Measuring arteriolar-to-venous ratio in retinal photography of patients with hypertension: Development and application of a new semi-automated method (2024)

Search for other works by this author on:

Article history

Abstract

Retinal vessels offer a unique opportunity to investigate noninvasively the relationship between arteriolar characteristics and cardiovascular disease. The association between retinal vessel abnormalities and high blood pressure (BP) has been described since the pioneering work of Keith, Wegener, and Baker¹ and has been demonstrated in larger contemporaneous cohorts^2,3 Moreover, arteriolar narrowing predicts the incidence of diabetes,⁴ coronary artery disease,⁵ and stroke.⁶ In two recent reports, narrowed retinal arterioles were shown to be independently associated with long-term risk of hypertension,^7,8 strengthening the idea that arteriolar constriction and narrowing may play a critical role in the earliest stages of the disease.

Because recognition of retinal abnormalities by direct ophthalmoscopy is subject to a large interobserver variation,^9–12 several investigators have developed quantitative methods to detect retinal microvascular changes documented by retinal photographs.^13–18 Reliable methods are available,^18,,20 but most of them seem to require a well-trained observer to recognize the border of the vessels.

In this report, we describe a new semi-automated method to evaluate the diameter of retinal vessels that does not depend on the identification of the vascular border, and we present an example of its use in the evaluation of patients with hypertension.

Methods

Our data derive from a cross-sectional analysis in a cohort study of patients attending to the outpatient hypertension clinic of the Hospital de Clínicas de Porto Alegre, Brazil. Details of the systematic evaluation of these patients and results of this investigation have been reported elsewhere.^21–23 For this investigation, all patients consecutively examined between 1993 and 1999 underwent retinal photography along with the complete evaluation of our protocol. The mean of six BP measurements, taken during three different visits, was used to diagnose hypertension. Patients were considered to having hypertension if their systolic BP (SBP) was ≥140 mm Hg or their diastolic BP (DBP) was ≥90 mm Hg, or if they were taking BP-lowering medications. Patients presenting with secondary hypertension, diabetes mellitus, or other systemic or ocular disease involving the retina were excluded.

In the total, 58 patients that met the eligibility criteria and had retinography suitable for analysis were evaluated. The photographs were analyzed by wall projection-micrometric method, in a previous study of the association between optic fundi abnormalities and stages of hypertension.²⁴ For the development of this new method, retinal photographs were digitized to determine the vessel diameter by the microdensitometric method.

Retinal photography

After pharmacologic pupil dilation (20 min after instillation of tropicamine), retinography was obtained using Topcon TVR-50 retinal camera (Topcon, Japan) in a 35-degree angle focusing in the posterior polo and the four quadrants (superior-nasal, inferior-nasal, superior-temporal, and inferior-temporal).

The color slides were digitized in a 35-mm film scanner Hewlett Packard model Photo Smart 20S (Hewlett Packard, Andover, MA) with a resolution of 600 dots per inch. Images were stored in 24 bits (true color). To enhance contrast of the retinal vessels against the retinal pigment epithelium, the green channel of the bitmap was selected.

Microdensitometric method

Image processor measurement

A custom-designed software program was built to perform vessel width measurements. The user defines the optic disc setting the center and average external boundary. Two concentric zones were delimited: inner zone (A), ranging from one half of the disc diameter to one full disc diameter; and outer zone (B), between one and two disc diameters from the margin.

Vessel measurements were done in a single vessel mode (arteriole or venule) and arteriolar-to-venous (A/V) ratio mode. The single-vessel mode was used to measure non-parallel pairs of vessels, whereas the A/V ratio mode was used for parallel pairs of vessels. In the single-vessel mode, the operator identified the type of vessel (arteriole or venule) to be automatically measured. In the A/V ratio mode, the operator determined two parallel adjacent vessels, and the program considered the wider vessel a vein and the thinner vessel an artery. We measured the most visible pair of vessels in each quadrant present in the photograph, and the program averaged these measurements.

