Search

Background/aimsTo evaluate the performances of deep learning (DL) algorithms for detection of presence and extent pterygium, based on colour anterior segment photographs (ASPs) taken from slit-lamp and hand-held cameras.MethodsReferable pterygium was defined as having extension towards the cornea from the limbus of >2.50 mm or base width at the limbus of >5.00 mm. 2503 images from the Singapore Epidemiology of Eye Diseases (SEED) study were used as the development set. Algorithms were validated on an internal set from the SEED cohort (629 images (55.3% pterygium, 8.4% referable pterygium)), and tested on two external clinic-based sets (set 1 with 2610 images (2.8% pterygium, 0.7% referable pterygium, from slit-lamp ASP); and set 2 with 3701 images, 2.5% pterygium, 0.9% referable pterygium, from hand-held ASP).ResultsThe algorithm’s area under the receiver operating characteristic curve (AUROC) for detection of any pterygium was 99.5%(sensitivity=98.6%; specificity=99.0%) in internal test set, 99.1% (sensitivity=95.9%, specificity=98.5%) in external test set 1 and 99.7% (sensitivity=100.0%; specificity=88.3%) in external test set 2. For referable pterygium, the algorithm’s AUROC was 98.5% (sensitivity=94.0%; specificity=95.3%) in internal test set, 99.7% (sensitivity=87.2%; specificity=99.4%) in external set 1 and 99.0% (sensitivity=94.3%; specificity=98.0%) in external set 2.ConclusionDL algorithms based on ASPs can detect presence of and referable-level pterygium with optimal sensitivity and specificity. These algorithms, particularly if used with a handheld camera, may potentially be used as a simple screening tool for detection of referable pterygium. Further validation in community setting is warranted.Synopsis/precisDL algorithms based on ASPs can detect presence of and referable-level pterygium optimally, and may be used as a simple screening tool for the detection of referable pterygium in community screenings.

BackgroundTo develop computer-aided detection (CADe) of ORL abnormalities in the retinal pigmented epithelium, interdigitation zone and ellipsoid zone via optical coherence tomography (OCT).MethodsIn this retrospective study, healthy participants with normal ORL, and patients with abnormality of ORL including choroidal neovascularisation (CNV) or retinitis pigmentosa (RP) were included. First, an automatic segmentation deep learning (DL) algorithm, CADe, was developed for the three outer retinal layers using 120 handcraft masks of ORL. This automatic segmentation algorithm generated 4000 segmentations, which included 2000 images with normal ORL and 2000 (1000 CNV and 1000 RP) images with focal or wide defects in ORL. Second, based on the automatically generated segmentation images, a binary classifier (normal vs abnormal) was developed. Results were evaluated by area under the receiver operating characteristic curve (AUC).ResultsThe DL algorithm achieved an AUC of 0.984 (95% CI 0.976 to 0.993) for individual image evaluation in the internal test set of 797 images. In addition, performance analysis of a publicly available external test set (n=968) had an AUC of 0.957 (95% CI 0.944 to 0.970) and a second clinical external test set (n=1124) had an AUC of 0.978 (95% CI 0.970 to 0.986). Moreover, the CADe highlighted well normal parts of ORL and omitted highlights in abnormal ORLs of CNV and RP.ConclusionThe CADe can use OCT images to segment ORL and differentiate between normal ORL and abnormal ORL. The CADe classifier also performs visualisation and may aid future physician diagnosis and clinical applications.

