0.862) and AIP at 0.318 (male AUC=0.853, female AUC=0.825) and 0.34 (male AUC=0.858, female AUC=0.831). Gender similarity in AUC (male AUC=0.907 versus female AUC=0.906) was observed only when the TyG-index cut-off point equals 8.93.

CONCLUSION

We have established a simple and effective non-invasive machine learning model that has good clinical value for predicting cardiovascular risk in the general population.

Retinal Fractal Dimension Is a Potential Biomarker for Systemic Health-Evidence From a Mixed-Age, Primary-Care Population

Purpose

To investigate whether fractal dimension (FD), a retinal trait relating to vascular complexity and a potential "oculomics" biomarker for systemic disease, is applicable to a mixed-age, primary-care population.

Methods

We used cross-sectional data (96 individuals; 183 eyes; ages 18-81 years) from a university-based optometry clinic in Glasgow, Scotland, to study the association between FD and systemic health. We computed FD from color fundus images using Deep Approximation of Retinal Traits (DART), an artificial intelligence-based method designed to be more robust to poor image quality.

Results

Despite DART being designed to be more robust, a significant association (P < 0.001) between image quality and FD remained. Consistent with previous literature, age was associated with lower FD (P < 0.001 univariate and when adjusting for image quality). However, FD variance was higher in older patients, and some patients over 60 had FD comparable to those of patients in their 20s. Prevalent systemic conditions were significantly (P = 0.037) associated with lower FD when adjusting for image quality and age.

Conclusions

Our work suggests that FD as a biomarker for systemic health extends to mixed-age, primary-care populations. FD decreases with age but might not substantially decrease in everyone. This should be further investigated using longitudinal data. Finally, image quality was associated with FD, but it is unclear whether this finding is measurement error caused by image quality or confounded by age and health. Future work should investigate this to clarify whether adjusting for image quality is appropriate.

Translational Relevance

FD could potentially be used in regular screening settings, but questions around image quality remain.

The Use of the RETeval Portable Electroretinography Device for Low-Cost Screening: A Mini-Review

Electroretinography (ERG) provides crucial insights into retinal function and the integrity of the visual pathways. However, ERG assessments classically require a complicated technical background with costly equipment. In addition, the placement of corneal or conjunctival electrodes is not always tolerated by the patients, which restricts the measurement for pediatric evaluations. In this short review, we give an overview of the use of the RETeval portable ERG device (LKC Technologies, Inc., Gaithersburg, MD, USA), a modern portable ERG device that can facilitate screening for diseases involving the retina and the optic nerve. We also review its potential to provide ocular biomarkers in systemic pathologies, such as Alzheimer's disease and central nervous system alterations, within the framework of oculomics.

CF-Loss: Clinically-relevant feature optimised loss function for retinal multi-class vessel segmentation and vascular feature measurement

Characterising clinically-relevant vascular features, such as vessel density and fractal dimension, can benefit biomarker discovery and disease diagnosis for both ophthalmic and systemic diseases. In this work, we explicitly encode vascular features into an end-to-end loss function for multi-class vessel segmentation, categorising pixels into artery, vein, uncertain pixels, and background. This clinically-relevant feature optimised loss function (CF-Loss) regulates networks to segment accurate multi-class vessel maps that produce precise vascular features. Our experiments first verify that CF-Loss significantly improves both multi-class vessel segmentation and vascular feature estimation, with two standard segmentation networks, on three publicly available datasets. We reveal that pixel-based segmentation performance is not always positively correlated with accuracy of vascular features, thus highlighting the importance of optimising vascular features directly via CF-Loss. Finally, we show that improved vascular features from CF-Loss, as biomarkers, can yield quantitative improvements in the prediction of ischaemic stroke, a real-world clinical downstream task. The code is available at https://github.com/rmaphoh/feature-loss.

A Siamese ResNeXt network for predicting carotid intimal thickness of patients with T2DM from fundus images

Objective

To develop and validate an artificial intelligence diagnostic model based on fundus images for predicting Carotid Intima-Media Thickness (CIMT) in individuals with Type 2 Diabetes Mellitus (T2DM).

Methods

In total, 1236 patients with T2DM who had both retinal fundus images and CIMT ultrasound records within a single hospital stay were enrolled. Data were divided into normal and thickened groups and sent to eight deep learning models: convolutional neural networks of the eight models were all based on ResNet or ResNeXt. Their encoder and decoder modes are different, including the standard mode, the Parallel learning mode, and the Siamese mode. Except for the six unimodal networks, two multimodal networks based on ResNeXt under the Parallel learning mode or the Siamese mode were embedded with ages. Performance of eight models were compared via the confusion matrix, precision, recall, specificity, F1 value, and ROC curve, and recall was regarded as the main indicator. Besides, Grad-CAM was used to visualize the decisions made by Siamese ResNeXt network, which is the best performance.

Results

Performance of various models demonstrated the following points: 1) the RexNeXt showed a notable improvement over the ResNet; 2) the structural Siamese networks, which extracted features parallelly and independently, exhibited slight performance enhancements compared to the traditional networks. Notably, the Siamese networks resulted in significant improvements; 3) the performance of classification declined if the age factor was embedded in the network. Taken together, the Siamese ResNeXt unimodal model performed best for its superior efficacy and robustness. This model achieved a recall rate of 88.0% and an AUC value of 90.88% in the validation subset. Additionally, heatmaps calculated by the Grad-CAM algorithm presented concentrated and orderly mappings around the optic disc vascular area in normal CIMT groups and dispersed, irregular patterns in thickened CIMT groups.

Conclusion

We provided a Siamese ResNeXt neural network for predicting the carotid intimal thickness of patients with T2DM from fundus images and confirmed the correlation between fundus microvascular lesions and CIMT.

Screening of Moyamoya Disease From Retinal Photographs: Development and Validation of Deep Learning Algorithms

BACKGROUND

Moyamoya disease (MMD) is a rare and complex pathological condition characterized by an abnormal collateral circulation network in the basal brain. The diagnosis of MMD and its progression is unpredictable and influenced by many factors. MMD can affect the blood vessels supplying the eyes, resulting in a range of ocular symptoms. In this study, we developed a deep learning model using real-world data to assist a diagnosis and determine the stage of the disease using retinal photographs.

METHODS

This retrospective observational study conducted from August 2006 to March 2022 included 498 retinal photographs from 78 patients with MMD and 3835 photographs from 1649 healthy participants. Photographs were preprocessed, and an ResNeXt50 model was developed. Model performance was measured using receiver operating curves and their area under the receiver operating characteristic curve, accuracy, sensitivity, and F1-score. Heatmaps and progressive erasing plus progressive restoration were performed to validate the faithfulness.

RESULTS

Overall, 322 retinal photographs from 67 patients with MMD and 3752 retinal photographs from 1616 healthy participants were used to develop a screening and stage prediction model for MMD. The average age of the patients with MMD was 44.1 years, and the average follow-up time was 115 months. Stage 3 photographs were the most prevalent, followed by stages 4, 5, 2, 1, and 6 and healthy. The MMD screening model had an average area under the receiver operating characteristic curve of 94.6%, with 89.8% sensitivity and 90.4% specificity at the best cutoff point. MMD stage prediction models had an area under the receiver operating characteristic curve of 78% or higher, with stage 3 performing the best at 93.6%. Heatmap identified the vascular region of the fundus as important for prediction, and progressive erasing plus progressive restoration result shows an area under the receiver operating characteristic curve of 70% only with 50% of the important regions.

CONCLUSIONS

This study demonstrated that retinal photographs could be used as potential biomarkers for screening and staging of MMD and the disease stage could be classified by a deep learning algorithm.

Retinal vessel diameters in intermediate age-related macular degeneration using en face optical coherence tomography

CLINICAL RELEVANCE

Clinical assessment of age-related macular degeneration (AMD) relies on biomarkers that do not necessarily reflect the contributions of vascular dysfunction. Validation of clinically accessible methods of measuring retinal vascular integrity could provide a more holistic understanding of AMD-related changes to facilitate appropriate care.

BACKGROUND

There is conflicting evidence if retinal vessel calibre is significantly altered in the early stages of AMD. This study examined the outer and inner diameters of first order retinal vessels in intermediate AMD eyes using en face optical coherence tomography (OCT).