For each measurement, a working area was constructed around the vessel of interest after setting an axis along its direction. Width measurements were performed in a plane perpendicular to this axis. An edge detection procedure was applied to the selected image. A double convolution was done using an approximation of the Sobel operator.^25,26 Sub-pixel resolution was obtained via cubic spline fitting of the edge of vessel walls. For each pair of adjacent pixels, 10 intermediate points were generated in the perpendicular direction of the wall. Vessel widths were automatically extracted from this synthetic curve, with cut-points defined where the exterior wall position equals the double of mean noise along the slice.

To test reproducibility of the microdensitometric method, 20% of the retinal photographs were randomly select and re-analyzed by the same observer (intraobserver reliability) and by a different observer (interobserver reliability). To access validity of the semi-automated method, the retinographies were also analyzed using standard grading.

Standard retinography analysis by micrometric method

The retinography was projected into a white screen 2 meters from the projector. All measurements were made in millimeters using a ruler. Arteriolar narrowing was estimated by A/V ratio. The vessels widths were examined at the same zones delimited for the microdensitometric method (inner and outer zones).

Arteriolar narrowing estimation

Arteriolar narrowing was considered present if the A/V ratio, as determined by both microdensitometric and micrometric methods, was ≥0.67 (arteriolar width equal to or less than two-thirds of the width of the venules). This cut-off point was chosen because it corresponds to the 75th percentile of the A/V distribution, as measured by the microdensitometric method.

Statistical analysis

Intra- and interobserver reliability of A/V ratio measurements were tested by intraclass correlation coefficient. The agreement between the two methods of determining A/V ratio measurements was tested using the intraclass correlation coefficient, and the agreement in determining the presence or absence of arteriolar narrowing was tested using the κ statistic. Sensitivity, specificity, and predicted values for the microdensitometric method were calculated using the results obtained by the micrometric method as the reference test.

The association between systolic, diastolic, and mean arterial BP and A/V ratio was tested using multiple regression analysis. The Statistical Program for Social Sciences, version 11.0 (SPSS Inc., Chicago, IL) and Epidat, version 2.1, PAHO-WHO software were used for analysis.

The institutional review board approved this study, and all patients gave their written consent to participate.

Results

As shown in Table 1, the participants included in the analysis were predominantly middle-aged, overweight, and female. More than 50% were smokers.

Figure 1 presents an example and the description of the output of our software for determining the arterial and venous widths. As can shown in Fig. 1, the participation of the observer is restricted to the selection of the point for measurement of vessel width. The software calculates the width of arterioles and veins, and the A/V ratios both for the inner and outer zone. If the pair of vessels was not parallel, the single-vessel mode was used: that is, the width of each vessel was measured separately and A/V ratio was then calculated. The software permits the examiner to see clearly the thickness of the vessels. The present analysis was restricted to the measurement of the external diameter of arterioles and venules.

Semi-automated measuring module of custom-made software. (Superior right window [1]:) The observer determines the center and margin of the optic nerve head. The software automatically delimits two concentric zones: the inner zone (A), ranging from one half of the disc diameter to one full disc diameter, and the outer zone (B), between one and two disc diameters from the margin. (Superior left window [2]:) The observer chooses the vessel segment to be measured and draws a line parallel to it. The software automatically rotates the image to make all vessel diameter measurements perpendicular to the vessel axis. (Inferior left window [3]:) (Right:) The software runs an edge detection routine and the observer chooses the point at which the vessel diameter (red line) should be determined. (Left:) The program automatically plots a line corresponding to the pixel intensity of the segment chosen by the observer. (Top:) Diameter of the first vessel (Va = 160), diameter of the second vessel (Vb = 129), and respective ratios (Va/Vb = 1.24 and Vb/Va = 0.74).

The intraclass correlation coefficient between micrometric and microdensitometric methods for A/V ratio measurement was 0.63 (P < .001) for the inner zone and 0.72 for the outer zone (P < .001). The global agreement was 59% for the inner zone and 83% for the outer zone. As shown in Table 2, the agreement between the methods to diagnose arteriolar narrowing reached a higher κ coefficient for the outer zone.