AimsTo evaluate the normative profiles for neuroretinal rim area (RA) in a multiethnic Asian population.MethodsSubjects were recruited from the Singapore Epidemiology of Eye Diseases (2009–2015) study and underwent standardised examinations. RA measurements were performed using Cirrus high-definition optical coherence tomography (Carl Zeiss Meditec). Multivariable linear regression with generalised estimating equation model was used to evaluate the associations between demographic, systemic and ocular factors with RA.ResultsA total of 9394 eyes from 5116 subjects (1724 Chinese, 1463 Malay, 1929 Indian) were included in the final analysis. The mean (±SD) of RA was 1.28 (±0.23) mm2for Chinese, 1.33 (±0.26) mm2for Malays, and 1.23 (±0.23) mm2for Indians. The 5th percentile value for RA was 0.94 mm2for Chinese, 0.96 mm2for Malay, and 0.89 mm2for Indian. In multivariable analysis, following adjustment for age, gender, body mass index, diabetes mellitus, hyperlipidaemia, history of cataract surgery, axial length, intraocular pressure (IOP) and disc area, Indian eyes have smaller RA when compared with Malays (β=−0.074; 95% CI −0.090 to −0.058; p<0.001) and Chinese (β=−0.035; 95% CI −0.051 to −0.019; p<0.001), respectively. Additionally, older age (per decade, β=−0.022), male gender (β=−0.031), longer axial length (per mm, β=−0.025), spherical equivalent (per negative dioptre, β=−0.005), higher IOP (per mm Hg, β=−0.009) were associated with smaller RA (all p≤0.004).ConclusionIn this multiethnic population-based study, we observed significantly smaller RA in Indian eyes, compared with Chinese and Malays. This indicates the need of a more refined ethnic-specific RA normative databases among Asians.

By 2050, almost 5 billion people globally are projected to have myopia, of whom 20% are likely to have high myopia with clinically significant risk of sight-threatening complications such as myopic macular degeneration. These are diagnoses that typically require specialist assessment or measurement with multiple unconnected pieces of equipment. Artificial intelligence (AI) approaches might be effective for risk stratification and to identify individuals at highest risk of visual loss. However, unresolved challenges for AI medical studies remain, including paucity of transparency, auditability, and traceability.

Coronary artery calcium (CAC) score is a clinically validated marker of cardiovascular disease risk. We developed and validated a novel cardiovascular risk stratification system based on deep-learning-predicted CAC from retinal photographs.

In current approaches to vision screening in the community, a simple and efficient process is needed to identify individuals who should be referred to tertiary eye care centres for vision loss related to eye diseases. The emergence of deep learning technology offers new opportunities to revolutionise this clinical referral pathway. We aimed to assess the performance of a newly developed deep learning algorithm for detection of disease-related visual impairment.

BackgroundThe ability of deep learning (DL) algorithms to identify eyes with neovascular age-related macular degeneration (nAMD) from optical coherence tomography (OCT) scans has been previously established. We herewith evaluate the ability of a DL model, showing excellent performance on a Korean data set, to generalse onto an American data set despite ethnic differences. In addition, expert graders were surveyed to verify if the DL model was appropriately identifying lesions indicative of nAMD on the OCT scans.MethodsModel development data set—12 247 OCT scans from South Korea; external validation data set—91 509 OCT scans from Washington, USA. In both data sets, normal eyes or eyes with nAMD were included. After internal testing, the algorithm was sent to the University of Washington, USA, for external validation. Area under the receiver operating characteristic curve (AUC) and precision–recall curve (AUPRC) were calculated. For model explanation, saliency maps were generated using Guided GradCAM.ResultsOn external validation, AUC and AUPRC remained high at 0.952 (95% CI 0.942 to 0.962) and 0.891 (95% CI 0.875 to 0.908) at the individual level. Saliency maps showed that in normal OCT scans, the fovea was the main area of interest; in nAMD OCT scans, the appropriate pathological features were areas of model interest. Survey of 10 retina specialists confirmed this.ConclusionOur DL algorithm exhibited high performance for nAMD identification in a Korean population, and generalised well to an ethnically distinct, American population. The model correctly focused on the differences within the macular area to extract features associated with nAMD.

