High-altitude chronic hypoxia can induce excessive erythrocytosis (EE, defined as a haemoglobin concentration of ≥21 g/dL in men), leading to hyperviscosity and promoting endothelial dysfunction. We aimed to assess whether EE affects the retinal vascular phenotype and the ophthalmological vascular response to CO2. We conducted an ophthalmological cross-sectional study among highlanders permanently living at 5100 m (La Rinconada, Peru). The central retinal artery equivalent (CRAE), central retinal vein equivalent (CRVE) and retinal vessel tortuosity were measured using semi-automatic imaging software (VAMPIRE) from the diameters of the six largest arteries and veins on fundus images. Choroidal blood flow was assessed using laser Doppler flowmetry. Measurements were performed at rest and during a hypercapnic challenge (+10.1 ± 1.4 mmHg end-tidal CO2). Among the 62 included highlanders, 38 (61%) had EE. Compared with non-EE, highlanders with EE exhibited higher CRVE (278 ± 25 vs. 249 ± 24 µm, P < 0.001), with no other significant ophthalmological differences. Resting CRVE was significantly correlated (all P-values < 0.001) with haematocrit (r = 0.56), haemoglobin concentration (r = 0.65), blood volume (r = 0.55) and the arterial partial pressure of carbon dioxide (r = 0.46). Hypercapnia led to a moderate overall decrease in CRVE (-8.6 ± 22.0 µm, P = 0.02), without a specific effect of EE, and induced no other retinal vascular or choroidal blood flow changes in either group. We observed larger retinal vein diameters in EE highlanders compared with non-EE. Although hypercapnia is known to increase retinal vessel diameter in healthy lowlanders, it selectively decreased CRVE in highlanders, irrespective of EE status. These findings suggest a retinal vascular dysfunction in highlanders, probably induced by chronic exposure to severe hypoxia.

Morphological measurements of retinal vessel parameters are commonly obtained from fundus images using semi-automatic software packages. Since image quality, and image processing methods are potential sources of variation of such measurements, the aim of this study was to determine quantitatively the effect of using different fundus cameras and image resolutions on measurements of retinal vascular parameters. Fifty-four digital fundus images of 27 healthy subjects were acquired using two non-mydriatic cameras, Topcon TRC NW6S and Canon CR-2. Central retinal artery and vein equivalent (CRAE, CRVE), arteriolar and venular tortuosity (TortA and TortV) and fractal dimension (FD) were calculated using VAMPIRE software. First, VAMPIRE measurements were compared between Topcon images (3008 × 2000 pixels) and Canon images after resizing manually (CR2r, 3008 × 2000 px). The effect of different resolutions was studied using Canon images at full resolution (4147 × 2764 px, 100%) and then scaled at 83.8%, 75% and 50% in JPEG format using the dedicated Canon software. When comparing images acquired by CR2r and Topcon at the same resolution, all parameters were significantly affected. When using images with different resolutions from the same camera, significant increases of CRAE and CRVE and significant decreases of FD were found for decreasing resolutions. In conclusion, vascular measurements from images from different fundus cameras or at different resolutions are not, in general, directly comparable.