Sensitivity, specificity, and predictive values of the microdensitometric method to detect arteriolar narrowing are shown in Table 3. Noticeably, sensitivity, specificity, and predictive values were higher in the outer than in the inner zone.

We found that SBP and mean arterial BP were inversely and significantly associated with A/V ratio at the outer zone. For every increment of 10 mm Hg in the SBP, there was a decrease of 0.03 units of A/V ratio. The association between diastolic DBP and A/V ratio showed the same trend; however, this did not reach statistical significance (Table 4).

Discussion

In this article, we present a new semi-automated method to determine the A/V retinal vessels ratio that is mostly independent of the observer. Direct ophthalmoscopy is subject to large intra- and interobserver variation.^9–11 Poor agreement between an ophthalmologist and a cardiologist to detect arteriolar narrowing using direct ophthalmoscopy has also been identified.¹²

In the micrometric methods currently available, the observer must locate the retinal vessel edge from an enlarged or projected retinal image and must determine the vessel diameter with the aid of some measuring device. These procedures are susceptible to measurement bias.^27,28 In our microdensitometric method, the observer needs only to identify the areas in the inner and outer zone where the software takes the measurements and calculates the average A/V ratio. The κ coefficients for intraobserver agreements were close to 1 for the inner and outer zones, demonstrating the precision of this method. The interobserver coefficient was also satisfactory. They compared favorably with a method tested in a population-based study²⁹ and with the method used in the Atherosclerosis Risk in Communities (ARIC) study.¹⁹

The sensitivity for arteriolar narrowing detection of our method was satisfactory for both the inner and outer zones, but it had low specificity, especially in the inner zone. For the analysis of sensitivity and specificity, however, the reference test was the micrometric method. The low intraobserver reliability of this method suggests that it may not be the best standard to measure A/V ratio, and that computer-assisted method will be the gold standard in the future.

Outer-zone measurements were more reliable and accurate than inner-zone measurements. Arteriolar narrowing usually appears first in the pre-arteriole capillaries; it is more prominent in second- and third-order arterioles and is less frequent near the optic disc.³⁰ For this reason, first-order arterioles should be measured with some distance from the disc. Most of the studies in the literature describe measurements made from one and one-half-disc diameters from the margin of the papilla. Furthermore, color photography depicts the blood column. Therefore, such measurements give only estimates of the real vessel diameter. Our method measures the width of the vessel and may permit measurement of the vessel thickness. The implication of these different measurements should be investigated in studies with larger sample sizes.

Recent studies have shown a strong association between decreased A/V ratio and the incidence of diabetes,⁴ coronary artery disease,⁵ stroke,⁶ and hypertension.^7,8 In the ARIC study, generalized arteriolar narrowing was associated with hypertension, related not only to current but also to past BP levels, even after adjustment for current BP.³ Previous studies suggested that absolute measures of retinal arteriolar and venular diameters predict different outcomes; thus, the A/V ratio may not capture the separate information of arterioles and venules. Most recent studies on the risks of generalized arteriolar narrowing, however, have used the AV ratio, because the venular diameters vary little.¹⁴

We explored the performance of our method of measuring the AV ratio in a relatively small sample of patients with hypertension. Our findings are in accordance with those described in the ARIC study and other larger cohorts. We found an inverse and statistically significant association between SBP and A/V ratio, independently of age. An increment of 10 mm Hg in SBP corresponded to a decrease in 0.02 units of A/V ratio. Beside this observation, the evaluation of retinal vessels diameters may give rise to a new field of research, looking at the association between abnormalities in the retinal microcirculation in the pre-hypertensive stage with consideration to future development of hypertension and also results of pharmacologic interventions.

In summary, we described a new, highly reliable, semi-automated technique to analyze A/V ratio on digitized retinography that is mostly independent of the observer. Its simplicity in the detection of retinal microvascular abnormalities might have clinical and epidemiologic utility in identifying individuals at higher risk of end-organ damage and cardiovascular events, among patients with and without hypertension.

References

© American Journal of Hypertension, Ltd. 2005

Issue Section:

ORIGINAL CONTRIBUTIONS

Email alerts

Citing articles via

More from Oxford Academic