Deep learning (DL), a subset of artificial intelligence (AI) based on deep neural networks, has made significant breakthroughs in medical imaging, particularly for image classification and pattern recognition. In ophthalmology, applying DL for glaucoma assessment with optical coherence tomography (OCT), including OCT traditional reports, two-dimensional (2D) B-scans, and three-dimensional (3D) volumetric scans, has increasingly raised research interests. Studies have demonstrated that using DL for interpreting OCT is efficient, accurate, and with good performance for discriminating glaucomatous eyes from normal eyes, suggesting that incorporation of DL technology in OCT for glaucoma assessment could potentially address some gaps in the current practice and clinical workflow. However, further research is crucial in tackling some existing challenges, such as annotation standardization (i.e., setting a standard for ground truth labelling among different studies), development of DL-powered IT infrastructure for real-world implementation, prospective validation in unseen datasets for further evaluation of generalizability, cost-effectiveness analysis after integration of DL, the AI “black box” explanation problem. This review summarizes recent studies on the application of DL on OCT for glaucoma assessment, identifies the potential clinical impact arising from the development and deployment of the DL models, and discusses future research directions.

Background: To describe the design, implementation, and evaluation of a nurse-led intravitreal injection (NL-IVT) programme in a Singapore tertiary eye hospital. Methods: Patients requiring anti-vascular endothelial growth factor (VEGF) IVT were recruited. Implementation and evaluation were done in the Singapore National Eye Centre, a tertiary centre. To assess safety, nurse injectors recorded details of procedures performed and complications for an 8-month period from February 2019. To evaluate patient experience, we used a modified patient questionnaire and recorded both patients’ waiting time and IVT procedure duration. A retrospective audit of IVTs before and after the introduction of NL-IVT was performed from January 2017 to September 2019. Cost difference between NL-IVT and standard doctor-led (DL) IVT was evaluated. Results: A total of 8599 NL-IVTs were performed. No cases of severe complication were detected in the follow-up. A total of 135 patients who received NL-IVT and DL-IVT were surveyed. General satisfaction, interpersonal manner, financial aspect, time spent with injector, and staff competence were higher in NL-IVTs than in DL-IVTs (p?&lt;?0.05). There were no differences in terms of technical quality and communication. For 934 patients, waiting time was significantly shorter in NL-IVT (3.6?±?10.3?min) compared with DL-IVTs (35.3?±?32.3?min); on average, 19.7??min were saved through NL-IVT (p?&lt;?0.01). The cost difference per IVT between NL-IVT and DL-IVT is estimated at 286 SGD (163 GBP). Conclusion: With a well-designed training programme, NL-IVT is a safe, acceptable, and cost savings procedure. With increasing demand for IVT, NL-IVT provides an alternative model of care for healthcare systems globally.

Though the domain of big data and artificial intelligence in health care continues to evolve, there is a lack of systemic methods to improve data quality and streamline the preparation process. To address this, we aimed to develop an automated sorting system (RetiSort) that accurately labels the type and laterality of retinal photographs.

Supplemental Digital Content is available in the text.

Background/AimThe aim of this study was to estimate the nationwide incidence of clinically diagnosed exudative age-related macular degeneration (AMD) and associated use of ranibizumab and aflibercept in South Korea.MethodsIn this retrospective, population-based cohort study, claims data for 2010–2015 were analysed in a randomly selected sample of 519 661 adults aged ≥40 years. The incidence per 10 000 person-years was estimated, along with the 95% CI. Incident exudative AMD was defined based on the registration code for rare intractable diseases. Use of ranibizumab and aflibercept and the incidence of exudative AMD were recorded.ResultsNine hundred and twelve patients were newly diagnosed with exudative AMD in 2010–2015. The 6-year incidence in the general population aged ≥40 years was 2.9 (95% CI 2.8 to 3.0) per 10 000 person-years. The incidence was highest in individuals aged 75–79 years (12.0, 95% CI 10.3 to 13.8). The incidence was higher in men than in women in all age groups. Six hundred and twenty-five (69%) of the 912 newly diagnosed patients started ranibizumab or aflibercept as a first-line treatment. The average number of injections administered was 6.1 (SD 3.9; minimum of 1 injection and maximum government-supported limit of 14) during 2010–2015; the number increased with increasing government funding support (from 5 to 10 and from 10 to 14 in 2013 and 2014, respectively).ConclusionsThis study describes the incidence of exudative AMD in South Korea and its treatment under the national health insurance system in this country. Its findings could be used for reference purposes and be useful when planning treatment for exudative AMD.