Background Identifying high cardiovascular (CV) risk in type 2 diabetes is critical, as cardiovascular diseases (CVD) drives most morbidity and mortality. Biomarkers such as NT-proBNP and high-sensitivity cardiac troponins detect subclinical cardiac stress and inform personalized prevention. Emerging evidence links retinal vessel calibre and microvascular features from fundus images to alterations in left ventricular structure and function. This study examined whether quantitative retinal vascular features (RVFs) in type 2 diabetes mellitus (T2DM) are associated with these cardiac biomarkers. Methods This study included 421 individuals with T2DM (mean age 61.17 ± 6.81 years) from the PRIME Cohort, part of The Malaysian Cohort project. All participants underwent retinal imaging and cardiac biomarker assessment, including high-sensitivity troponin T (HS-troponin) and N-terminal pro–B-type natriuretic peptide (NT-proBNP). Macula-centred retinal images were analyzed using the Vascular Assessment and Measurement Platform for Images of the Retina (VAMPIRE) to evaluate RVFs, including central retinal arteriole/venule equivalent (CRAE, CRVE), arteriovenous ratio (AVR), arteriole/venule tortuosity (TORTA/V), and arteriole/venule fractal dimension (FDA/V). Incident CVD was recorded at baseline and during follow-up. Linear regression was used to analyse the relationships between RVFs and cardiac markers (HS-troponin and NTP-proBNP), while logistic regression was used to examine associations between RVFs and incident CVD. Results After adjusting for age and gender, AVR (adjusted β (aβ) − 0.709; 95% CI − 1.307, − 0.112; P = 0.020), FDA (aβ − 1.202; 95% CI − 1.849, − 0.555; P < 0.001), and FDV (aβ − 1.148; 95% CI − 1.786, − 0.509; P < 0.001) were significantly associated with HS-troponin. FDA (aβ − 26.225; 95% CI − 44.992, − 7.528; P = 0.006), FDV (aβ − 23.501; 95% CI − 41.956, − 5.045; P = 0.013), and TORTV (aβ 25.814; 95% CI 2.292, 49.337; P = 0.032) were significantly associated with NT-proBNP. Conclusions Fractal dimensions of arterioles and venules were associated with both HS-troponin and NT-proBNP, while venule tortuosity was associated only with NT-proBNP. Our study provides evidence that machine learning-derived retinal vessel features are associated with cardiac biomarkers of alterations in left ventricular structure and function.

Purpose:The aim of this study was to characterize changes in retinal vessel diameters and choroidal blood flow in healthy lowlanders during a high-altitude expedition. Methods:Ocular examination, fundus images acquired using a handheld camera, and laser Doppler flowmetry (LDF) measurements within the subfoveal choroid (blood flow = ChBF, blood velocity = ChVel, and blood volume = ChVol) were carried out at 200 m and after 9 days at 5100 m in 11 healthy participants. Fundus images were analyzed with the semi-automatic software Vessel Assessment and Measurement Platform for Images of the Retina (VAMPIRE) version 3.2 to quantify retinal vessel parameters: the central retinal artery equivalent (CRAE), the central retinal vein equivalent (CRVE), and arterial and venular tortuosity. Hematocrit and hemoglobin concentrations were also measured at both altitudes. Results:Corneal thickness increased slightly at altitude (median = 536 µm, interquartile range = 25-75%: [521-571] at 200 m vs. 561 µm [540-574] at 5100 m, P = 0.004). No participant was affected by high-altitude retinopathy. From 200 m to 5100 m, ChVol and ChBF decreased significantly (-31% [43-22], P = 0.003 and -13% [22-8], P = 0.01, respectively), ChVel increased (+17% [10-44], P = 0.003), and CRVE (+10% [3-14], P = 0.04) and venular tortuosity (+142% [71-168], P = 0.04) increased significantly. The altitude-induced increase in hematocrit correlated negatively with the decrease in ChBF (r = -0.88, P < 0.001) and positively with the increase in CRVE (r = 0.88, P = 0.01). Conclusions:Acute high-altitude exposure leads to a decrease of ChBF (partly related to a decrease in blood volume) and an increase in retinal vein diameter and tortuosity. The physiological consequences of these changes on retinal blood flow and retinal function remain to be explored.

We present the SEoul Retinal Vessel Assessment Library (SERVAL), a novel software platform for precise quantitative measurement of vascular structures in fundus images. SERVAL integrates deep learning-based automatic artery and vein mask initialization, subpixel vessel centerline and boundary refinement, and interactive editing tools within a user-friendly graphical interface. From the refined artery and vein delineations, it enables accurate computation of a wide range of vessel assessment metrics, facilitating better characterization of complex vascular structures. We evaluate SERVAL through: (1) comparative analyses with existing platforms, highlighting its superior precision and structural detail; (2) longitudinal image studies demonstrating measurement consistency; and (3) a usability study confirming its clinical practicality. We expect SERVAL to serve as a valuable tool in clinical research, supporting the development of novel vascular biomarkers and diagnostic metrics for retinal and systemic diseases.

