A Retinal Oct-Angiography and Cardiovascular STAtus (RASTA) Dataset on Swept-Source Microvascular Imaging for Cardiovascular Risk Assessment

1. Summary

Cardiovascular diseases (CVD) remain the leading cause of death worldwide with 9 million deaths from heart disease reported in 2019 [1]. Pathophysiological mechanisms involved in the development of CVD begin years before the appearance of any symptoms [2]. Thus, researchers have been investigating early biomarkers to help screen and diagnose CVD before the onset of symptoms or major cardiovascular events. The retinal vascular network could be a good candidate since the retinal microvasculature may share the same physiological and anatomical characteristics as the cerebral and coronary microvasculature [3]. Associations between retina vascular features and CVD were first demonstrated with fundus photographs [4,5]. These associations were subsequently confirmed with other retinal imaging such as retinal swept source optical coherence tomography-angiography (SS OCT-A) [6,7]. SS OCT-A enables noninvasive assessment of the retinal microvascular network. It is thus possible to study the different vascular plexi (superficial capillary plexus, deep capillary plexus, and choriocapillaris plexus) and the avascular zone using quantitative data. Quantification of retinal vascular density by SS OCT-A could therefore be compared to a window into the integrity of the systemic microcirculation.

The cardiovascular risk profile of patients can be estimated with numerous score models such as the Framingham Risk score (FRS) for 10-year CVD risk calculation, the Pooled Cohort Equations (PCE), the American Heart Association risk score (AHA risk score) for a moderate-risk population, or the SCORE2 to predict the 10-year risk of first-onset CVD in European populations [8,9,10,11]. The CHA₂DS₂-VASc clinical score, which is universally known and easy to calculate, is an embolic risk stratification tool originally used to assess the risk of stroke in patients with non-valvular atrial fibrillation [12]. It has been recently presented as an effective model for evaluating the cardiovascular risk profile regardless of the arrhythmic status of patients [13,14,15,16,17,18,19]. Several datasets containing images of retinal fundus photographs are publicly available (i.e. MESSIDOR, STARE project, DRIVE, E-ophtha, and EyePACS) [20,21,22,23,24]. However, SS OCT-A datasets are less widespread [25]. To the best of our knowledge, the Retinal oct-Angiography and cardiovascular STAtus (RASTA) dataset is the first publicly available dataset that provides systematic cardiovascular data and complete SS OCT-A retinal imaging. The RASTA dataset is hosted on https://rasta.u-bourgogne.fr/.

2. Ethics Approval

The RASTA dataset was acquired in the Department of Ophthalmology at the University Hospital of Dijon, France, and consists of actual clinical acquisitions from different registered clinical studies. The RASTA dataset was anonymized and processed in accordance with the rules established by the Ethics Committee of the University Hospital of Dijon. All administrative information included in the metadata has been removed, making it untraceable. Thus, in accordance with the French law it was not necessary to obtain ethical approval.

3. Data Description

3.1. Data Composition

The RASTA dataset is a new publicly available SS OCT-A retinal image dataset consisting of 499 participants for 2005 en face images and 814 angiocubes combined with clinical and demographic characteristics. Information on data accessibility and specifications is provided in Table 1. Each participant was identified by an anonymized ID and was then included in one of three groups according to their cardiovascular risk category as follows:

• Low cardiovascular risk – CHA₂DS₂-VASc = [0; 1]
• Intermediate cardiovascular risk – CHA₂DS₂-VASc = [2; 3]
• High cardiovascular risk – CHA₂DS₂-VASc = [3; 9]

Table 1. Specifications table.

Subject Area	Biomedical Imaging, Ophthalmology
More specific subject area	Retinal OCT-A volumes analysis for cardiovascular risk prediction
Type of data	Image, CSV
How data were acquired	Swept-source OCT-A Instrument name : PLEX Elite 9000® (Carl Zeiss Meditec Inc., Dublin, OH, USA)
Data format	DICOM for volumes, Bitmap for en face images
Experimental factors	Pupillary dilatation with tropicamide 0.5% if signal strength < 8/10
Experimental features	Macular angiography 6 x 6-mm
Main data source location	University Hospital of Dijon, Dijon 21000, France
Data accessibility	https://rasta.u-bourgogne.fr/

Table 1. Specifications table.