IMPORTANCE: Information on the incidence of central serous chorioretinopathy (CSC) in individuals who receive corticosteroids is scarce but clinically important because these agents are useful and widely used. OBJECTIVE: To estimate the annual and 5-year incidence of CSC in South Korea in the overall population and in those who have used corticosteroid medications. DESIGN, SETTING, AND PARTICIPANTS: A cohort study of a population-based random sample included East Asian adults for whom records are held in the Korean National Health Insurance Service database for calendar years 2011 through 2015. The data analysis was performed from July 1, 2017 to January 5, 2018. EXPOSURES: Any type of corticosteroid use from 2002 through 2015. MAIN OUTCOMES AND MEASURES: Incidence of CSC. RESULTS: The data set contained data from 868 939 adults (4 117 768 person-years). From 2011 through 2015, 1423 individuals (among whom the mean [SD] age was 46.8 [16.4] years and 1091 [76.7%] were male) with newly diagnosed CSC were identified. From 2002 to 2015, 783 099 individuals in the data set (90.1%) had ever used corticosteroids. The incidence of CSC per 10 000 person-years was 3.5 (5.4 in men; 1.6 in women) among the total population, 2.5 (3.0 in men; 1.2 in women) in those who had never used corticosteroids, and 3.6 (5.7 in men; 1.6 in women) among those who had ever used corticosteroids. The risk of CSCR with individuals who had ever used corticosteroids was estimated as an adjusted hazard ratio of 1.81 (95% CI, 1.47-2.23) compared with those who have never used these drugs. Current or recent corticosteroid use showed a positive association with the incidence of CSC (depending on duration of use, adjusted hazard ratio ranged from 1.54 to 2.15). Corticosteroid use in 2006 through 2009 was associated with an increased incidence of CSC after 2011 (adjusted hazard ratio 1.57 [95% CI, 1.13-2.18]). CONCLUSIONS AND RELEVANCE: In 2002 through 2015, 90.1% of adults in Korea received corticosteroids at least once. Although there was a clear difference in relative risk, this data analysis could not replicate the more than 30-fold increase in the risk ratio of CSC that has been reported previously. The incidence of CSC in the most vulnerable group, middle-aged men, was estimated to be approximately 1 case per 1000 corticosteroid users in the year following medication use. The overall incidence among those who had ever used corticosteroids and those who had never used these drugs was 2.5 and 3.6 per 10 000 person-years, respectively. This study provides additional evidence to support the potential role of corticosteroids in CSC.

IMPORTANCE: It has been suggested that physical activity (PA) is associated with reduced risk for early age-related macular degeneration (AMD). Systematic evaluation has been examining the association between lifestyle and neovascular AMD in an East Asian population, with a particular focus on past vigorous PA. OBJECTIVE: To investigate the association between neovascular AMD and past PA, particularly a history of vigorous exercise, in the overall study population and among 2 a priori–defined subgroups. DESIGN, SETTING, AND PARTICIPANTS: In this propensity score–matched cohort study, individuals between ages 45 and 79 years who were included in the South Korean National Health Insurance Service database from 2002 through 2013 were evaluated. Physical activity and incident neovascular AMD were recorded at baseline (2002-2003) and at follow-up (August 1, 2009, to December 31, 2013), respectively. Using a 1:1 propensity score-matched analysis, the incidence of neovascular AMD was compared using hazard ratios (HRs) for neovascular AMD between 105 980 participants who did and 105 980 who did not (no-PA) engage in vigorous PA. The data analysis was performed from April 19, 2017, to June 5, 2017. EXPOSURES: Physical activity. MAIN OUTCOMES AND MEASURES: Incident cases of neovascular AMD. RESULTS: Of the 211 960 participants (92 036 [43.4%] women; mean [SD] age, 55.1 [7.8] years), neovascular AMD was detected at follow-up in 250 (0.24%) individuals who engaged in past vigorous PA and in 198 (0.19%) of those who did not (HR, 1.23; 95% CI, 1.02-1.49). In subgroup analysis, vigorous PA was associated with a greater HR for neovascular AMD in participants aged 45 to 64 years (HR, 1.30; 95% CI, 1.04-1.63) and in men (HR, 1.36; 95% CI, 1.09-1.69). In the high-PA (≥5 times/wk: HR, 1.54; 95% CI, 1.15-2.06) and moderate-PA (1-4 times/wk: HR, 1.28; 95% CI, 1.01-1.63) groups, there was a greater incidence of neovascular AMD in the vigorous PA than in the no-PA group for men; no association was found for women. CONCLUSIONS AND RELEVANCE: Self-reported past vigorous PA in men aged 45 to 64 years was associated with an increased risk for neovascular AMD. To our knowledge, no previous study has reported such an association; replication of the results would seem warranted to strengthen the likelihood of a cause and effect relationship.