To investigate the associations between retinal vessel parameters and normal-tension glaucoma (NTG). We conducted a case–control study with a prospective cohort, allowing to record 23 cases of NTG. We matched NTG patient with one primary open-angle glaucoma (POAG) and one control per case by age, systemic hypertension, diabetes, and refraction. Central retinal artery equivalent (CRAE), central retinal venule equivalent (CRVE), Arteriole-To-Venule ratio (AVR), Fractal Dimension and tortuosity of the vascular network were measured using VAMPIRE software. Our sample consisted of 23 NTG, 23 POAG, and 23 control individuals, with a median age of 65 years (25–75th percentile, 56–74). No significant differences were observed in median values for CRAE (130.6 µm (25–75th percentile, 122.8; 137.0) for NTG, 128.4 µm (124.0; 132.9) for POAG, and 135.3 µm (123.3; 144.8) for controls, P = .23), CRVE (172.1 µm (160.0; 188.3), 172.8 µm (163.3; 181.6), and 175.9 µm (167.6; 188.4), P = .43), AVR (0.76, 0.75, 0.74, P = .71), tortuosity and fractal parameters across study groups. Vascular morphological parameters were not significantly associated with retinal nerve fiber layer thickness or mean deviation for the NTG and POAG groups. Our results suggest that vascular dysregulation in NTG does not modify the architecture and geometry of the retinal vessel network.

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.

We aimed to compare measurements from three of the most widely used software packages in the literature and to generate conversion algorithms for measurement of the central retinal artery equivalent (CRAE) and central retinal vein equivalent (CRVE) between SIVA and IVAN and between SIVA and VAMPIRE. We analyzed 223 retinal photographs from 133 human participants using both SIVA, VAMPIRE and IVAN independently for computing CRAE and CRVE. Agreement between measurements was assessed using Bland–Altman plots and intra-class correlation coefficients. A conversion algorithm between measurements was carried out using linear regression, and validated using bootstrapping and root-mean-square error. The agreement between VAMPIRE and IVAN was poor to moderate: The mean difference was 20.2 µm (95% limits of agreement, LOA, −12.2–52.6 µm) for CRAE and 21.0 µm (95% LOA, −17.5–59.5 µm) for CRVE. The agreement between VAMPIRE and SIVA was also poor to moderate: the mean difference was 36.6 µm (95% LOA, −12.8–60.4 µm) for CRAE, and 40.3 µm (95% LOA, 5.6–75.0 µm) for CRVE. The agreement between IVAN and SIVA was good to excellent: the mean difference was 16.4 µm (95% LOA, −4.25–37.0 µm) for CRAE, and 19.3 µm (95% LOA, 0.09–38.6 µm) for CRVE. We propose an algorithm converting IVAN and VAMPIRE measurements into SIVA-estimated measurements, which could be used to homogenize sets of vessel measurements obtained with different software packages.

PURPOSE:To identify the retinal vessel vasculature parameters associated with birdshot chorioretinopathy (BSCR).METHODS:This retrospective observational study included 28 prevalent cases of BSCR with a median time from diagnosis of 6 years and 28 controls matched for age, arterial hypertension, diabetes and refraction. Forty-five-degree fundus images of both dilated eyes were acquired with a fundus camera (Canon CR-2, Tokyo, Japan). The summary diameter of the arterial retinal vessels (central retinal artery equivalent, CRAE), venous retinal vessels (central retinal vein equivalent, CRVE), vascular tortuosity and fractal dimension (FD) were measured using VAMPIRE software. Retinal vasculitis was characterized using fluorescein angiography and active choroiditis using indocyanine green angiography.RESULTS:At baseline, BSCR was associated with lower FD compared with matched controls (mean difference, -0.04; 95% confidence interval [CI], -0.06 to -0.02, p < 0.001). No other VAMPIRE parameters (CRAE, CRVE, arterial and venous tortuosity) differed. Among BSCR patients, retinal vein vasculitis was associated with higher CRAE (mean difference, 21 μ; 95% CI, 2.6-40, p = 0.03), venous tortuosity (geometric mean ratio, 1.79; 95% CI, 1.18-2.72, p = 0.007) and FD (mean difference, -0.04; 95% CI, -0.06 to -0.01, p = 0.007). Resolution of retinal vein vasculitis during follow-up was paralleled by decrease in CRAE, CRVE and venous tortuosity values and increase in venous FD, respectively.CONCLUSION:BSCR is associated with lower FD value, suggesting that chronic retinal inflammation induces microvascular remodelling. Efficient treatment of retinal vasculitis may reverse changes in retinal vascular parameters. Changes in retinal vascular parameters could be potentially useful for assessing patients with BSCR disease.