METHODS

Retinal en face (6 × 6 mm) OCT images were examined in a single eye of participants with intermediate AMD (n = 46) versus normal macula (n = 43) for arterioles (all identifiable) and venules (40/46 and 39/43 identifiable). All participants were aged ≥50 years without diabetes mellitus, hypertension, or other systemic vascular disease.

RESULTS

Intra- and inter-grader agreement was good-to-excellent for all en face OCT measurements of arteriole and venule diameters (intraclass correlation coefficient = 0.87 to 0.99). Arteriolar outer diameters (82.3 ± 19.8 µm vs 73.8 ± 16.1 µm; p < 0.05) and inner diameters (35.1 ± 8.4 µm vs 31.5 ± 8.1 µm; p < 0.05) were significantly greater in AMD eyes compared to normal eyes. Venular inner diameter was significantly greater (43.1 ± 9.5 µm vs 39.2 ± 10.1 µm; p < 0.05), but outer diameter remained unchanged (p = 0.17) in AMD eyes compared to normal eyes.

CONCLUSION

Arteriolar dilation and altered venular inner diameter were observed in intermediate AMD eyes. These results support further investigation of vascular contributions to AMD in the early stages of disease, possibly using the en face OCT imaging modality.

Deep Learning and Machine Learning Algorithms for Retinal Image Analysis in Neurodegenerative Disease: Systematic Review of Datasets and Models

Purpose

Retinal images contain rich biomarker information for neurodegenerative disease. Recently, deep learning models have been used for automated neurodegenerative disease diagnosis and risk prediction using retinal images with good results.

Methods

In this review, we systematically report studies with datasets of retinal images from patients with neurodegenerative diseases, including Alzheimer's disease, Huntington's disease, Parkinson's disease, amyotrophic lateral sclerosis, and others. We also review and characterize the models in the current literature which have been used for classification, regression, or segmentation problems using retinal images in patients with neurodegenerative diseases.

Results

Our review found several existing datasets and models with various imaging modalities primarily in patients with Alzheimer's disease, with most datasets on the order of tens to a few hundred images. We found limited data available for the other neurodegenerative diseases. Although cross-sectional imaging data for Alzheimer's disease is becoming more abundant, datasets with longitudinal imaging of any disease are lacking.

Conclusions

The use of bilateral and multimodal imaging together with metadata seems to improve model performance, thus multimodal bilateral image datasets with patient metadata are needed. We identified several deep learning tools that have been useful in this context including feature extraction algorithms specifically for retinal images, retinal image preprocessing techniques, transfer learning, feature fusion, and attention mapping. Importantly, we also consider the limitations common to these models in real-world clinical applications.

Translational Relevance

This systematic review evaluates the deep learning models and retinal features relevant in the evaluation of retinal images of patients with neurodegenerative disease.

Deep learning prediction of steep and flat corneal curvature using fundus photography in post-COVID telemedicine era

Recently, fundus photography (FP) is being increasingly used. Corneal curvature is an essential factor in refractive errors and is associated with several pathological corneal conditions. As FP-based examination systems have already been widely distributed, it would be helpful for telemedicine to extract information such as corneal curvature using FP. This study aims to develop a deep learning model based on FP for corneal curvature prediction by categorizing corneas into steep, regular, and flat groups. The EfficientNetB0 architecture with transfer learning was used to learn FP patterns to predict flat, regular, and steep corneas. In validation, the model achieved a multiclass accuracy of 0.727, a Matthews correlation coefficient of 0.519, and an unweighted Cohen's κ of 0.590. The areas under the receiver operating characteristic curves for binary prediction of flat and steep corneas were 0.863 and 0.848, respectively. The optic nerve and its peripheral areas were the main focus of the model. The developed algorithm shows that FP can potentially be used as an imaging modality to estimate corneal curvature in the post-COVID-19 era, whereby patients may benefit from the detection of abnormal corneal curvatures using FP in the telemedicine setting.

Ocular biomarkers of cognitive decline based on deep-learning retinal vessel segmentation

BACKGROUND

The current literature shows a strong relationship between retinal neuronal and vascular alterations in dementia. The purpose of the study was to use NFN+ deep learning models to analyze retinal vessel characteristics for cognitive impairment (CI) recognition.

METHODS

We included 908 participants from a community-based cohort followed for over 15 years (the prospective KaiLuan Study) who underwent brain magnetic resonance imaging (MRI) and fundus photography between 2021 and 2022. The cohort consisted of both cognitively healthy individuals (N = 417) and those with cognitive impairment (N = 491). We employed the NFN+ deep learning framework for retinal vessel segmentation and measurement. Associations between Retinal microvascular parameters (RMPs: central retinal arteriolar / venular equivalents, arteriole to venular ratio, fractal dimension) and CI were assessed by Pearson correlation. P < 0.05 was considered statistically significant. The correlation between the CI and RMPs were explored, then the correlation coefficients between CI and RMPs were analyzed. Random Forest nonlinear classification model was used to predict whether one having cognitive decline or not. The assessment criterion was the AUC value derived from the working characteristic curve.

RESULTS

The fractal dimension (FD) and global vein width were significantly correlated with the CI (P < 0.05). Age (0.193), BMI (0.154), global vein width (0.106), retinal vessel FD (0.099), and CRAE (0.098) were the variables in this model that were ranked in order of feature importance. The AUC values of the model were 0.799.

CONCLUSIONS

Establishment of a predictive model based on the extraction of vascular features from fundus images has a high recognizability and predictive power for cognitive function and can be used as a screening method for CI.

Automated Retinal Vessel Analysis Based on Fundus Photographs as a Predictor for Non-Ophthalmic Diseases-Evolution and Perspectives

The study of retinal vessels in relation to cardiovascular risk has a long history. The advent of a dedicated tool based on digital imaging, i.e., the retinal vessel analyzer, and also other software such as Integrative Vessel Analysis (IVAN), Singapore I Vessel Assessment (SIVA), and Vascular Assessment and Measurement Platform for Images of the Retina (VAMPIRE), has led to the accumulation of a formidable body of evidence regarding the prognostic value of retinal vessel analysis (RVA) for cardiovascular and cerebrovascular disease (including arterial hypertension in children). There is also the potential to monitor the response of retinal vessels to therapies such as physical activity or bariatric surgery. The dynamic vessel analyzer (DVA) remains a unique way of studying neurovascular coupling, helping to understand the pathogenesis of cerebrovascular and neurodegenerative conditions and also being complementary to techniques that measure macrovascular dysfunction. Beyond cardiovascular disease, retinal vessel analysis has shown associations with and prognostic value for neurological conditions, inflammation, kidney function, and respiratory disease. Artificial intelligence (AI) (represented by algorithms such as QUantitative Analysis of Retinal vessel Topology and siZe (QUARTZ), SIVA-DLS (SIVA-deep learning system), and many others) seems efficient in extracting information from fundus photographs, providing prognoses of various general conditions with unprecedented predictive value. The future challenges will be integrating RVA and other qualitative and quantitative risk factors in a unique, comprehensive prediction tool, certainly powered by AI, while building the much-needed acceptance for such an approach inside the medical community and reducing the "black box" effect, possibly by means of saliency maps.

Association of preoperative retinal microcirculation and perioperative outcomes in patients undergoing congenital cardiac surgery

BACKGROUND

Microcirculatory dysfunction is associated with increased morbidity and mortality in cardiac surgery patients. This study aimed to investigate the association between preoperative retinal microcirculation evaluated using optical coherence tomography angiography (OCTA) and perioperative outcomes in patients with congenital heart disease (CHD).

METHODS

This prospective, observational study was performed from May 2017 to January 2021. OCTA was used to automatically quantify the vessel density (VD) of the superficial capillary plexus, deep capillary plexus (DCP), and radial peripapillary capillary (RPC) preoperatively. The primary outcome was excessive postoperative bleeding, defined as bleeding volume > 75th percentile for 24-hour postoperative chest tube output. The secondary outcome was composite adverse outcomes, including one or more operative mortalities, early postoperative complications, and prolonged length of stay. The association between retinal VD and outcomes was assessed using Poisson regression.

