322

INTRODUCTION

The retinal microvasculature offers a unique oppor-tunity to study the human circulatory system in-vivo and non-invasively.1,2 Several parameters of the retinal vascular geometry, including vascular caliber, tortu-osity, and branching angle, have been proposed to

reflect the state of the peripheral circulation in previ-ous experiments.3–5 Therefore, it has been suggested that these parameters may be altered early in the course of vascular disease.6–11

We have recently developed new imaging software, Singapore ‘I’ Vessel Analysis (SIVA), which measures a range of retinal vascular geometric parameters from

Ophthalmic Epidemiology, 19(5), 322–327, 2012© 2012 Informa Healthcare USA, Inc.ISSN: 0928-6586 print/1744-5086 onlineDOI: 10.3109/09286586.2012.702258

*Joint first author. Both authors contributed equally to this studyReceived 24 September 2011; revised 06 January 2012; accepted 01 March 2012

Correspondence: Jie Jin Wang, MMed (ClinEpi), MAppStat, PhD, Professor, NHMRC Senior Research Fellow, Centre for Eye Research Australia, University of Melbourne, 32 Gisborne Street, Victoria 3002, Australia. Tel: +61 2 9845 5006. Fax: +61 2 9845 8345. E-mail: jiejin.wang@sydney.edu.au

24September2011

06January2012

01March2012

© 2012 Informa Healthcare USA, Inc.

2012

Ophthalmic Epidemiology

1744-5086

10.3109/09286586.2012.702258

19

322

327

5

ORIGINAL ARTICLE

Correlation and Reproducibility of Retinal Vascular Geometric Measurements for Stereoscopic Retinal

Images of the Same EyesMuhammad Bayu Sasongko1,2,*, Lauren A. B. Hodgson1,*, Tien Y. Wong1,3,4,5,

Ryo Kawasaki1, Carol Y. Cheung3,4,5, Wynne Hsu6, Mong Li Lee6, Peter Q. F. Lau6, Paul Mitchell7, and Jie Jin Wang1,7

1Centre for Eye Research Australia, Royal Victorian Eye and Ear Hospital, University of Melbourne, Australia, 2Department of Ophthalmology, Faculty of Medicine, Gadjah Mada University, Yogyakarta, Indonesia, 3Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, 4Department of Ophthalmology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, 5Centre for Quantitative Medicine, Office of Clinical Sciences, Duke-NUS Graduate Medical School, Singapore, 6School of Computing, National University of Singapore, Singapore,

and 7Centre for Vision Research, Westmead Millennium Institute, University of Sydney, Sydney, Australia

ABSTRACT

Purpose: To assess the correlation and reproducibility of retinal vascular geometric measurements obtained from two stereo-paired fundus images.

Methods: Thirty stereoscopic pairs of color optic disc-centered photographs from the Blue Mountains Eye Study were analyzed. Side-by-side grading was performed by a single grader, using semi-automated computer software to quantify the following retinal geometric parameters: (1) retinal arteriolar/venular caliber (CRAE/CRVE); (2) arteriole-to-venule ratio (AVR); (3) branching angle; and (4) tortuosity. We used Pearson correla-tion (r), intra-class correlation coefficient (ICC), and Bland-Altman plots to assess within-pair correlation and reproducibility for each parameter measured.

Results: Inter- and intra-grader r and ICC were high (all r > 0.90 and ICC > 0.90), except for branching angle (ICCs between 0.69–0.83). There was no significant difference between within-pair means of all retinal vascular geometric parameters, before and after excluding poor quality images. CRAE, CRVE, AVR, and arteriolar and venular tortuosity showed very high within-pair correlation and agreement (all r > 0.80 and all ICC > 0.90 respectively). Arteriolar and venular branching angles demonstrated moderate within-pair correlation (r = 0.65 and r = 0.62, respectively) and within-pair agreement (ICC = 0.76 and ICC = 0.77, respectively).

Conclusions: Use of computer-assisted software to measure retinal vascular geometric parameters from paired fundus images was highly repeatable and is robust to differences in photographic angles of paired stereo images. Such measurements can be applied to evaluate temporal changes in longitudinal studies.