Objectives Improved identification of individuals with type 2 diabetes at high cardiovascular risk could help in selection of newer cardiovascular risk-reducing therapies. The aim of this study was to determine whether retinal vascular parameters, derived from retinal screening photographs, alone and in combination with a genome-wide polygenic risk score for coronary heart disease (CHD PRS) would have independent prognostic value over traditional CV risk assessment in patients without prior cardiovascular disease. Research Design and Methods Patients in the GoDARTS study were linked to retinal photographs, prescriptions, and outcomes. Retinal photographs were analysed using VAMPIRE software, a semi-automated AI platform, to compute arterial and venous fractal dimension, tortuosity and diameter. CHD PRS was derived from previously published data. Multivariable Cox regression was used to evaluate the association between retinal vascular parameters and major adverse cardiovascular events (MACE) at 10 years compared to the pooled cohort equations (PCE) risk score. Results 5,152 individuals were included. 1,017 individuals suffered a MACE. Reduced arterial fractal dimension and diameter and increased venous tortuosity each independently predicted MACE. A risk score combining these parameters significantly predicted MACE after adjustment for age, sex, PCE and the CHD PRS (HR 1.11 per SD increase; 95% CI 1.04-1.18, p=0.002) with similar accuracy to PCE (AUC 0.663 vs. 0.658, p=0.33). A model incorporating retinal parameters and PRS improved MACE prediction compared to PCE (AUC 0.686 vs. 0.658, p<0.001). Conclusions Retinal parameters alone and in combination with genome-wide CHD PRS have independent and incremental prognostic value compared to traditional CV risk assessment in type 2 diabetes.

The aim of this prospective study was to compare retinal vascular diameter measurements taken from standard fundus images and adaptive optics (AO) images. We analysed retinal images of twenty healthy subjects with 45-degree funduscopic colour photographs (CR-2 Canon fundus camera, Canon™) and adaptive optics (AO) fundus images (rtx1 camera, Imagine Eyes®). Diameters were measured using three software applications: the VAMPIRE (Vessel Assessment and Measurement Platform for Images of the REtina) annotation tool, IVAN (Interactive Vessel ANalyzer) for funduscopic colour photographs, and AO_Detect_Artery™ for AO images. For the arterial diameters, the mean difference between AO_Detect_Artery™ and IVAN was 9.1 µm (−27.4 to 9.2 µm, p = 0.005) and the measurements were significantly correlated (r = 0.79). The mean difference between AO_Detect_Artery™ and VAMPIRE annotation tool was 3.8 µm (−34.4 to 26.8 µm, p = 0.16) and the measurements were poorly correlated (r = 0.12). For the venous diameters, the mean difference between the AO_Detect_Artery™ and IVAN was 3.9 µm (−40.9 to 41.9 µm, p = 0.35) and the measurements were highly correlated (r = 0.83). The mean difference between the AO_Detect_Artery™ and VAMPIRE annotation tool was 0.4 µm (−17.44 to 25.3 µm, p = 0.91) and the correlations were moderate (r = 0.41). We found that the VAMPIRE annotation tool, an entirely manual software, is accurate for the measurement of arterial and venular diameters, but the correlation with AO measurements is poor. On the contrary, IVAN, a semi-automatic software tool, presents slightly greater differences with AO imaging, but the correlation is stronger. Data from arteries should be considered with caution, since IVAN seems to significantly under-estimate arterial diameters.