AimTo evaluate the incidence and prevalence of uveitis and associated risk factors in South Korea.MethodsFor this retrospective national cohort study, approximately 1 000 000 Korean residents were randomly selected from the Korean National Health Insurance Service database. Uveitis was defined according to the Korean Classification of Diseases. The uveitis incidence and prevalence were calculated. Sociodemographic factors and comorbidities associated with uveitis were evaluated via Cox proportional regression models.ResultsA total of 1 094 440 subjects were evaluated over 7 051 346 person-years (mean follow-up: 6.44 years). Overall, 7447 newly developed uveitis cases were identified during the period of 2007–2013; the average incidence of uveitis was 10.6 per 10 000 person-years (95% CI 10.3 to 10.8). The average incidences of anterior and non-anterior uveitis were 9.0 and 1.5 per 10 000 person-years, respectively. The prevalence rates of uveitis, anterior uveitis and non-anterior uveitis were 17.3, 15.0 and 2.3 per 10 000 persons, respectively, during the period of 2007–2013. Increasing age, male sex, residing in a relatively rural area and high income were associated with uveitis, along with Behcet’s disease, juvenile idiopathic arthritis, ankylosing spondylitis, systemic lupus erythematous, ulcerative colitis and tuberculosis.ConclusionsThe overall incidence of uveitis in Korea was similar to those reported in Taiwan and the USA. Despite a potentially inaccurate disease definition, claims data may be useful for monitoring the substantial uveitis burden in South Korea.

BackgroundAlthough several studies showed the association between stroke and open-angle glaucoma (OAG), there is still lack of longitudinal studies based on large populations. Therefore, in this study, we investigated the risk of stroke after OAG diagnosis over a 10-year follow-up period.MethodsWe performed a retrospective nationwide propensity score-matched cohort study. OAG and comparison groups were selected from a large database from the Korean National Health Insurance Service, comprising 1 025 340 random subjects. The OAG group comprised patients with an initial diagnosis of OAG between January 2004 and December 2007 (n=1520), and the comparison group comprised randomly selected patients (five per glaucoma patient; n=7570). Each cohort was tracked until 2013 for stroke development. Cox proportional hazard regression analysis was performed to determine possible associationResultsOAG was associated with increased stroke incidence (HR=1.20, 95% CI 1.03 to 1.40). Hypertension, diabetes mellitus, chronic renal failure, atrial fibrillation, hyperlipidaemia, increasing age and male gender also increased the incidences of stroke. Risk of stroke for patients with OAG was greater in the older age group (≥65 years, HR=1.23, 95% CI 1.02 to 1.47) than in the younger age group (<65 years, HR=1.12, 95% CI 0.86 to 1.46), and greater in males (HR=1.31, 95% CI 1.06 to 1.60) than in females (HR=1.10, 95% CI 0.88 to 1.38).ConclusionsPatients who were diagnosed with OAG were more likely to experience subsequent stroke than comparison group without OAG, and the risk was greater for older adults and males.