Subject Area	Biomedical Imaging, Ophthalmology
More specific subject area	Retinal OCT-A volumes analysis for cardiovascular risk prediction
Type of data	Image, CSV
How data were acquired	Swept-source OCT-A Instrument name : PLEX Elite 9000® (Carl Zeiss Meditec Inc., Dublin, OH, USA)
Data format	DICOM for volumes, Bitmap for en face images
Experimental factors	Pupillary dilatation with tropicamide 0.5% if signal strength < 8/10
Experimental features	Macular angiography 6 x 6-mm
Main data source location	University Hospital of Dijon, Dijon 21000, France
Data accessibility	https://rasta.u-bourgogne.fr/

For each participant, we included the images of their corresponding SS OCT-A 6 x 6-mm angiocubes and en face (two-dimensional) images. Angiocubes were identified on the basis of their side only. En face images (Figure 1) were identified on the basis of their plexus followed by their side as follows:

• « sup » for superficial plexus or « deep » for deep plexus or « cc » for choriocapillaris plexus
• « OD » for right eye or « OS » for left eye

Figure 1. Right eye en face images of (a) superficial plexus, (b) deep plexus and (c) choriocapillaris plexus.

Figure 1. Right eye en face images of (a) superficial plexus, (b) deep plexus and (c) choriocapillaris plexus.

The RASTA dataset is composed of four different single-center studies with the Ophthalmology Department of the University Hospital of Dijon as the principal investigator since 2018 and one multicenter study conducted by 14 investigative health centers since 2021. All of the studies required the collection of cardiovascular history and anthropometric data. The aims of these studies are described as follows:

• « AnomAlies Rétiniennes précoces au cours du Diabète de type 1 » (AwARD; Early Retinal Anomalies in Type 1 Diabetes) [26]: to specify early retinal microvascular abnormalities by measuring the area of the central retinal avascular zone on SS OCT-A images of patients with type 1 diabetes without diabetic retinopathy (ID-RCB: 2017-A02724-49)
  95 eyes of 95 patients, from 02/23/2018 to 02/28/2020
• RETINORM: control group of the AwARD study
  137 eyes of 75 volunteers, from 04/12/2021 to 11/25/2021
• « Retinal Microvascular Changes in Familial Hypercholesterolemia: Analysis with Swept-Source Optical Coherence Tomography Angiography » (FAMILIPO) [27]: to analyze the association between retinal vascular density and the presence of atherosclerosis assessed with the Coronary Artery Calcium score and compare SS OCT-A quantitative parameters between patients with familial hypercholesterolemia (FH) and healthy volunteers from the AwARD study without a history of FH
  162 eyes of 81 patients with FH, from 10/21/2020 to 10/27/2021
• « Obstructive sleep apnea and Retinal vascular NETwork » (ORNET): to describe retinal microvascular characteristics with SS OCT-A in a population with obstructive sleep apnea syndrome (OSAS) and to compare them with healthy volunteers (ID-RCB: 2018-A02204-51)
  159 eyes of 79 patients with OSAS and 62 eyes of 33 volunteers without OSAS, from 07/01/2020 to 02/14/2023
• « Réseau Microvasculaire Rétinien et Chirurgie Cardiaque de revascularisation coronarienne » (MRCC; Retinal Microvascular Network and Coronary Revascularization Cardiac Surgery): to study, in patients scheduled for coronary revascularization cardiac surgery with extracorporeal circulation, the discriminative capacity of retinal vascular density to predict the occurrence of acute renal failure defined by the KDIGO criterion [28] within 7 days of surgery (ID-RCB: 2021-A02895-36)
  33 eyes of 33 patients, from 06/07/2022 to 03/06/2023
• « Giant cell arteritis study » (GIANT): to describe retinal microvasculature on SS OCT-A in patients with giant cell arteritis without ophthalmological symptom
  56 eyes of 40 patients, from 11/21/2017 to 10/18/2022
• « Evaluation intelligente de la Rétinopathie diabétique » (EviRed; Intelligent Assessment of Diabetic Retinopathy): to propose SS OCT-A analysis to better predict the risk of diabetic retinopathy than the current classification of diabetic retinopathy mainly based on fundus photography (ANR: 18-RHUS-0008)
  118 eyes of 63 patients without diabetic retinopathy, from 06/01/2021 to 01/19/2022