Background Chest x-rays are widely used in clinical practice; however, interpretation can be hindered by human error and a lack of experienced thoracic radiologists. Deep learning has the potential to improve the accuracy of chest x-ray interpretation. We therefore aimed to assess the accuracy of radiologists with and without the assistance of a deep learning model. Methods In this retrospective study, a deep-learning model was trained on 821 681 images (284 649 patients) from five data sets from Australia, Europe, and the USA. 2568 enriched chest x-ray cases from adult patients (>= 16 years) who had at least one frontal chest x-ray were included in the test dataset; cases were representative of inpatient, outpatient, and emergency settings. 20 radiologists reviewed cases with and without the assistance of the deep-learning model with a 3-month washout period. We assessed the change in accuracy of chest x-ray interpretation across 127 clinical findings when the deep-learning model was used as a decision support by calculating area under the receiver operating characteristic curve (AUC) for each radiologist with and without the deep-learning model. We also compared AUCs for the model alone with those of unassisted radiologists. If the lower bound of the adjusted 95% CI of the difference in AUC between the model and the unassisted radiologists was more than -0middot05, the model was considered to be non-inferior for that finding. If the lower bound exceeded 0, the model was considered to be superior. Findings Unassisted radiologists had a macroaveraged AUC of 0middot713 (95% CI 0middot645-0middot785) across the 127 clinical findings, compared with 0middot808 (0middot763-0middot839) when assisted by the model. The deep-learning model statistically significantly improved the classification accuracy of radiologists for 102 (80%) of 127 clinical findings, was statistically non-inferior for 19 (15%) findings, and no findings showed a decrease in accuracy when radiologists used the deep learning model. Unassisted radiologists had a macroaveraged mean AUC of 0middot713 (0middot645-0middot785) across all findings, compared with 0middot957 (0middot954-0middot959) for the model alone. Model classification alone was significantly more accurate than unassisted radiologists for 117 (94%) of 124 clinical findings predicted by the model and was non-inferior to unassisted radiologists for all other clinical findings. Interpretation This study shows the potential of a comprehensive deep-learning model to improve chest x-ray interpretation across a large breadth of clinical practice. Funding Annalise.ai. Copyright (c) 2021 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license.

Introduction: The aim of the study was to estimate the phenotype of retinal vessels using central retinal artery equivalent (CRAE), central retinal vein equivalent (CRVE), tortuosity, and fractal analysis in the unaffected contralateral eye of patients with central or branch retinal vein occlusion (CRVO or BRVO). Methods: Thirty-four patients suffering from CRVO, 15 suffering from BRVO, and 49 controlled matched subjects had a fundus image analyzed using the VAMPIRE software. The intraclass correlation coefficient and a Bland-Altman plot were done for the reproducibility study. Results: There was a lack of evidence of difference between the control group and the CRVO group for CRAE (p = 0.06), CRVE (p = 0.3), and arterio-venule ratio (AVR, p = 0.6). Contralateral eyes of CRVO exhibited a significantly higher arterial and minimum arterial tortuosity values (p = 0.012), as compared with control eyes. Contralateral eyes of patients with a history of BRVO had a significantly higher CRAE (p = 0.02), AVR (p = 0.006), and minimal arterial tortuosity (p = 0.05). Fractal analysis showed that contralateral eyes of BRVO had higher values of fractal parameters (D0a, p = 0.005). Conclusion: This study suggests that CVRO or BRVO is not triggered by the same retinal vascular phenotypes in the contralateral eye. The morphology of retinal vasculature may be associated with the occurrence of RVO, independently of known risk factors.

Background and Objectives: This paper reports a quantitative analysis of the effects of joint photographic experts group (JPEG) image compression of retinal fundus camera images on automatic vessel segmentation and on morphometric vascular measurements derived from it, including vessel width, tortuosity and fractal dimension. Methods: Measurements are computed with vascular assessment and measurement platform for images of the retina (VAMPIRE), a specialized software application adopted in many international studies on retinal biomarkers. For reproducibility, we use three public archives of fundus images (digital retinal images for vessel extraction (DRIVE), automated retinal image analyzer (ARIA), high-resolution fundus (HRF)). We generate compressed versions of original images in a range of representative levels. Results: We compare the resulting vessel segmentations with ground truth maps and morphological measurements of the vascular network with those obtained from the original (uncompressed) images. We assess the segmentation quality with sensitivity, specificity, accuracy, area under the curve and Dice coefficient. We assess the agreement between VAMPIRE measurements from compressed and uncompressed images with correlation, intra-class correlation and Bland-Altman analysis. Conclusions: Results suggest that VAMPIRE width-related measurements (central retinal artery equivalent (CRAE), central retinal vein equivalent (CRVE), arteriolar-venular width ratio (AVR)), the fractal dimension (FD) and arteriolar tortuosity have excellent agreement with those from the original images, remaining substantially stable even for strong loss of quality (20% of the original), suggesting the suitability of VAMPIRE in association studies with compressed images. ? 2021 Elsevier B.V. All rights reserved. Background and Objectives: This paper reports a quantitative analysis of the effects of joint photographic experts group (JPEG) image compression of retinal fundus camera images on automatic vessel segmentation and on morphometric vascular measurements derived from it, including vessel width, tortuosity and fractal dimension. Methods: Measurements are computed with vascular assessment and measurement platform for images of the retina (VAMPIRE), a specialized software application adopted in many international studies on retinal biomarkers. For reproducibility, we use three public archives of fundus images (digital retinal images for vessel extraction (DRIVE), automated retinal image analyzer (ARIA), high-resolution fundus (HRF)). We generate compressed versions of original images in a range of representative levels. Results: We compare the resulting vessel segmentations with ground truth maps and morphological measurements of the vascular network with those obtained from the original (uncompressed) images. We assess the segmentation quality with sensitivity, specificity, accuracy, area under the curve and Dice coefficient. We assess the agreement between VAMPIRE measurements from compressed and uncompressed images with correlation, intra-class correlation and Bland-Altman analysis. Conclusions: Results suggest that VAMPIRE width-related measurements (central retinal artery equivalent (CRAE), central retinal vein equivalent (CRVE), arteriolar-venular width ratio (AVR)), the fractal dimension (FD) and arteriolar tortuosity have excellent agreement with those from the original images, remaining substantially stable even for strong loss of quality (20% of the original), suggesting the suitability of VAMPIRE in association studies with compressed images.