RESULTS

In total, 173 CHD patients who underwent cardiac surgery were included (mean age, 26 years). Among them, 43 (24.9%) and 46 (26.6%) developed excessive postoperative bleeding and composite adverse outcomes, respectively. A lower VD of DCP (odds ratio [OR], 1.24; 95% confidence interval [CI], 1.08-1.43; P = 0.003) was independently associated with excessive postoperative bleeding, and a lower VD of RPC (OR, 1.97; 95% CI, 1.08-3.57; P = 0.027), and DCP (OR, 2.17; 95% CI, 1.08-4.37; P = 0.029) were independently associated with the postoperative composite adverse outcomes.

CONCLUSION

Preoperative retinal hypoperfusion was independently associated with an increased risk of perioperative adverse outcomes in patients with CHD, suggesting that retinal microcirculation evaluation could provide valuable information about the outcomes of cardiac surgery, thereby aiding physicians in tailoring individualized treatment.

Retina Oculomics in Neurodegenerative Disease

Ophthalmic biomarkers have long played a critical role in diagnosing and managing ocular diseases. Oculomics has emerged as a field that utilizes ocular imaging biomarkers to provide insights into systemic diseases. Advances in diagnostic and imaging technologies including electroretinography, optical coherence tomography (OCT), confocal scanning laser ophthalmoscopy, fluorescence lifetime imaging ophthalmoscopy, and OCT angiography have revolutionized the ability to understand systemic diseases and even detect them earlier than clinical manifestations for earlier intervention. With the advent of increasingly large ophthalmic imaging datasets, machine learning models can be integrated into these ocular imaging biomarkers to provide further insights and prognostic predictions of neurodegenerative disease. In this manuscript, we review the use of ophthalmic imaging to provide insights into neurodegenerative diseases including Alzheimer Disease, Parkinson Disease, Amyotrophic Lateral Sclerosis, and Huntington Disease. We discuss recent advances in ophthalmic technology including eye-tracking technology and integration of artificial intelligence techniques to further provide insights into these neurodegenerative diseases. Ultimately, oculomics opens the opportunity to detect and monitor systemic diseases at a higher acuity. Thus, earlier detection of systemic diseases may allow for timely intervention for improving the quality of life in patients with neurodegenerative disease.

Deep learning-based fundus image analysis for cardiovascular disease: a review

It is well established that the retina provides insights beyond the eye. Through observation of retinal microvascular changes, studies have shown that the retina contains information related to cardiovascular disease. Despite the tremendous efforts toward reducing the effects of cardiovascular diseases, they remain a global challenge and a significant public health concern. Conventionally, predicting the risk of cardiovascular disease involves the assessment of preclinical features, risk factors, or biomarkers. However, they are associated with cost implications, and tests to acquire predictive parameters are invasive. Artificial intelligence systems, particularly deep learning (DL) methods applied to fundus images have been generating significant interest as an adjunct assessment tool with the potential of enhancing efforts to prevent cardiovascular disease mortality. Risk factors such as age, gender, smoking status, hypertension, and diabetes can be predicted from fundus images using DL applications with comparable performance to human beings. A clinical change to incorporate DL systems for the analysis of fundus images as an equally good test over more expensive and invasive procedures may require conducting prospective clinical trials to mitigate all the possible ethical challenges and medicolegal implications. This review presents current evidence regarding the use of DL applications on fundus images to predict cardiovascular disease.

The association between cardiovascular risk profile and ocular microvascular changes in patients with non-ST elevation myocardial infarction

PURPOSE

We aimed to evaluate the association between ocular microvasculature (vascular density) on optical coherence tomography-angiography (OCT-A) and the cardiovascular risk profile of patients hospitalized for non-ST-elevation myocardial infarction (NSTEMI) patients.

METHODS

Patients admitted to the intensive care unit with the diagnosis of NSTEMI and undergoing coronary angiography were divided into 3 groups as low, intermediate, and high risk according to the SYNTAX score. OCT-A imaging was performed in all three groups. Right-left selective coronary angiography images of all patients were analyzed. The SYNTAX and TIMI risk scores of all patients were calculated.

RESULTS

This study included opthalmological examination of 114 NSTEMI patients. NSTEMI patients with high SYNTAX risk scores had significantly lower deep parafoveal vessel density (DPD) than patients with low-intermediate SYNTAX risk scores (p < 0.001). ROC curve analysis found that a DPD threshold below 51.65 % was moderately associated with high SYNTAX risk scores in patients with NSTEMI. In addition, NSTEMI patients with high TIMI risk scores had significantly lower DPD than patients with low-intermediate TIMI risk scores (p < 0.001).

CONCLUSIONS

OCT-A may be a non-invasive useful tool to assess the cardiovascular risk profile of NSTEMI patients with a high SYNTAX and TIMI score.

Potential Ocular Biomarkers for Early Detection of Alzheimer's Disease and Their Roles in Artificial Intelligence Studies

Alzheimer's disease (AD) is the leading cause of dementia worldwide. Early detection is believed to be essential to disease management because it enables physicians to initiate treatment in patients with early-stage AD (early AD), with the possibility of stopping the disease or slowing disease progression, preserving function and ultimately reducing disease burden. The purpose of this study was to review prior research on the use of eye biomarkers and artificial intelligence (AI) for detecting AD and early AD. The PubMed database was searched to identify studies for review. Ocular biomarkers in AD research and AI research on AD were reviewed and summarized. According to numerous studies, there is a high likelihood that ocular biomarkers can be used to detect early AD: tears, corneal nerves, retina, visual function and, in particular, eye movement tracking have been identified as ocular biomarkers with the potential to detect early AD. However, there is currently no ocular biomarker that can be used to definitely detect early AD. A few studies that used AI with ocular biomarkers to detect AD reported promising results, demonstrating that using AI with ocular biomarkers through multimodal imaging could improve the accuracy of identifying AD patients. This strategy may become a screening tool for detecting early AD in older patients prior to the onset of AD symptoms.

A foundation model for generalizable disease detection from retinal images

Medical artificial intelligence (AI) offers great potential for recognizing signs of health conditions in retinal images and expediting the diagnosis of eye diseases and systemic disorders1. However, the development of AI models requires substantial annotation and models are usually task-specific with limited generalizability to different clinical applications2. Here, we present RETFound, a foundation model for retinal images that learns generalizable representations from unlabelled retinal images and provides a basis for label-efficient model adaptation in several applications. Specifically, RETFound is trained on 1.6 million unlabelled retinal images by means of self-supervised learning and then adapted to disease detection tasks with explicit labels. We show that adapted RETFound consistently outperforms several comparison models in the diagnosis and prognosis of sight-threatening eye diseases, as well as incident prediction of complex systemic disorders such as heart failure and myocardial infarction with fewer labelled data. RETFound provides a generalizable solution to improve model performance and alleviate the annotation workload of experts to enable broad clinical AI applications from retinal imaging.

A Datasheet for the INSIGHT University Hospitals Birmingham Retinal Vein Occlusion Data Set

Purpose

Retinal vein occlusion (RVO) is the second leading cause of visual loss due to retinal disease. Retinal vein occlusion increases the risk of cardiovascular mortality and the risk of stroke. This article describes the data contained within the INSIGHT eye health data set for RVO and cardiovascular disease.

Design

Data set descriptor for routinely collected eye and systemic disease data.

Participants

All people who had suffered an RVO aged ≥ 18 years old, attending the Ophthalmology Clinic at Queen Elizabeth Hospital, University Hospitals Birmingham (UHB) National Health Service (NHS) Trust were included.

Methods

The INSIGHT Health Data Research Hub for Eye Health is an NHS-led ophthalmic bioresource. It provides researchers with safe access to anonymized routinely collected data from contributing NHS hospitals to advance research for patient benefit. This report describes the INSIGHT UHB RVO and major adverse cardiovascular events data set, a data set of ophthalmology and systemic data derived from the United Kingdom's largest acute care trust.

Main Outcome Measures

This data set consists of routinely collected data from the hospital's electronic patient records. The data set primarily includes structured data (relating to their hospital eye care and any cardiovascular data held for the individual) and OCT ocular images. Further details regarding the available data points are available in the supplementary information.