A CSV file contains each ID in alphanumerical order with the corresponding characteristics. Each medical diagnosis has been confirmed by a panel of medical experts according to the guidelines of the French National Authority for Health (Haute Autorité de Santé). The information available in the CSV file is illustrated in Figure 2, with the following explanation fo each column:

-

ID: participant’s anonymous identity code

-

Age: age in years at inclusion

-

Sex: 0 if male gender, 1 if female gender

-

Congestive heart failure: presence of heart failure/moderate-severe cardiac dysfunction with left ventricular ejection fraction ≤ 40%

-

Hypertension: presence of hypertension confirmed by ambulatory blood pressure measurement with a systolic blood pressure ≥ 135 mmHg and/or diastolic blood pressure ≥ 85 mmHg

-

Diabetes mellitus: presence of diabetes mellitus confirmed by a single blood glucose sample ≥ 2 g/l or confirmed by a second blood glucose sample ≥ 1.26 g/l when the first one is ≥ 1.26 g/L and < 2 g/L

-

Stroke: prior stroke or transient ischemic attack or thromboembolism

-

Vascular disease: presence of vascular disease (e.g., peripheral artery disease, myocardial infarction, aortic plaque) confirmed by Doppler ultrasonography, coronary angiography/cardiac magnetic resonance imaging (MRI)/myocardial perfusion scintigraphy, or computed tomography angiography, respectively

-

Body mass index: body mass divided by the square of height, in kg/m²

-

CHA₂DS₂-VASc: cardiovascular score prediction

-

Obstructive sleep apnea syndrome: presence of obstructive sleep apnea syndrome confirmed by respiratory polygraphy or polysomnography

-

Smoking: previous or active Smoking

-

Dyslipidemia: presence of dyslipidemia confirmed by two blood samples with HDL-c < 0.35g/L or LDL-c > 1.30 g/L and/or TG > 1.5 g/L for patients with cardiovascular risk, and two blood samples with HDL-c < 0.35g/L or LDL-c > 1.60 g/L and/or TG > 1.5 g/L for patients without cardiovascular risk

-

OD: oculus dexter (right eye)

-

OS: oculus sinister (left eye)

-

Fovea Avascular Zone (FAZ) in superficial plexus:

○

FAZ_RL: raw length (perimeter) of the FAZ in mm

○

FAZ_Ci: circularity index of the FAZ ranging from 0 (most irregular circle shape) to 1 (perfect circle shape)

○

FAZ_RS: raw size (area) of the FAZ in mm²

Figure 2. Sample CSV files.

Figure 2. Sample CSV files.

-

Vessel density (VD): total length of perfused vasculature per unit area in a region of measurement in units of mm^-1. It consists in untangling the entire vasculature in the retina, and measuring its length and then dividing it by the area it originally occupied, ranging from a minimum of 0 (no vessels) and an unbounded maximum

○

Dens_Ave_Sup: VD average in superficial plexus

○

Dens_Circle3mm_Sup: VD in a circle of 3-mm diameter in superficial plexus

○

Dens_Circle6mm_Sup: VD in a circle of 6-mm diameter in superficial plexus

○

Dens_Ave_Deep: VD average in deep plexus

○

Dens_Circle3mm_Deep: VD in a circle of 3-mm diameter in deep plexus

○

Dens_Circle6mm_Deep: VD in a circle of 6-mm diameter in deep plexus

-

Perfusion density (PD): total area of perfused vasculature per unit area in a region of measurement ranging from 0 (no perfusion) to 1 (fully perfused)