The eye affords a unique opportunity to inspect a rich part of the human microvasculature non-invasively via retinal imaging. Retinal blood vessel segmentation and classification are prime steps for the diagnosis and risk assessment of microvascular and systemic diseases. A high volume of techniques based on deep learning have been published in recent years. In this context, we review 158 papers published between 2012 and 2020, focussing on methods based on machine and deep learning (DL) for automatic vessel segmentation and classification for fundus camera images. We divide the methods into various classes by task (segmentation or artery-vein classification), technique (supervised or unsupervised, deep and non-deep learning, hand-crafted methods) and more specific algorithms (e.g. multiscale, morphology). We discuss advantages and limitations, and include tables summarising results at-a-glance. Finally, we attempt to assess the quantitative merit of DL methods in terms of accuracy improvement compared to other methods. The results allow us to offer our views on the outlook for vessel segmentation and classification for fundus camera images.

The eye a ords a unique opportunity to inspect a rich part of the human microvasculature non-invasively via retinal imaging. Retinal blood vessel segmentation and classi cation are prime steps for the diagnosis and risk assessment of microvascular and systemic diseases. A high volume of techniques based on deep learning have been published in recent years. In this context, we review 158 papers published between 2012 and 2020, focussing on methods based on machine and deep learning (DL) for automatic vessel segmentation and classi cation for fundus camera images. We divide the methods into various classes by task (segmentation or artery-vein classi cation), technique class (supervised or unsupervised, deep and non-deep learning, hand-crafted methods) and more speci c algorithms (e.g. multiscale, morphology). We discuss advantages and limitations, and include tables summarising results at-a-glance. Finally, we attempt to assess the quantitative merit of DL methods in terms of accuracy improvement compared to other methods. The results allow us to o er our views on the outlook for vessel segmentation and classi cation for fundus camera images.

Cardiovascular diseases are a public health concern; they remain the leading cause of morbidity and mortality in patients with type 2 diabetes. Phenotypic information available from retinal fundus images and clinical measurements, in addition to genomic data, can identify relevant biomarkers of cardiovascular health. In this study, we assessed whether such biomarkers stratified risks of major adverse cardiac events (MACE). A retrospective analysis was carried out on an extract from the Tayside GoDARTS bioresource of participants with type 2 diabetes (n = 3,891). A total of 519 features were incorporated, summarising morphometric properties of the retinal vasculature, various single nucleotide polymorphisms (SNPs), as well as routine clinical measurements. After imputing missing features, a predictive model was developed on a randomly sampled set (n = 2,918) using L1-regularised logistic regression (lasso). The model was evaluated on an independent set (n = 973) and its performance associated with overall hazard rate after censoring (log-rank p < 0.0001), suggesting that multimodal features were able to capture important knowledge for MACE risk assessment. We further showed through a bootstrap analysis that all three sources of information (retinal, genetic, routine clinical) offer robust signal. Particularly robust features included: tortuousity, width gradient, and branching point retinal groupings; SNPs known to be associated with blood pressure and cardiovascular phenotypic traits; age at imaging; clinical measurements such as blood pressure and high density lipoprotein. This novel approach could be used for fast and sensitive determination of future risks associated with MACE.