Results

At the time point of this analysis (September 30, 2022) the data set was composed of clinical data from 1521 patients, from Medisoft records inception. The data set includes 2196 occurrences of RVO affecting 2026 eyes, longitudinal eye follow-up clinical parameters, over 6217 eye-related procedures, and 982 encountered complications. The data set contains information on 2534 major adverse cardiovascular event occurrences, their subtype, number experienced per patient, and chronological relation to RVO event. Longitudinal follow-up data including laboratory results, regular medications, and all-cause mortality are also available within the data set.

Conclusions

This data set descriptor article summarizes the data set contents, the process of its curation, and potential uses. The data set is available through the structured application process that ensures research studies are for patient benefit. Further information regarding the data repository and contact details can be found at https://www.insight.hdrhub.org/.

Financial Disclosures

Proprietary or commercial disclosure may be found after the references.

Learning from small data: Classifying sex from retinal images via deep learning

Deep learning (DL) techniques have seen tremendous interest in medical imaging, particularly in the use of convolutional neural networks (CNNs) for the development of automated diagnostic tools. The facility of its non-invasive acquisition makes retinal fundus imaging particularly amenable to such automated approaches. Recent work in the analysis of fundus images using CNNs relies on access to massive datasets for training and validation, composed of hundreds of thousands of images. However, data residency and data privacy restrictions stymie the applicability of this approach in medical settings where patient confidentiality is a mandate. Here, we showcase results for the performance of DL on small datasets to classify patient sex from fundus images-a trait thought not to be present or quantifiable in fundus images until recently. Specifically, we fine-tune a Resnet-152 model whose last layer has been modified to a fully-connected layer for binary classification. We carried out several experiments to assess performance in the small dataset context using one private (DOVS) and one public (ODIR) data source. Our models, developed using approximately 2500 fundus images, achieved test AUC scores of up to 0.72 (95% CI: [0.67, 0.77]). This corresponds to a mere 25% decrease in performance despite a nearly 1000-fold decrease in the dataset size compared to prior results in the literature. Our results show that binary classification, even with a hard task such as sex categorization from retinal fundus images, is possible with very small datasets. Our domain adaptation results show that models trained with one distribution of images may generalize well to an independent external source, as in the case of models trained on DOVS and tested on ODIR. Our results also show that eliminating poor quality images may hamper training of the CNN due to reducing the already small dataset size even further. Nevertheless, using high quality images may be an important factor as evidenced by superior generalizability of results in the domain adaptation experiments. Finally, our work shows that ensembling is an important tool in maximizing performance of deep CNNs in the context of small development datasets.

Brain and Cognition Signature Fingerprinting Vascular Health in Diabetic Individuals: An International Multi-Cohort Study

OBJECTIVE

To evaluate the correlation between cognitive signatures and the risk of diabetic vascular complications and mortality, based on a multicountry prospective study.

METHODS

The participants comprised 27,773 diabetics from the UK Biobank (UKB) and 1307 diabetics from the Guangzhou Diabetic Eye Study (GDES) cohort. The exposures were brain volume and cognitive screening tests for UKB participants, whilst the global cognitive score (GCS) measuring orientation to time and attention, episodic memory, and visuospatial abilities were determined for GDES participants. The outcomes for the UKB group were mortality, as well as macrovascular (myocardial infarction [MI] and stroke), microvascular (end-stage renal disease [ESRD], and diabetic retinopathy [DR]) events. The outcomes for the GDES group were retinal and renal microvascular damage.

RESULTS

In the UKB group, a 1-SD reduction in brain gray matter volume was associated with 34%-77% higher risks of incident MI, ESRD, and DR. The presence of impaired memory was associated with 18%-73% higher risk of mortality and ESRD; impaired reaction was associated with 1.2-1.7-fold higher risks of mortality, stroke, ESRD, and DR. In the GDES group, the lowest GCS tertile exhibited 1.4-2.2-fold higher risk of developing referable DR and a twofold faster decline in renal function and retinal capillary density compared with the highest tertile. Restricting data analysis to individuals aged less than 65 years produced consistent results.

CONCLUSION

Cognitive decline significantly elevates the risk of diabetic vascular complications and is correlated with retinal and renal microcirculation damage. Cognitive screening tests are strongly recommended as routine tools for management of diabetes.

UK Biobank retinal imaging grading: methodology, baseline characteristics and findings for common ocular diseases

BACKGROUND/OBJECTIVES

This study aims to describe the grading methods and baseline characteristics for UK Biobank (UKBB) participants who underwent retinal imaging in 2009-2010, and to characterise individuals with retinal features suggestive of age-related macular degeneration (AMD), glaucoma and retinopathy.

METHODS

Non-mydriatic colour fundus photographs and macular optical coherence tomography (OCT) scans were manually graded by Central Administrative Research Facility certified graders and quality assured by clinicians of the Network of Ophthalmic Reading Centres UK. Captured retinal features included those associated with AMD (≥1 drusen, pigmentary changes, geographic atrophy or exudative AMD; either imaging modality), glaucoma (≥0.7 cup-disc ratio, ≥0.2 cup-disc ratio difference between eyes, other abnormal disc features; photographs only) and retinopathy (characteristic features of diabetic retinopathy with or without microaneurysms; either imaging modality). Suspected cases of these conditions were characterised with reference to diagnostic records, physical and biochemical measurements.

RESULTS

Among 68,514 UKBB participants who underwent retinal imaging, the mean age was 57.3 years (standard deviation 8.2), 45.7% were men and 90.6% were of White ethnicity. A total of 64,367 participants had gradable colour fundus photographs and 68,281 had gradable OCT scans in at least one eye. Retinal features suggestive of AMD and glaucoma were identified in 15,176 and 2184 participants, of whom 125 (0.8%) and 188 (8.6%), respectively, had a recorded diagnosis. Of 264 participants identified to have retinopathy with microaneurysms, 251 (95.1%) had either diabetes or hypertension.

CONCLUSIONS

This dataset represents a valuable addition to what is currently available in UKBB, providing important insights to both ocular and systemic health.

Association Between Retinal Features From Multimodal Imaging and Schizophrenia

Importance

The potential association of schizophrenia with distinct retinal changes is of clinical interest but has been challenging to investigate because of a lack of sufficiently large and detailed cohorts.

Objective

To investigate the association between retinal biomarkers from multimodal imaging (oculomics) and schizophrenia in a large real-world population.

Design, Setting, and Participants

This cross-sectional analysis used data from a retrospective cohort of 154 830 patients 40 years and older from the AlzEye study, which linked ophthalmic data with hospital admission data across England. Patients attended Moorfields Eye Hospital, a secondary care ophthalmic hospital with a principal central site, 4 district hubs, and 5 satellite clinics in and around London, United Kingdom, and had retinal imaging during the study period (January 2008 and April 2018). Data were analyzed from January 2022 to July 2022.

Main Outcomes and Measures

Retinovascular and optic nerve indices were computed from color fundus photography. Macular retinal nerve fiber layer (RNFL) and ganglion cell-inner plexiform layer (mGC-IPL) thicknesses were extracted from optical coherence tomography. Linear mixed-effects models were used to examine the association between schizophrenia and retinal biomarkers.

Results

A total of 485 individuals (747 eyes) with schizophrenia (mean [SD] age, 64.9 years [12.2]; 258 [53.2%] female) and 100 931 individuals (165 400 eyes) without schizophrenia (mean age, 65.9 years [13.7]; 53 253 [52.8%] female) were included after images underwent quality control and potentially confounding conditions were excluded. Individuals with schizophrenia were more likely to have hypertension (407 [83.9%] vs 49 971 [48.0%]) and diabetes (364 [75.1%] vs 28 762 [27.6%]). The schizophrenia group had thinner mGC-IPL (-4.05 μm, 95% CI, -5.40 to -2.69; P = 5.4 × 10-9), which persisted when investigating only patients without diabetes (-3.99 μm; 95% CI, -6.67 to -1.30; P = .004) or just those 55 years and younger (-2.90 μm; 95% CI, -5.55 to -0.24; P = .03). On adjusted analysis, retinal fractal dimension among vascular variables was reduced in individuals with schizophrenia (-0.14 units; 95% CI, -0.22 to -0.05; P = .001), although this was not present when excluding patients with diabetes.