○

Perf_Ave_Sup: PD average in superficial plexus

○

Perf_Circle3mm_Sup: PD in a circle of 3-mm diameter in superficial plexus

○

Perf_Circle6mm_Sup: PD in a circle of 6-mm diameter in superficial plexus

○

Perf_Ave_Deep: PD average in deep plexus

○

Perf_Circle3mm_Deep: PD in a circle of 3-mm diameter in deep plexus

○

Perf_Circle6mm_Deep: PD in a circle of 6-mm diameter in deep plexus

3.2. Swept-Source OCT-A Acquisitions

OCT-A is a noninvasive imaging technique that provides three-dimensional visualization of the perfused vasculature of the retina and choroid. In contrast to standard structural OCT, OCT-A analyzes not only the intensity of the reflected light but also the temporal changes in the reflection caused by moving particles, such as erythrocytes flowing through vessels. These changes in the OCT signal are detected by repeatedly capturing OCT images at each point on the retina and allowing for the creation of image contrast between perfused vessels and static surrounding tissues [29]. To acquire such data, various algorithms have been established by several manufacturers, making resultant images different in appearance from one another. Such variances in the output of each device may result in different interpretations of the clinical diagnosis. More specifically, the success of an algorithm may be dependent on the number of repeated OCT scans at each retinal location, and on the sensitivity of the algorithm to differentiate particles in motion from static tissue. In addition to these considerations, each device may also differ with regard to acquisition speed and the retinal boundaries that are applied to differentiate various vascular plexi (using en face images generated from slabs). Moreover, while each unique OCT-A algorithm is subject to slightly different limitations that are attributed to its overall approach, there are certain confounding factors and/or limitations that impact all algorithms and are innate characteristics of this imaging modality.

The acquisition of OCT-A volume scans provides a three-dimensional cube of data that includes structural OCT and OCT-A images. A series of OCT section images (or B-scans) are acquired in order to create this cube of data. An initial review of these data is usually based on images that are generated from slabs of the cube. Slabs are sections of three-dimensional volumetric data. In the case of OCT-A slabs, the section is delimited by anterior and posterior retinal and choroidal boundaries. The OCT-A signal between these boundaries is displayed as a two-dimensional en face image, showing perfused vasculature. It is referred to as an en face image due to the transversal slab orientation; the resulting image gives the impression of looking onto the retina.

With the PLEX Elite 9000® instrument from Zeiss (Carl Zeiss Meditec Inc., Dublin, OH, USA), a 6 × 6-mm (~21° × 21°) scan pattern provides a relatively large overview of the retinal and choroidal circulation, ideal for detection of vascular abnormalities that may not be present at the avascular central macula. This high-speed scan has an isotropic lateral resolution of 11.7 μm/pixel (512 A-scans × 512 B-scans), and it can offer the resolution needed to visualize small capillaries. Considering the small diameter of these smallest capillaries (approximately 12 μm), a lower-resolution scan may limit the confidence or relibaility in image interpretation. The specifications of PLEX Elite 9000® instrument are resumed in Table 2. Finally, this high-resolution scan facilitates a more detailed and confident evaluation of vascular abnormalities at the capillary level.

Table 2. OCT Device specifications.

Model	Constructor			Technology	Hardware
PLEX Elite 9000®	Carl Zeiss Meditec Inc, Dublin, USA			Swept Source Optical Coherence Tomography	Optical Micro AngioGraphy (OMAG)
FOV	Wave length	Slew rate	Axial scan depth	Optical axial resolution	Optical transversal resolution	Number of images in dataset
56°	1040-1060 nm	100 000 A-scans/sec	3.0 mm	6.3 µm	20 µm	2005 en face images 814 angiocubes

Table 2. OCT Device specifications.