KEYWORDS: Correlation, Reproducibility, Retina, Retinal imaging, Retinal vascular geometry, Stereoscopic

0928-6586

Reproducibility of Stereo Retinal Vessel Geometry

M. B. Sasongko et al.

Oph

thal

mic

Epi

dem

iol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

Reproducibility of Stereo Retinal Vessel Geometry 323

© 2012 Informa Healthcare USA, Inc.

digital fundus photographs. We have shown that subtle changes in caliber, branching angle and tortuosity, measured using SIVA, are associated with diabetic risk factors,7 diabetic microvascular complications,12–14 and cardiovascular disease risk factors in generally healthy populations before the manifestation of clinical disease.6,15 These findings support the potential clinical utility of SIVA to identify early, sub-clinical changes of small vessel structure in some systemic diseases.

However, whether these retinal vascular parameters change over time is not known. Understanding tem-poral change in longitudinal studies is important for retinal vascular imaging to be used in monitoring dis-ease development and progression and in testing thera-peutic interventions. Considering that retinal vascular geometric parameters rely on their three-dimensional nature, potential error may occur in the measurement of these parameters from multiple two-dimensional plane fundus images without consistent photographic angle. Availability of retinal imaging software that is robust to variation in photographic angle is important for its application in longitudinal studies and clinical trials. In this study, we tested the impact from different photo-graphic angles in measurement of geometric parameters of the retinal vasculature, using the SIVA program.

MATERIALS AND METHODS

Measurement of Retinal Vascular Geometry using SIVA

Thirty pairs of stereoscopic retinal photographs of right eyes taken of Blue Mountains Eye Study (BMES) baseline participants (1992–1994) were selected. Ethics approval was granted from the Western Sydney Area Health Service Human Research Ethics Committee and written consent was obtained from all study partici-pants. To obtain a stereo effect, image pairs were taken while the camera was manually shifted laterally, with the first image of the pair taken when the camera shifted temporally, and the second image of the pair when the camera was shifted nasally. The distance between the two positions of the camera was a few centimeters, and the time interval between the two images was less than a minute.16 We digitized these color positive film pho-tographs using a Canon CanoScan FS2710 slide scanner (Canon Corporation, Tokyo, Japan) to produce. TIFF images at 3888 × 2595 pixels (size 9.6 MB).

For measuring retinal microvascular geometric properties, fundus photographs centered on the optic disc were viewed on two 20-inch monitors with a resolution of 1280 × 1024. A trained and experienced grader (LH) opened each pair of images and per-formed simultaneous grading of both paired images using SIVA software version 3.0. This grading method was done to ensure vessel selections between paired images were identical, without any knowledge of the

measurement results. Grading procedures have been detailed previously.7,13,14 The program measured retinal microvascular geometric parameters within a concen-tric zone between the optic disc margin and two optic disc diameters away from the optic disc margin. The grader allowed the software to automatically detect the centre of the optic disc and divide the region into three concentric subzones (A, B, C) surrounding the optic disc, each zone corresponding to 0.5, 1.0 and 2.0 optic disc diameters away from the optic disc margin. Once the optic disc and the three concentric subzones were considered appropriately located, the grader executed the program to generate a line tracing of the retinal vessels from each image. This software also has an automated function to appropriately identify arteri-oles and venules, indicated by two different colors of the lines generated by the program; red for arterioles and blue for venules. The grader subsequently checked each image graded that all arterioles and venules were correctly identified, based on information of parent vessels, crossing between arterioles and venules and the color of the vessels. In addition, the grader also examined the line tracing along each vessel to avoid erroneous selection of artifacts or shadows that are commonly present in poor quality or blurry images. Correction could be made if necessary to ensure the vessel selections and line tracings were appropriate and exactly identical between paired images (Figure 1).

Images were considered of poor quality if they were blurred, there was incomplete representation of zone C or there were less than six gradable vessels, and ungradable if there were less than four large arteri-oles or venules gradable. Measurements were based on summaries of arterioles and venules separately. The software combined the individual measurement into summary indices for caliber, tortuosity, and branch-ing angles separately. The technical details of these parameters are given in previous publications.3,17–19 Briefly, vessel caliber is represented by central retinal arteriolar equivalent (CRAE) and central retinal venu-lar equivalent (CRVE).17,18 Vessel tortuosity, a reflection of the shape of the vessel, is expressed as a tortuosity index.3 We used two measures of vessel tortuosity, simple and curvature. Simple tortuosity was a ratio between the actual length of a vessel segment and the shortest distance within the same segment. Curvature tortuosity was calculated from the integral of the total squared curvature along the path of the vessel, divided by the total arc length.3 Branching angle (in degrees) represents the angle between two daughter vessels.19

To assess intra-grader reliability, the primary grader (LH) re-graded one image from each of the 30 stereo image pairs following exactly the same protocol as explained above, masked to previous grading results. An additional grader (MBS) also graded the same images following the same protocol, masked to previ-ous grading results and those graded by the primary grader.