Conclusions and Relevance

In this study, patients with schizophrenia had measurable differences in neural and vascular integrity of the retina. Differences in retinal vasculature were mostly secondary to the higher prevalence of diabetes and hypertension in patients with schizophrenia. The role of retinal features as adjunct outcomes in patients with schizophrenia warrants further investigation.

A deep learning model for novel systemic biomarkers in photographs of the external eye: a retrospective study

BACKGROUND

Photographs of the external eye were recently shown to reveal signs of diabetic retinal disease and elevated glycated haemoglobin. This study aimed to test the hypothesis that external eye photographs contain information about additional systemic medical conditions.

METHODS

We developed a deep learning system (DLS) that takes external eye photographs as input and predicts systemic parameters, such as those related to the liver (albumin, aspartate aminotransferase [AST]); kidney (estimated glomerular filtration rate [eGFR], urine albumin-to-creatinine ratio [ACR]); bone or mineral (calcium); thyroid (thyroid stimulating hormone); and blood (haemoglobin, white blood cells [WBC], platelets). This DLS was trained using 123 130 images from 38 398 patients with diabetes undergoing diabetic eye screening in 11 sites across Los Angeles county, CA, USA. Evaluation focused on nine prespecified systemic parameters and leveraged three validation sets (A, B, C) spanning 25 510 patients with and without diabetes undergoing eye screening in three independent sites in Los Angeles county, CA, and the greater Atlanta area, GA, USA. We compared performance against baseline models incorporating available clinicodemographic variables (eg, age, sex, race and ethnicity, years with diabetes).

FINDINGS

Relative to the baseline, the DLS achieved statistically significant superior performance at detecting AST >36·0 U/L, calcium <8·6 mg/dL, eGFR <60·0 mL/min/1·73 m2, haemoglobin <11·0 g/dL, platelets <150·0 × 103/μL, ACR ≥300 mg/g, and WBC <4·0 × 103/μL on validation set A (a population resembling the development datasets), with the area under the receiver operating characteristic curve (AUC) of the DLS exceeding that of the baseline by 5·3-19·9% (absolute differences in AUC). On validation sets B and C, with substantial patient population differences compared with the development datasets, the DLS outperformed the baseline for ACR ≥300·0 mg/g and haemoglobin <11·0 g/dL by 7·3-13·2%.

INTERPRETATION

We found further evidence that external eye photographs contain biomarkers spanning multiple organ systems. Such biomarkers could enable accessible and non-invasive screening of disease. Further work is needed to understand the translational implications.

FUNDING

Google.

Analysis of Plasma Metabolic Profile on Ganglion Cell-Inner Plexiform Layer Thickness With Mortality and Common Diseases

Importance

The neural retina is considered a unique window to systemic health, but its biological link with systemic health remains unknown.

Objective

To investigate the independent associations of retinal ganglion cell-inner plexiform layer thickness (GCIPLT) metabolic profiles with rates of mortality and morbidity of common diseases.

Design, Setting, and Participants

This cohort study evaluated UK Biobank participants enrolled between 2006 and 2010, and prospectively followed them up for multidisease diagnosis and mortality. Additional participants from the Guangzhou Diabetes Eye Study (GDES) underwent optical coherence tomography scanning and metabolomic profiling and were included for validation.

Main Outcomes and Measures

Systematic analysis of circulating plasma metabolites to identify GCIPLT metabolic profiles; prospective associations of these profiles with mortality and morbidity of 6 common diseases with their incremental discriminative value and clinical utility.

Results

Among 93 838 community-based participants (51 182 [54.5%] women), the mean (SD) age was 56.7 (8.1) years and mean (SD) follow-up was 12.3 (0.8) years. Of 249 metabolic metrics, 37 were independently associated with GCIPLT, including 8 positive and 29 negative associations, and most were associated with the rates of future mortality and common diseases. These metabolic profiles significantly improved the models for discriminating type 2 diabetes over clinical indicators (C statistic: 0.862; 95% CI, 0.852-0.872 vs clinical indicators only, 0.803; 95% CI, 0.792-0.814; P < .001), myocardial infarction (0.792; 95% CI, 0.775-0.808 vs 0.768; 95% CI, 0.751-0.786; P < .001), heart failure (0.803; 95% CI, 0.786-0.820 vs 0.790; 95% CI, 0.773-0.807; P < .001), stroke (0.739; 95% CI, 0.714-0.764 vs 0.719; 95% CI, 0.693-0.745; P < .001), all-cause mortality (0.747; 95% CI, 0.734-0.760 vs 0.724; 95% CI, 0.711-0.738; P < .001), and cardiovascular disease mortality (0.790; 95% CI, 0.767-0.812 vs 0.763; 95% CI, 0.739-0.788; P < .001). Additionally, the potential of GCIPLT metabolic profiles for risk stratification of cardiovascular diseases were further confirmed in the GDES cohort using a different metabolomic approach.

Conclusions and Relevance

In this prospective study of multinational participants, GCIPLT-associated metabolites demonstrated the potential to inform mortality and morbidity risks. Incorporating information on these profiles may facilitate individualized risk stratification for these health outcomes.

Using Artificial Intelligence to Analyse the Retinal Vascular Network: The Future of Cardiovascular Risk Assessment Based on Oculomics? A Narrative Review

The healthcare burden of cardiovascular diseases remains a major issue worldwide. Understanding the underlying mechanisms and improving identification of people with a higher risk profile of systemic vascular disease through noninvasive examinations is crucial. In ophthalmology, retinal vascular network imaging is simple and noninvasive and can provide in vivo information of the microstructure and vascular health. For more than 10 years, different research teams have been working on developing software to enable automatic analysis of the retinal vascular network from different imaging techniques (retinal fundus photographs, OCT angiography, adaptive optics, etc.) and to provide a description of the geometric characteristics of its arterial and venous components. Thus, the structure of retinal vessels could be considered a witness of the systemic vascular status. A new approach called "oculomics" using retinal image datasets and artificial intelligence algorithms recently increased the interest in retinal microvascular biomarkers. Despite the large volume of associated research, the role of retinal biomarkers in the screening, monitoring, or prediction of systemic vascular disease remains uncertain. A PubMed search was conducted until August 2022 and yielded relevant peer-reviewed articles based on a set of inclusion criteria. This literature review is intended to summarize the state of the art in oculomics and cardiovascular disease research.

Retinal microvascular changes in patients with coronary artery disease and apnea

BACKGROUND

Optical coherence tomography angiography (OCT-A) allowed visualization of capillary level of retina; however, the relationship between coronary vascular status and retinal microvascular changes in patients with apnea is not known well. Our aim was to assess the retinal OCT-A parameters in patients with ischemia and angiographically proven microvascular disease and compare them with obstructive coronary disease in patients with apnea.

METHODS

Our observational study included 185 eyes of 185 patients, 123 eyes of patients with apnea (72 eyes from mild OSAS, 51 eyes from moderate to severe OSAS) and 62 eyes from healthy controls. Radial scans of the macula and OCT-A scans of the central macula (superficial (SCP) and deep (DCP) capillary plexuses) were performed on all participants. All participants had documented sleep apnea disorder within 2 years prior to coronary angiography. Patients were grouped by severity of apnea and coronary atherosclerosis (50 % stenosis cut-off value for obstructive coronary artery disease). Patients presented with myocardial ischemia and without coronary artery occlusion (<50 % diameter reduction or FFR > 0.80) constitute the microvascular coronary artery (INOCA) group.

RESULTS

Compared to healthy controls, patients with apnea showed deterioration in vascular density in all regions of the retina, regardless of obstructive or microvascular coronary artery disease on the ischemia background. This study has provided important observations of a high prevalence of INOCA in patients with OSAS and the presence of OSAS was a significant independent predictor of functional coronary artery disease. The relative decreases in vascular densities were more pronounced in the DCP layer according to SCP layer of macula. Only FAZ area values were significantly different according to the severity of OSAS (0.27 (0.11-0.62) and 0.23 (0.07-0.50) (p = 0.012)).