BackgroundFew longitudinal studies have evaluated the relationship between cigarette smoking and risk of neovascular age-related macular degeneration (AMD) among Asian populations. This study aimed to prospectively evaluate the association between cigarette smoking and risk of neovascular AMD among Korean men.MethodsMen between the ages of 45 and 79 years included in the Korea National Health Insurance Service database from 2002 through 2013. We compared hazard ratios (HR) for neovascular AMD between 64 560 past/current and 64 560 never smokers by 1:1 propensity-matched analysis and 85 267 past/current and 72 347 never smokers by unmatched cohort and propensity-adjusted analysis.ResultsThe risk of neovascular AMD among past/current smokers was 50% higher than that among never smokers (propensity-adjusted whole cohort analysis: HR, 1.48; 95% CI 1.22 to 1.79; propensity-matched analysis: HR, 1.50; 95% CI 1.22 to 1.84), with the risk more pronounced among current than past smokers (current vs past smokers: propensity-adjusted whole cohort analysis, HR, 1.66; 95% CI 1.35 to 2.04 vs HR, 1.15, 95% CI 0.87 to 1.52; propensity-matched analysis, HR, 1.65; 95% CI 1.32 to 2.05 vs HR, 1.21; 95% CI 0.90 to 1.63). Duration of smoking and daily cigarette consumption was associated with the incidence of neovascular AMD in a dose-dependent manner (p<0.001 for trend).ConclusionsCigarette smoking is associated with a strong risk of neovascular AMD among Korean men. These data highlight the public health impact of smoking on blindness in Asia.

Supplemental Digital Content is Available in the Text.In this nationwide retrospective propensity score matched cohort study, ranibizumab treatment for neovascular age-related macular degeneration did not increase the risk of stroke compared to comorbidity-matched controls or sociodemographic-matched controls.

Preoperative apparent diffusion coefficient histogram analysis is a useful imaging biomarker that holds incremental prognostic value over methylguanine-DNA methyltransferase status after surgery for predicting overall and progression-free survival in patients with glioblastoma.

Introduction:Currently in the United Kingdom, cardiovascular disease (CVD) risk assessment is based on the QRISK3 score, with a benchmark of 10% for 10-year CVD risk determining clinical intervention. Yet, effects on clinical practice are limited and the barriers call for a simple, non-invasive risk stratification tool. Retinal photography is becoming increasingly acceptable as a non-invasive imaging tool for CVD. Previously, we developed a novel CVD risk stratification system based on retinal photographs predicting future CVD risk. This study aimed to further validate our biomarker, Reti-CVD, 1)to detect risk group of ≥10% in 10-year CVD risk and 2)enhance risk assessment in individuals with QRISK3 of 7.5%-10% (termed as borderline-QRISK3 group) using the UK Biobank.Methods:RetiCVD scores were calculated and stratified into three risk groups based on optimized cut-off values from the UK Biobank. We used Cox proportional-hazards models to evaluate the ability of RetiCVD to predict CVD events in the general population. C-statistics was used to assess the prognostic value of adding RetiCVD to QRISK3 in borderline-QRISK3 group and three vulnerable subgroups.Results:Among 45,233 participants with no history of CVD, 6.7% had CVD events during the 11-year follow-up. RetiCVD was associated with an increased risk of CVD (adjusted hazard ratio [HR] 1.34 (95% confidence interval [CI], 1.26-1.42) with a 14.7% (95% CI, 13.6-15.9%) 10-year CVD risk in RetiCVD-high-risk group. The 10-year CVD risk of the borderline-QRISK3 group was greater than 10% in RetiCVD-high-risk group (12.8% in non-statin cohort [n=16240], 11.5% in stage 1 hypertension cohort [n=5102], and 14.9% in middle-aged cohort [n=11,474]). C statistics increased by 0.013 (0.010-0.017) in non-statin cohort, 0.017 (0.010-0.024) in stage 1 hypertension cohort, and 0.022 (0.017-0.027) in middle-aged cohort for CVD event prediction after adding RetiCVD to QRISK3.Conclusion:RetiCVD has the potential to identify individuals with ≥10% 10-year CVD risk who are likely to benefit from earlier, upstream preventative CVD interventions. For borderline-QRISK3 individuals, RetiCVD could be used as a risk enhancer tool to help improve discernment accuracy, especially in early vulnerable adult groups.