Oph

thal

mic

Epi

dem

iol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

324 M. B. Sasongko et al.

Ophthalmic Epidemiology

Statistical Analysis

We performed statistical analyses using Intercooled STATA 10.1 for Windows (Stata Corp., TX). All parame-ters were analyzed continuously. Paired 2-sample t-tests were used to compare mean values within paired ste-reoscopic images and to obtain intra- and inter-grader mean difference for each parameter. We also calculated Pearson correlation (r) and intra-class correlation (ICC) coefficients to assess within-pair, intra-, and inter-grader correlation and agreement for each parameter measured. Bland-Altman plots were used to depict the limits of agreement by showing mean difference with 95% confidence intervals (CI) for within-paired stereo-scopic images.

RESULTS

Table 1 shows that the mean value of each parameter within paired images was very similar and also shows that none of the within-pair mean differences for each parameter was significant. These results remained identical before and after excluding poor quality images.

Table 2 shows within-stereoscopic paired image r and ICC of each parameter. CRAE, CRVE, arteriole-to-venule ratio (AVR), and arteriolar and venular tortuos-ity showed high within-pair correlation and agreement, with all r > 0.80 and all ICC > 0.90. However, arteriolar and venular branching angle measurements demon-strated moderate within-pair correlation (r = 0.65 and r = 0.62, respectively) and within-pair agreement (ICC = 0.76 and ICC = 0.77, respectively). Excluding poor quality images from the analysis slightly improved the correlation and agreement in measurement of arteriolar and venular branching angles, and did not make any difference in measures of other parameters.

Figure 2 (Bland-Altman plots) shows that all parameters had good-to-excellent within-pair agree-ment; overall there were less than three data points among 30 pairs of images (<10%) outside the limit of agreement.

Table 3 shows that intra- and inter-grader mean differences were not significant for all parameters. Table 4 shows that inter- and intra-grader r and ICC of each parameter were very high (all r > 0.90 and ICC > 0.90), except for branching angle (ICC 0.78 for both intra- and inter-grader arteriolar branching angle; 0.83 and 0.69 for intra- and inter-grader venular branching

FIGURE 1 Graded stereo image pair showing the zones of measurement and vessels that were not included in grading due to lack of visibility in the second image (indicated by asterisks*).

TABLE 1 Mean value of each parameter within paired images and within-pair mean difference of retinal vascular parameters.

VariablesMean right image (SD)

Mean left image (SD)

Mean difference (95% CI) P value

Mean difference (95% CI)* P value*

Arterioles CRAE 146 (10.7) 147 (10.3) −1.34 (−6.76–4.09) 0.62 −3.67 (−9.98–2.54) 0.25Branching angle 81.4 (8.65) 83.5 (11.9) −2.07 (−7.44–3.29) 0.44 −2.83 (−8.76–3.09) 0.34S tortuosity (×102) 1.10 (0.03) 1.10 (0.02) −0.02 (−1.31–1.25) 0.97 −0.33 (−1.51–0.86) 0.58C tortuosity (×105) 7.59 (1.78) 7.87 (1.80) −0.28 (−1.21–0.64) 0.55 −0.35 (−1.31–0.60) 0.46

Venules CRVE 221 (20.4) 222 (19.5) −1.01 (−11.3–9.32) 0.85 −2.81 (−13.4–7.78) 0.59Branching angle 84.3 (12.3) 79.7 (11.8) 4.60 (−1.62–10.8) 0.14 2.41 (−4.90–9.73) 0.51S tortuosity (×102) 1.10 (0.03) 1.10 (0.03) −0.01 (−1.36–1.34) 0.99 −0.82 (−2.18–0.54) 0.23C tortuosity (×105) 8.25 (1.29) 8.58 (1.31) −0.33 (−1.00–0.35) 0.32 −0.66 (−1.38–0.56) 0.07

Combined AVR (×102) 0.67 (0.05) 0.67 (0.06) −0.34 (−3.27–2.57) 0.81 −0.90 (−3.89–2.09) 0.55*Excluding poor quality images; P-value was estimated using paired t-test.SD, standard deviation; CI, confidence interval; CRAE, central retinal arteriolar equivalent; CRVE, central retinal venular equivalent; S tortuosity, simple tortuosity; C tortuosity, curvature tortuosity; AVR, arteriole-to-venule ratio.