CONCLUSIONS

In patients with apnea, OCT-A can be used as a noninvasive tool to define coronary artery involvement, with similar retinal microvascular changes both in obstructive and microvascular coronary artery group. In patients with OSAS, we observed a high prevalence of microvascular coronary disease, supporting pathophysiological role of OSAS in ischemia of this group of patients.

Effect of Hypercholesterolemia, Systemic Arterial Hypertension and Diabetes Mellitus on Peripapillary and Macular Vessel Density on Superficial Vascular Plexus in Glaucoma

BACKGROUND/AIMS

Vascular factors are involved in the development of glaucoma, including diseases such as hypercholesterolemia (HC), systemic arterial hypertension (SAH), and diabetes mellitus (DM). The aim of this study was to determine the effect of glaucoma disease on peripapillary vessel density (sPVD) and macular vessel density (sMVD) on the superficial vascular plexus, controlling differences on comorbidities such as SAH, DM and HC between glaucoma patients and normal subjects.

METHODS

In this prospective, unicenter, observational cross-sectional study, sPVD and sMVD were measured in 155 glaucoma patients and 162 normal subjects. Differences between normal subjects and glaucoma patients' groups were analyzed. A linear regression model with 95% confidence and 80% statistical power was performed.

RESULTS

Parameters with greater effect on sPVD were glaucoma diagnosis, gender, pseudophakia and DM. Glaucoma patients had a sPVD 1.2% lower than healthy subjects (Beta slope 1.228; 95%CI 0.798-1.659, p < 0.0001). Women presented 1.19% more sPVD than men (Beta slope 1.190; 95%CI 0.750-1.631, p < 0.0001), and phakic patients presented 1.7% more sPVD than men (Beta slope 1.795; 95%CI 1.311-2.280, p < 0.0001). Furthermore, DM patients had 0.9% lower sPVD than non-diabetic patients (Beta slope 0.925; 95%CI 0.293-1.558, p = 0.004). SAH and HC did not affect most of the sPVD parameters. Patients with SAH and HC showed 1.5% lower sMVD in the outer circle than subjects without those comorbidities (Beta slope 1.513; 95%CI 0.216-2.858, p = 0.021 and 1.549; 95%CI 0.240-2.858, p = 0.022 respectively.

CONCLUSIONS

Glaucoma diagnosis, previous cataract surgery, age and gender seem to have greater influence than the presence of SAH, DM and HC on sPVD and sMVD, particularly sPVD.

A Datasheet for the INSIGHT Birmingham, Solihull and Black Country Diabetic Retinopathy Screening Dataset

Purpose

Diabetic retinopathy (DR) is the most common microvascular complication associated with diabetes mellitus (DM), affecting approximately 40% of this patient population. Early detection of DR is vital to ensure monitoring of disease progression and prompt sight saving treatments as required. This article describes the data contained within the INSIGHT Birmingham, Solihull, and Black Country Diabetic Retinopathy Dataset.

Design

Dataset descriptor for routinely collected eye screening data.

Participants

All diabetic patients aged 12 years and older, attending annual digital retinal photography-based screening within the Birmingham, Solihull, and Black Country Eye Screening Programme.

Methods

The INSIGHT Health Data Research Hub for Eye Health is a National Health Service (NHS)-led ophthalmic bioresource that provides researchers with safe access to anonymized, routinely collected data from contributing NHS hospitals to advance research for patient benefit. This report describes the INSIGHT Birmingham, Solihull, and Black Country DR Screening Dataset, a dataset of anonymized images and linked screening data derived from the United Kingdom's largest regional DR screening program.

Main Outcome Measures

This dataset consists of routinely collected data from the eye screening program. The data primarily include retinal photographs with the associated DR grading data. Additional data such as corresponding demographic details, information regarding patients' diabetic status, and visual acuity data are also available. Further details regarding available data points are available in the supplementary information, in addition to the INSIGHT webpage included below.

Results

At the time point of this analysis (December 31, 2019), the dataset comprised 6 202 161 images from 246 180 patients, with a dataset inception date of January 1, 2007. The dataset includes 1 360 547 grading episodes between R0M0 and R3M1.

Conclusions

This dataset descriptor article summarizes the content of the dataset, how it has been curated, and what its potential uses are. Data are available through a structured application process for research studies that support discovery, clinical evidence analyses, and innovation in artificial intelligence technologies for patient benefit. Further information regarding the data repository and contact details can be found at https://www.insight.hdrhub.org/.

Financial Disclosures

Proprietary or commercial disclosure may be found after the references.

Retinal nerve fiber layer thinning as a novel fingerprint for cardiovascular events: results from the prospective cohorts in UK and China

BACKGROUND

Retinal structural abnormalities have been found to serve as biomarkers for cardiovascular disease (CVD). However, the association between retinal nerve fiber layer (RNFL) thickness and the incidence of CVD events remains inconclusive, and relevant longitudinal studies are lacking. Therefore, we aimed to examine this link in two prospective cohort studies.

METHODS

A total of 25,563 participants from UK Biobank who were initially free of CVD were included in the current study. Another 635 participants without retinopathy at baseline from the Chinese Guangzhou Diabetes Eye Study (GDES) were adopted as the validation set. Measurements of RNFL thickness in the macular (UK Biobank) and peripapillary (GDES) regions were obtained from optical coherence tomography (OCT). Adjusted hazard ratios (HRs), odd ratios (ORs), and 95% confidence intervals (CI) were calculated to quantify CVD risk.

RESULTS

Over a median follow-up period of 7.67 years, 1281 (5.01%) participants in UK Biobank developed CVD events. Each 5-μm decrease in macular RNFL thickness was associated with an 8% increase in incident CVD risk (HR = 1.08, 95% CI: 1.01-1.17, p = 0.033). Compared with participants in the highest tertile of RNFL thickness, the risk of incident CVD was significantly increased in participants in the lowest thickness tertile (HR = 1.18, 95% CI: 1.01-1.38, p = 0.036). In GDES, 29 (4.57%) patients developed CVD events within 3 years. Lower average peripapillary RNFL thickness was also associated with a higher CVD risk (OR = 1.35, 95% CI: 1.11-1.65, p = 0.003). The additive net reclassification improvement (NRI) was 21.8%, and the absolute NRI was 2.0% by addition of RNFL thickness over the Framingham risk score. Of 29 patients with incident CVD, 7 were correctly reclassified to a higher risk category while 1 was reclassified to a lower category, and 21 high risk patients were not reclassified.

CONCLUSIONS

RNFL thinning was independently associated with increased incident cardiovascular risk and improved reclassification capability, indicating RNFL thickness derived from the non-invasive OCT as a potential retinal fingerprint for CVD event across ethnicities and health conditions.

TRIAL REGISTRATION

ISRCTN 15853192.

Deep Learning in Optical Coherence Tomography Angiography: Current Progress, Challenges, and Future Directions

Optical coherence tomography angiography (OCT-A) provides depth-resolved visualization of the retinal microvasculature without intravenous dye injection. It facilitates investigations of various retinal vascular diseases and glaucoma by assessment of qualitative and quantitative microvascular changes in the different retinal layers and radial peripapillary layer non-invasively, individually, and efficiently. Deep learning (DL), a subset of artificial intelligence (AI) based on deep neural networks, has been applied in OCT-A image analysis in recent years and achieved good performance for different tasks, such as image quality control, segmentation, and classification. DL technologies have further facilitated the potential implementation of OCT-A in eye clinics in an automated and efficient manner and enhanced its clinical values for detecting and evaluating various vascular retinopathies. Nevertheless, the deployment of this combination in real-world clinics is still in the "proof-of-concept" stage due to several limitations, such as small training sample size, lack of standardized data preprocessing, insufficient testing in external datasets, and absence of standardized results interpretation. In this review, we introduce the existing applications of DL in OCT-A, summarize the potential challenges of the clinical deployment, and discuss future research directions.