Background:In our previous study, we developed a deep-learning-based novel cardiovascular disease (CVD) risk stratification system based on retinal photographs, Reti-CVD. This study aims to further validate Reti-CVD in the US population of the Age-Related Eye Disease Studies (AREDS).Methods:Our testing cohort comprises of AREDS participants who have undergone retinal photography. We evaluated the ability of Reti-CVD to predict atherosclerotic CVD (ASCVD) events (non-fatal and fatal events, and fatal events) using Cox proportional-hazards models. Reti-CVD scores were then calculated and stratified into three groups based on optimized cut-off values from AREDS. Upon further stratifying Reti-CVD scores, we evaluated whether adding Reti-CVD to traditional risk factors could improve risk prediction.Results:Among 3,555 participants, 282 (7.9%) had non-fatal and fatal ASCVD events, and 84 (2.4%) had fatal ASCVD during the 13-year follow-up. Reti-CVD was significantly associated with an increased risk of ASCVD, demonstrating an adjusted hazard ratio (HR) trend of 1.13 (95% CI, 1.03-1.24) for non-fatal and fatal ASCVD events, and an adjusted HR trend of 1.27 (1.07-1.52) for fatal ASCVD events. Reti-CVD significantly improved the overall predictive performance of traditional risk models with a continuous net reclassification index (NRI) of 0.247 (0.106-0.364) for non-fatal and fatal ASCVD events, and 0.232 (0.107-0.359) for fatal ASCVD events. In heatmaps, traditional risk features such as arteriovenous nicking and arteriolar narrowing were well-detected by the Reti-CVD algorithm.Conclusions:A deep-learning-based, retinal photograph-derived new CVD biomarker, Reti-CVD, can be utilized as a risk stratification tool in predicting ASCVD risk in the US population. With the improvements observed on the current risk stratification by traditional risk models, Reti-CVD has considerable potential as an enhanced CVD risk stratification tool applicable in the general US population.

Introduction:The Korean Ministry of Food and Drug Safety (K-MFDS) has authorized over 110 artificial intelligence (AI)-software as medical devices (AI-SaMDs) for diagnostic purposes. Herein, we clinically validate the efficacy of a three-tier cardiovascular disease (CVD) risk stratification system developed using Reti-CVD, an AI-SaMD that utilizes retinal images to estimate CVD risk. The aim of this study is to evaluate the efficacy of RetiCVD in predicting future CVD risk based on retrospective analysis of a prior prospective cohort study.Methods:This clinical study was a single-center, retrospective, conformity-design, confirmatory analysis of prior prospective cohort study. For primary endpoint, retinal images were evaluated by Reti-CVD to validate its three-tier CVD risk stratification system. For secondary endpoints, cardiac CT-measured coronary artery calcium (CAC, 0, >0-100, and >100), carotid intima-media thickness (CIMT, <90thand ≥90thpercentile), and brachial-ankle pulse wave velocity (baPWV, <1800 and ≥1800 cm/s) were also measured as independent variables for future CVD risk. The cumulative incidence of non-fatal and/or fatal CVD events was evaluated. Cox proportional hazards models were used to estimate the hazard ratios (HR) trends.Results:In this clinical cohort (n=1106), 33 (3.0%) participants had non-fatal or fatal CVD events over 5 years, and RetiCVD scores were significantly associated with increased CVD risk (HR trend=2.02, 95% CI, 1.26-3.24). In a multivariable Cox model incorporating RetiCVD scores, CAC, CIMT, baPWV, and other traditional risk factors, RetiCVD scores for the high-risk tier were significantly associated with increased CVD risk (low risk as a reference; HR=3.56 [1.34-9.51] in high risk), while other biomarkers showed trends toward association: CAC showed a HR of 2.45 [0.88-6.84] in CAC of >100 compared to zero CAC (reference); CIMT showed a HR of 1.50 (0.64-3.51) in CIMT of ≥90thpercentile compared to <90thpercentile; and baPWV showed a HR of 1.27 (0.53-3.03) in baPWV of ≥1800 compared to <1800 cm/s.Conclusions:This is the first K-MFDS approved AI-software that can potentially stratify future risk as biomarkers, which can have major implications such as identifying high-risk individuals.

To describe the normative quantitative parameters of macular retinal vasculature and their systemic and ocular associations, using optical coherence tomography angiography (OCTA).

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.