Oph

thal

mic

Epi

dem

iol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

Reproducibility of Stereo Retinal Vessel Geometry 325

© 2012 Informa Healthcare USA, Inc.

angle, respectively). We also demonstrated that the magnitude of intra- and inter-grader differences for each parameter were smaller than the corresponding differences within paired stereoscopic images, shown by the slightly higher ICCs and r values for intra- and inter-grader reproducibility than those for paired ste-reoscopic images.

DISCUSSION

In this study, we demonstrated that the SIVA program had excellent within-pair correlation and agreement for vessel caliber and tortuosity, and moderate correlation and agreement for branching angle measurements. We also showed that excluding poor quality images slightly improved the reproducibility and within-pair correlation for branching angle measurements. These findings suggest that quantitative measurement of retinal vascular geometry using SIVA is robust and reproducible for different photographic angles, which is likely to be the case if multiple images are taken from the same subjects/eyes over a period of time, as in most longitudinal studies.

The retinal vasculature is a complex peripheral vas-cular system, which has unique vessel structural and physiological characteristics, differing from most other peripheral vasculature.2 For example, the retinal circula-tion has higher blood flow volumes and retinal arteries and small arterioles have scarce muscular components compared to other microcirculatory beds.20 Assuming the blood acts as a Newtonian fluid with constant viscosity and adopts Poiseuille’s law, Murray proposed that in an optimal state of circulation, there is an optimal vascular architecture that enables the most efficient blood flow transport with minimum energy spent.4 The geometry of vascular beds follow this principle of efficiency to maxi-mize blood flow and diffusion.21 The retinal vasculature may therefore have structural parameters that adapt to hemodynamic variations and are sensitive to subtle pathophysiological sequences in the early stages of vari-ous conditions.

TABLE 2 Pearson correlation and intra-class correlation (ICC) of retinal vascular parameters within paired stereoscopic images.

Variables

Pearson Correlation, r ICC (95% CI)

All (n = 30) Good (n = 17) All (n = 30) Good (n = 17)Arterioles CRAE 0.927 0.919 0.905 (0.800–0.955) 0.953 (0.892–0.980)

Branching angle 0.646 0.751 0.761 (0.499–0.886) 0.849 (0.650–0.935)Simple tortuosity 0.984 0.982 0.992 (0.983–0.996) 0.990 (0.976–0.996)

Venules Curvature tortuosity 0.954 0.963 0.977 (0.951–0.989) 0.966 (0.922–0.985)CRVE 0.942 0.972 0.970 (0.937–0.986) 0.979 (0.951–0.991)Branching angle 0.620 0.712 0.765 (0.506–0.888) 0.830 (0.608–0.927)Simple tortuosity 0.972 0.979 0.972 (0.970–0.993) 0.989 (0.975–0.995)

Combined Curvature tortuosity 0.837 0.846 0.911 (0.814–0.958) 0.915 (0.804–0.963)AVR 0.904 0.935 0.947 (0.890–0.975) 0.967 (0.923–0.986)

CI, confidence interval; CRAE, central retinal arteriolar equivalent; CRVE, central retinal venular equivalent; AVR, arteriole-to-venule ratio.

FIGURE 2 Bland-Altman plots for various within-pair retinal vascular parameters.

Oph

thal

mic

Epi

dem

iol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

326 M. B. Sasongko et al.

Ophthalmic Epidemiology

In line with this theory, we previously demonstrated that variations in retinal vascular geometry (eg, nar-rower/wider caliber, increased tortuosity, larger branch-ing angle), measured using the SIVA program were associated with a range of systemic cardiovascular or diabetes risk factors before the clinical manifestation of these conditions.6,7 Although the evidence is cross-sec-tional in nature, these findings suggest that assessment of retinal vascular geometry using the SIVA program might provide insights into early, pre-clinical pathophysiologi-cal processes involved in the development of systemic vascular conditions. Evidence from longitudinal studies, however, will be needed to confirm this hypothesis.