Deep Reinforcement Learning-Based Retinal Imaging in Alzheimer's Disease: Potential and Perspectives

Alzheimer's disease (AD) remains a global health challenge in the 21st century due to its increasing prevalence as the major cause of dementia. State-of-the-art artificial intelligence (AI)-based tests could potentially improve population-based strategies to detect and manage AD. Current retinal imaging demonstrates immense potential as a non-invasive screening measure for AD, by studying qualitative and quantitative changes in the neuronal and vascular structures of the retina that are often associated with degenerative changes in the brain. On the other hand, the tremendous success of AI, especially deep learning, in recent years has encouraged its incorporation with retinal imaging for predicting systemic diseases. Further development in deep reinforcement learning (DRL), defined as a subfield of machine learning that combines deep learning and reinforcement learning, also prompts the question of how it can work hand in hand with retinal imaging as a viable tool for automated prediction of AD. This review aims to discuss potential applications of DRL in using retinal imaging to study AD, and their synergistic application to unlock other possibilities, such as AD detection and prediction of AD progression. Challenges and future directions, such as the use of inverse DRL in defining reward function, lack of standardization in retinal imaging, and data availability, will also be addressed to bridge gaps for its transition into clinical use.

An Overview of Deep-Learning-Based Methods for Cardiovascular Risk Assessment with Retinal Images

Cardiovascular diseases (CVDs) are one of the most prevalent causes of premature death. Early detection is crucial to prevent and address CVDs in a timely manner. Recent advances in oculomics show that retina fundus imaging (RFI) can carry relevant information for the early diagnosis of several systemic diseases. There is a large corpus of RFI systematically acquired for diagnosing eye-related diseases that could be used for CVDs prevention. Nevertheless, public health systems cannot afford to dedicate expert physicians to only deal with this data, posing the need for automated diagnosis tools that can raise alarms for patients at risk. Artificial Intelligence (AI) and, particularly, deep learning models, became a strong alternative to provide computerized pre-diagnosis for patient risk retrieval. This paper provides a novel review of the major achievements of the recent state-of-the-art DL approaches to automated CVDs diagnosis. This overview gathers commonly used datasets, pre-processing techniques, evaluation metrics and deep learning approaches used in 30 different studies. Based on the reviewed articles, this work proposes a classification taxonomy depending on the prediction target and summarizes future research challenges that have to be tackled to progress in this line.

Application of Deep Learning to Retinal-Image-Based Oculomics for Evaluation of Systemic Health: A Review

The retina is a window to the human body. Oculomics is the study of the correlations between ophthalmic biomarkers and systemic health or disease states. Deep learning (DL) is currently the cutting-edge machine learning technique for medical image analysis, and in recent years, DL techniques have been applied to analyze retinal images in oculomics studies. In this review, we summarized oculomics studies that used DL models to analyze retinal images-most of the published studies to date involved color fundus photographs, while others focused on optical coherence tomography images. These studies showed that some systemic variables, such as age, sex and cardiovascular disease events, could be consistently robustly predicted, while other variables, such as thyroid function and blood cell count, could not be. DL-based oculomics has demonstrated fascinating, "super-human" predictive capabilities in certain contexts, but it remains to be seen how these models will be incorporated into clinical care and whether management decisions influenced by these models will lead to improved clinical outcomes.

Integration of Artificial Intelligence into the Approach for Diagnosis and Monitoring of Dry Eye Disease

Dry eye disease (DED) is one of the most common diseases worldwide that can lead to a significant impairment of quality of life. The diagnosis and treatment of the disease are often challenging because of the lack of correlation between the signs and symptoms, limited reliability of diagnostic tests, and absence of established consensus on the diagnostic criteria. The advancement of machine learning, particularly deep learning technology, has enabled the application of artificial intelligence (AI) in various anterior segment disorders, including DED. Currently, many studies have reported promising results of AI-based algorithms for the accurate diagnosis of DED and precise and reliable assessment of data obtained by imaging devices for DED. Thus, the integration of AI into clinical approaches for DED can enhance diagnostic and therapeutic performance. In this review, in addition to a brief summary of the application of AI in anterior segment diseases, we will provide an overview of studies regarding the application of AI in DED and discuss the recent advances in the integration of AI into the clinical approach for DED.

Development and validation of a deep learning algorithm based on fundus photographs for estimating the CAIDE dementia risk score

BACKGROUND

the Cardiovascular Risk Factors, Aging, and Incidence of Dementia (CAIDE) dementia risk score is a recognised tool for dementia risk stratification. However, its application is limited due to the requirements for multidimensional information and fasting blood draw. Consequently, an effective and non-invasive tool for screening individuals with high dementia risk in large population-based settings is urgently needed.

METHODS

a deep learning algorithm based on fundus photographs for estimating the CAIDE dementia risk score was developed and internally validated by a medical check-up dataset included 271,864 participants in 19 province-level administrative regions of China, and externally validated based on an independent dataset included 20,690 check-up participants in Beijing. The performance for identifying individuals with high dementia risk (CAIDE dementia risk score ≥ 10 points) was evaluated by area under the receiver operating curve (AUC) with 95% confidence interval (CI).

RESULTS

the algorithm achieved an AUC of 0.944 (95% CI: 0.939-0.950) in the internal validation group and 0.926 (95% CI: 0.913-0.939) in the external group, respectively. Besides, the estimated CAIDE dementia risk score derived from the algorithm was significantly associated with both comprehensive cognitive function and specific cognitive domains.

CONCLUSIONS

this algorithm trained via fundus photographs could well identify individuals with high dementia risk in a population setting. Therefore, it has the potential to be utilised as a non-invasive and more expedient method for dementia risk stratification. It might also be adopted in dementia clinical trials, incorporated as inclusion criteria to efficiently select eligible participants.

Insights into Sickle Cell Disease through the Retinal Microvasculature: Adaptive Optics Scanning Light Ophthalmoscopy Correlates of Clinical OCT Angiography

PURPOSE

Clinical OCT angiography (OCTA) of the retinal microvasculature offers a quantitative correlate to systemic disease burden and treatment efficacy in sickle cell disease (SCD). The purpose of this study was to use the higher resolution of adaptive optics scanning light ophthalmoscopy (AOSLO) to elucidate OCTA features of parafoveal microvascular compromise identified in SCD patients.

DESIGN

Case series of 11 SCD patients and 1 unaffected control.

PARTICIPANTS

A total of 11 eyes of 11 SCD patients (mean age, 33 years; range, 23-44; 8 female, 3 male) and 1 eye of a 34-year-old unaffected control.

METHODS

Ten sequential 3 × 3 mm parafoveal OCTA full vascular slab scans were obtained per eye using a commercial spectral domain OCT system (Avanti RTVue-XR; Optovue). These were used to identify areas of compromised perfusion near the foveal avascular zone (FAZ), designated as regions of interest (ROIs). Immediately thereafter, AOSLO imaging was performed on these ROIs to examine the cellular details of abnormal perfusion. Each participant was imaged at a single cross-sectional time point. Additionally, 2 of the SCD patients were imaged prospectively 2 months after initial imaging to study compromised capillary segments across time and with treatment.

MAIN OUTCOME MEASURES

Detection and characterization of parafoveal perfusion abnormalities identified using OCTA and resolved using AOSLO imaging.

RESULTS

We found evidence of abnormal blood flow on OCTA and AOSLO imaging among all 11 SCD patients with diverse systemic and ocular histories. Adaptive optics scanning light ophthalmoscopy imaging revealed a spectrum of phenomena, including capillaries with intermittent blood flow, blood cell stasis, and sites of thrombus formation. Adaptive optics scanning light ophthalmoscopy imaging was able to resolve single sickled red blood cells, rouleaux formations, and blood cell-vessel wall interactions. OCT angiography and AOSLO imaging were sensitive enough to document improved retinal perfusion in an SCD patient 2 months after initiation of oral hydroxyurea therapy.

CONCLUSIONS

Adaptive optics scanning light ophthalmoscopy imaging was able to reveal the cellular details of perfusion abnormalities detected using clinical OCTA. The synergy between these clinical and laboratory imaging modalities presents a promising avenue in the management of SCD through the development of noninvasive ocular biomarkers to prognosticate progression and measure the response to systemic treatment.