Model	Constructor			Technology	Hardware
PLEX Elite 9000®	Carl Zeiss Meditec Inc, Dublin, USA			Swept Source Optical Coherence Tomography	Optical Micro AngioGraphy (OMAG)
FOV	Wave length	Slew rate	Axial scan depth	Optical axial resolution	Optical transversal resolution	Number of images in dataset
56°	1040-1060 nm	100 000 A-scans/sec	3.0 mm	6.3 µm	20 µm	2005 en face images 814 angiocubes

3.3. Quantitative OCT-A Vascular Features

All the angiocubes were segmented and analyzed on a cloud platform called the Advanced Research and Innovation Network (ARI Network). Quantification analysis was performed using the « Macular Vasculature Density v0.7.3.3 » algorithm. This algorithm quantifies the vascular density (vessel and perfusion) of superficial and deep retina layers; it also quantifies the foveal avascular zone (FAZ) of the superficial layer. The outputs offered are:

-

Superficial and deep slabs (angio and structure)

-

Vessel and perfusion traces for superficial and deep slabs

-

Superficial and deep vessel and perfusion density maps, color overlay images

-

FAZ superficial segmentation

-

Density and FAZ quantification results

3.4. Cardiovascular Data

Historical models for the prediction of CVD in the general population, such as FRS, PCE, and the recently updated AtheroSclerotic CVD (ASCVD) Risk Estimator Plus, may have some limitations when used for patients with an intermediate risk profile. The latest guidelines from the American College of Cardiology and American Heart Association (AHA) recommend the use of the ASCVD Risk Estimator Plus, which provides a 10-year CVD risk score based on certain risk factors (age, sex, ethnicity), bedside tests (e.g., blood pressure), and blood parameters (e.g., total cholesterol) [30]. However, even such risk stratification algorithms can have limited calibration and discriminative ability when externally validated [31,32]. Moreover, generating these scores requires invasive biological sampling and depends on significant input from healthcare professionals and laboratory testing.

The universally known CHA₂DS₂-VASc clinical score, which is simple and quick to calculate, is a risk stratification tool initially used to estimate the risk of stroke in people with non-rheumatic atrial fibrillation [12]. It is a risk factor-based approach, by defining definitive risk factors (previous stroke/transient ischemic attack [TIA]/thromboembolism [TE] and age ≥ 75 years) and combination risk factors (heart failure/moderate–severe cardiac dysfunction, hypertension, diabetes, vascular disease, female gender, and age 65–74 years), as shown in Table 3. As we wished to artificially categorize the neurocardiovascular risk of these individuals, high risk was defined as the presence of one definitive or two or more combination risk factors, intermediate risk was essentially defined as the presence of one combination risk factor, and low risk was defined as the presence of one or no risk factor (Table 4). Guidelines from the AHA and the European Society of Cardiology recommend the use of this stratification system for the indication of oral anticoagulant therapy. However, because all components of the CHA₂DS₂-VASc score are important cardiovascular risk factors, a recent cohort study demonstrated that an incrementally higher CHA₂DS₂-VASc score is associated with stroke in patients regardless of the presence of atrial fibrillation and can help identify patients at higher risk of mortality [13,14,15,16,17,18,19,33,34,35,36,37]. To date, there is no consensus regarding the use of the CHA₂DS₂-VASc score for global cardiovascular risk stratification, but it appears that a high score of >3 would be synonymous with high cardiovascular risk.