The technical difficulty in capturing retinal images from exactly the same photographic angle at different times poses a potential limitation of longitudinal stud-ies that attempt to assess changes over time in retinal vascular geometry using retinal imaging. Due to its three-dimensional structure, capturing structural infor-mation from multiple two-dimensional retinal images with different photographic angle might bias findings, which is likely to occur in longitudinal studies. In this study, we documented excellent reproducibility of caliber and tortuosity measurements on paired images from different photographic angles, reassuring the suit-ability of SIVA for use in longitudinal studies. A slightly lower correlation and agreement in branching angle

measurement was expected, due to its two-dimensional measurement of a three-dimensional structural charac-ter. Nevertheless, the correlation and reproducibility of branching angle measurement was moderate to high, which could be improved slightly using high quality images. Our findings support the overall suitability of the SIVA image program to be used in comparisons of images taken from different photographic angles. Furthermore, our findings indicated that the magni-tude of intra- and inter-grader variations is smaller than the variations due to photographic angles. These findings are particularly important when applied to longitudinal studies to ensure that changes in geomet-ric parameters are due to actual biological processes, not the result of photographic angle or intra- and inter-grader variations.

The strength of this study is its use of paired images taken from the same eye from a well-established epi-demiological study (BMES), with only a difference in photographic angle. A limitation of our study is that we performed side-by-side grading instead of using a masked manner, due to the fact that the exact same ves-sels and branches must be traced in both images of each pair, to ensure that these retinal vascular parameters were measured from the same vessels.

In conclusion, our study showed that quantitative measurement of retinal vascular geometric parameters

TABLE 3 Intra- and inter-grader mean difference of retinal vascular parameters.

VariablesIntra-grader mean difference Inter-grader mean difference

(95% CI) P value* (95% CI) P value*Arterioles CRAE 0.98 (−0.92–2.87) 0.30 0.67 (−1.12–2.45) 0.44

Branching angle 0.16 (−3.44–3.77) 0.93 0.18 (−3.41–3.78) 0.92S tortuosity (×102) 0.01 (−0.05–0.07) 0.80 0.06 (−0.11–0.24) 0.49C tortuosity (×105) −0.06 (−0.25–0.13) 0.52 −0.27 (−0.54–0.01) 0.16

Venules CRVE 0.21 (−1.41–1.83) 0.80 −1.02 (−3.70–1.65) 0.44Branching angle 0.21 (−3.94–4.37) 0.92 −0.81 (−3.99–2.35) 0.60S tortuosity (×102) 0.04 (−0.02–0.06) 0.64 0.10 (−0.08–0.27) 0.26C tortuosity (×105) −0.15 (−0.40–0.11) 0.25 −0.24 (−0.51–0.03) 0.08

Combined AVR (×102) 0.36 (−0.44–1.17) 0.36 0.58 (−0.48–1.65) 0.27*P-values were estimated using paired t-test.CI, confidence interval; CRAE, central retinal arteriolar equivalent; CRVE, central retinal venular equivalent; S tortuosity, simple tortu-osity; C tortuosity, curvature tortuosity; AVR, arteriole-to-venule ratio.

TABLE 4 Inter- and intra-grader Pearson correlation and intra-class correlation (ICC) of retinal vascular parameters.

VariablesPearson Correlation, r ICC (95% CI)

Inter-grader (n = 30) Intra-grader (n = 30) Inter-grader (n = 30) Intra-grader (n = 30)Arterioles CRAE 0.974 0.972 0.987 (0.972–0.994) 0.972 (0.942–0.987)

Branching angle 0.647 0.651 0.783 (0.545–0.897) 0.783 (0.544–0.894)Simple tortuosity 0.988 0.967 0.992 (0.982–0.996) 0.982 (0.963–0.992)Curvature tortuosity 0.961 0.949 0.980 (0.958–0.991) 0.974 (0.945–0.988)

Venules CRVE 0.940 0.929 0.991 (0.982–0.996) 0.974 (0.946–0.994)Branching angle 0.523 0.716 0.829 (0.641–0.919) 0.685 (0.337–0.850)Simple tortuosity 0.984 0.968 0.993 (0.986–0.997) 0.966 (0.930–0.984)Curvature tortuosity 0.885 0.886 0.937 (0.867–0.970) 0.933 (0.859–0.968)

Combined AVR 0.977 0.981 0.988 (0.975–0.994) 0.980 (0.978–0.995)CRAE, central retinal arteriolar equivalent; CRVE, central retinal venular equivalent; AVR, arteriolar-to-venular ratio.