Comparison of Retinal Imaging Techniques in Individuals with Pulmonary Artery Hypertension Using Vessel Generation Analysis

(1) Background: Retinal vascular imaging plays an essential role in diagnosing and managing chronic diseases such as diabetic retinopathy, sickle cell retinopathy, and systemic hypertension. Previously, we have shown that individuals with pulmonary arterial hypertension (PAH), a rare disorder, exhibit unique retinal vascular changes as seen using fluorescein angiography (FA) and that these changes correlate with PAH severity. This study aimed to determine if color fundus (CF) imaging could garner identical retinal information as previously seen using FA images in individuals with PAH. (2) Methods: VESGEN, computer software which provides detailed vascular patterns, was used to compare manual segmentations of FA to CF imaging in PAH subjects (n = 9) followed by deep learning (DL) processing of CF imaging to increase the speed of analysis and facilitate a noninvasive clinical translation. (3) Results: When manual segmentation of FA and CF images were compared using VESGEN analysis, both showed identical tortuosity and vessel area density measures. This remained true even when separating images based on arterial trees only. However, this was not observed with microvessels. DL segmentation when compared to manual segmentation of CF images showed similarities in vascular structure as defined by fractal dimension. Similarities were lost for tortuosity and vessel area density when comparing manual CF imaging to DL imaging. (4) Conclusions: Noninvasive imaging such as CF can be used with VESGEN to provide an accurate and safe assessment of retinal vascular changes in individuals with PAH. In addition to providing insight into possible future clinical translational use.

Radiomics-Based Assessment of OCT Angiography Images for Diabetic Retinopathy Diagnosis

Purpose

To evaluate the diagnostic accuracy of machine learning (ML) techniques applied to radiomic features extracted from OCT and OCT angiography (OCTA) images for diabetes mellitus (DM), diabetic retinopathy (DR), and referable DR (R-DR) diagnosis.

Design

Cross-sectional analysis of a retinal image dataset from a previous prospective OCTA study (ClinicalTrials.govNCT03422965).

Participants

Patients with type 1 DM and controls included in the progenitor study.

Methods

Radiomic features were extracted from fundus retinographies, OCT, and OCTA images in each study eye. Logistic regression, linear discriminant analysis, support vector classifier (SVC)-linear, SVC-radial basis function, and random forest models were created to evaluate their diagnostic accuracy for DM, DR, and R-DR diagnosis in all image types.

Main Outcome Measures

Area under the receiver operating characteristic curve (AUC) mean and standard deviation for each ML model and each individual and combined image types.

Results

A dataset of 726 eyes (439 individuals) were included. For DM diagnosis, the greatest AUC was observed for OCT (0.82, 0.03). For DR detection, the greatest AUC was observed for OCTA (0.77, 0.03), especially in the 3 × 3 mm superficial capillary plexus OCTA scan (0.76, 0.04). For R-DR diagnosis, the greatest AUC was observed for OCTA (0.87, 0.12) and the deep capillary plexus OCTA scan (0.86, 0.08). The addition of clinical variables (age, sex, etc.) improved most models AUC for DM, DR and R-DR diagnosis. The performance of the models was similar in unilateral and bilateral eyes image datasets.

Conclusions

Radiomics extracted from OCT and OCTA images allow identification of patients with DM, DR, and R-DR using standard ML classifiers. OCT was the best test for DM diagnosis, OCTA for DR and R-DR diagnosis and the addition of clinical variables improved most models. This pioneer study demonstrates that radiomics-based ML techniques applied to OCT and OCTA images may be an option for DR screening in patients with type 1 DM.

Financial Disclosures

Proprietary or commercial disclosure may be found after the references.

A Generic Pixel Pitch Calibration Method for Fundus Camera via Automated ROI Extraction

Pixel pitch calibration is an essential step to make the fundus structures in the fundus image quantitatively measurable, which is important for the diagnosis and treatment of many diseases, e.g., diabetes, arteriosclerosis, hereditary optic atrophy, etc. The conventional calibration approaches require the specific parameters of the fundus camera or several specially shot images of the chess board, but these are generally not accessible, and the calibration results cannot be generalized to other cameras. Based on automated ROI (region of interest) and optic disc detection, the diameter ratio of ROI and optic disc (ROI-disc ratio) is quantitatively analyzed for a large number of fundus images. With the prior knowledge of the average diameter of an optic disc in fundus, the pixel pitch can be statistically estimated from a large number of fundus images captured by a specific camera without the availability of chess board images or detailed specifics of the fundus camera. Furthermore, for fundus cameras of FOV (fixed field-of-view), the pixel pitch of a fundus image of 45° FOV can be directly estimated according to the automatically measured diameter of ROI in the pixel. The average ROI-disc ratio is approximately constant, i.e., 6.404 ± 0.619 in the pixel, according to 40,600 fundus images, captured by different cameras, of 45° FOV. In consequence, the pixel pitch of a fundus image of 45° FOV can be directly estimated according to the automatically measured diameter of ROI in the pixel, and results show the pixel pitches of Canon CR2, Topcon NW400, Zeiss Visucam 200, and Newvision RetiCam 3100 cameras are 6.825 ± 0.666 μm, 6.625 ± 0.647 μm, 5.793 ± 0.565 μm, and 5.884 ± 0.574 μm, respectively. Compared with the manually measured pixel pitches, based on the method of ISO 10940:2009, i.e., 6.897 μm, 6.807 μm, 5.693 μm, and 6.050 μm, respectively, the bias of the proposed method is less than 5%. Since our method doesn't require chess board images or detailed specifics, the fundus structures on the fundus image can be measured accurately, according to the pixel pitch obtained by this method, without knowing the type and parameters of the camera.

Deep learning for predicting uncorrected refractive error using posterior segment optical coherence tomography images

BACKGROUND/OBJECTIVES

This study aimed to evaluate a deep learning model for estimating uncorrected refractive error using posterior segment optical coherence tomography (OCT) images.

METHODS

In this retrospective study, we assigned healthy subjects to development (N = 688 eyes of 344 subjects) and test (N = 248 eyes of 124 subjects) datasets (prospective validation design). We developed and validated OCT-based deep learning models to estimate refractive error. A regression model based on a pretrained ResNet50 architecture was trained using horizontal OCT images to predict the spherical equivalent (SE). The performance of the deep learning model for detecting high myopia was also evaluated. A saliency map was generated using the Grad-CAM technique to visualize the characteristic features.

RESULTS

The developed model showed a low mean absolute error for SE prediction (2.66 D) and a significant Pearson correlation coefficient of 0.588 (P < 0.001) in the test dataset validation. To detect high myopia, the model yielded an area under the receiver operating characteristic curve of 0.813 (95% confidence interval [CI], 0.744-0.881) and an accuracy of 71.4% (95% CI, 65.3-76.9%). The inner retinal layers and relatively steepened curvatures were highlighted using a saliency map to detect high myopia.

CONCLUSION

A deep learning algorithm showed that OCT could potentially be used as an imaging modality to estimate refractive error. This method will facilitate the evaluation of refractive error to prevent clinicians from overlooking the risks associated with refractive error during OCT assessment.

k-SALSA: k-anonymous synthetic averaging of retinal images via local style alignment

The application of modern machine learning to retinal image analyses offers valuable insights into a broad range of human health conditions beyond ophthalmic diseases. Additionally, data sharing is key to fully realizing the potential of machine learning models by providing a rich and diverse collection of training data. However, the personallyidentifying nature of retinal images, encompassing the unique vascular structure of each individual, often prevents this data from being shared openly. While prior works have explored image de-identification strategies based on synthetic averaging of images in other domains (e.g. facial images), existing techniques face difficulty in preserving both privacy and clinical utility in retinal images, as we demonstrate in our work. We therefore introduce k-SALSA, a generative adversarial network (GAN)-based framework for synthesizing retinal fundus images that summarize a given private dataset while satisfying the privacy notion of k-anonymity. k-SALSA brings together state-of-the-art techniques for training and inverting GANs to achieve practical performance on retinal images. Furthermore, k-SALSA leverages a new technique, called local style alignment, to generate a synthetic average that maximizes the retention of fine-grain visual patterns in the source images, thus improving the clinical utility of the generated images. On two benchmark datasets of diabetic retinopathy (EyePACS and APTOS), we demonstrate our improvement upon existing methods with respect to image fidelity, classification performance, and mitigation of membership inference attacks. Our work represents a step toward broader sharing of retinal images for scientific collaboration. Code is available at https://github.com/hcholab/k-salsa.