Table 4. Risk Scheme used for neurocardiovascular risk stratification

Risk scheme	Low risk [0 ;1]		Intermediate risk [2 ;3]		High risk [4 ;9]
RASTA (2023)	One or no combination risk factor		One definitive risk factor and 1 or no combination risk factor, or 2 or 3 combination risk factors		Two definitive risk factors, or 1 definitive risk factor and ≥ 2 combination risk factors, or ≥ 4 combination risk factors
Definitive risk factors: previous stroke/TIA/TE, age > 75
Combination risk factors: heart failure/left ventricular ejection fraction ≤ 40%, hypertension, diabetes, vascular disease, female gender, age 65-74

Table 4. Risk Scheme used for neurocardiovascular risk stratification

Risk scheme	Low risk [0 ;1]		Intermediate risk [2 ;3]		High risk [4 ;9]
RASTA (2023)	One or no combination risk factor		One definitive risk factor and 1 or no combination risk factor, or 2 or 3 combination risk factors		Two definitive risk factors, or 1 definitive risk factor and ≥ 2 combination risk factors, or ≥ 4 combination risk factors
Definitive risk factors: previous stroke/TIA/TE, age > 75
Combination risk factors: heart failure/left ventricular ejection fraction ≤ 40%, hypertension, diabetes, vascular disease, female gender, age 65-74

Table 3. CHA₂DS₂-VASc point-based scoring system.

Risk Factor	Score
Congestive heart failure / Left ventricular dysfunction	1
Hypertension	1
Age ≥ 75 years	2
Diabetes mellitus	1
Stroke / TIA / TE	2
Vascular disease (prior myocardial infarction, peripheral artery disease, or aortic plaque)	1
Age 65-74 years	1
Sex category (i.e., female gender)	1

Table 3. CHA₂DS₂-VASc point-based scoring system.

Risk Factor	Score
Congestive heart failure / Left ventricular dysfunction	1
Hypertension	1
Age ≥ 75 years	2
Diabetes mellitus	1
Stroke / TIA / TE	2
Vascular disease (prior myocardial infarction, peripheral artery disease, or aortic plaque)	1
Age 65-74 years	1
Sex category (i.e., female gender)	1

4. Methods

Clinic and demographic data were collected at the inclusion of each participant in the study using a single medical interview common to each of the studies mentioned above. All information was verified by an investigating operator from the patients’ hospital medical chart, if available.

Each participant underwent an SS OCT-A examination of one or both eyes using the PLEX Elite 9000®. Examinations were acquired by three different trained operators and were performed under standard dark conditions. Pupillary dilatation was systematically performed with one eye drop of tropicamide 0.5% in both eyes if the B-scan signal strength was lower than 8/10. A 6 × 6-mm PLEX Elite 9000® angiography examination was performed for each of the included eyes. Only acquisitions with a signal strength greater than or equal to 8/10 were processed. Each angiocube and en face image were reviewed by an ophthalmologist without knowledge of the participant’s cardiovascular status. For volumetric acquisitions, if an acquisition was judged to be of poor quality or with too much noise after review, the participant was excluded from the database. En face images judged to be of insufficient quality by the ophthalmologist were deleted from the database.

5. Conclusion

Emerging modern imaging techniques such as SS OCT-A have created an unprecedented opportunity to comprehensively characterize the microscopic ophthalmic features associated with CVD, also known as the oculomics [38,39]. This oculomics revolution has opened up new avenues, including the use of the retina to obtain insights beyond the eye. Detecting microvascular changes before clinical manifestations can have predictive value, and ophthalmoscopic changes in the retinal microvasculature structure with SS OCT-A might represent a unique opportunity to fulfill this task.

Here, we introduce the first existing dataset of SS OCT-A images combined with cardiovascular data. Our dataset called RASTA contains volumetric acquisitions from 499 patients and 2005 segmented en face images with corresponding quantitative microvascular features and clinical cardiovascular data. The main interest of the RASTA dataset lies in the hybrid nature of the data that can strengthen collaborative research between ophthalmology and cardiology and refine the correlation between the vascular retinal network and cardiovascular diseases. Open access to medical imaging datasets remains a huge challenge for the community, which hinders the development of deep learning-based solutions as they require large datasets to reach efficient performances. Against this background, the RASTA dataset is the first contribution of publicly available SS OCT-A images with associated cardiovascular data.