Oph

thal

mic

Epi

dem

iol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

Reproducibility of Stereo Retinal Vessel Geometry 327

© 2012 Informa Healthcare USA, Inc.

from retinal images using SIVA was robust and repro-ducible for images of different photographic angles. This property allows quantitative assessment of changes over time that may occur in retinal vascular geometry in longitudinal studies.

ACKNOWLEDGMENTS

Financial Support: CERA receives Operational Infra-structure Support from the Victorian Government. This study was supported by the National Health and Medical Research Council Centre for Clinical Research Excellence #529923 – Translational Clinical Research in Major Eye Diseases.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the article.

REFERENCES

[1] Liew G, Wang JJ, Mitchell P, Wong TY. Retinal vascular imaging: a new tool in microvascular disease research. Circ Cardiovasc Imaging 2008;1(2):156–161.

[2] Sasongko MB, Wong TY, Wang JJ. Retinal microvascular structure: determinants and potential utility of novel imag-ing measurements. Expert Rev Ophthal 2010;5(3):353–363.

[3] Hart WE, Goldbaum M, Cote B, Kube P, Nelson MR. Measurement and classification of retinal vascular tortuos-ity. Int J Med Informatics 1999;53(2–3):239–252.

[4] Murray CD. The physiological principle of minimum work: I. The vascular system and the cost of blood volume. Proc Nat Acad Sci USA 1926;12(3):207–214.

[5] Zamir M. Optimality principles in arterial branching. J Theoret Biol 1976;62(1):227–251.

[6] Cheung CY, Zheng Y, Hsu W, et al. Retinal vascular tor-tuosity, blood pressure, and cardiovascular risk factors. Ophthalmology 2011;118(5):812–818.

[7] Sasongko MB, Wang JJ, Donaghue KC, et al. Alterations in retinal microvascular geometry in young type 1 diabetes. Diabetes Care 2010;33(6):1331–1336.

[8] Witt N, Wong TY, Hughes AD, et al. Abnormalities of retinal microvascular structure and risk of mortality from ischemic heart disease and stroke. Hypertension 2006;47(5): 975–981.

[9] Taarnhoj NC, Munch IC, Sander B, et al. Straight versus tortuous retinal arteries in relation to blood pressure and genetics. Br J Ophthalmol 2008;92(8):1055–1060.

[10] Tapp RJ, Williams C, Witt N, et al. Impact of size at birth on the microvasculature: the Avon Longitudinal Study of Parents and Children. Pediatrics 2007;120(5):e1225–1228.

[ 11] Sasongko MB, Wong TY, Wang JJ. Retinal arteriolar changes: intermediate pathways linking early life exposures to cardio-vascular disease? Microcirculation 2010;17(1):21–31.

[ 12] Benitez-Aguirre P, Craig ME, Sasongko MB, et al. Retinal vascular geometry predicts incident retinopathy in young people with type 1 diabetes: a prospective cohort study from adolescence. Diabetes Care 2011;34(7):1622–1627.

[ 13] Sasongko MB, Wong TY, Donaghue KC, et al. Retinal arterio-lar tortuosity is associated with retinopathy and early kid-ney dysfunction in type 1 diabetes mellitus. Am J Ophthalmol 2012;153(1):176–183.

[ 14] Sasongko MB, Wong TY, Nguyen TT, Cheung CY, Shaw JE, Wang JJ. Retinal vascular tortuosity in persons with diabetes and diabetic retinopathy. Diabetologia 2011;54(9): 2409–2416

[ 15] Cheung CY, Tay WT, Mitchell P, et al. Quantitative and quali-tative retinal microvascular characteristics and blood pres-sure. J Hypertension 2011;29(7):1380–1391.

[ 16] Cosatto VF, Liew G, Rochtchina E, et al. Retinal vascular fractal dimension measurement and its influence from imag-ing variation: results of two segmentation methods. Curr Eye Res 2010;35(9):850–856.

[ 17] Hubbard LD, Brothers RJ, King WN, et al. Methods for evaluation of retinal microvascular abnormalities associated with hypertension/sclerosis in the Atherosclerosis Risk in Communities Study. Ophthalmology 1999;106(12):2269–2280.

[ 18] Knudtson MD, Lee KE, Hubbard LD, Wong TY, Klein R, Klein BE. Revised formulas for summarizing retinal vessel diameters. Curr Eye Res 2003;27(3):143–149.

[ 19] Zamir M, Medeiros JA, Cunningham TK. Arterial bifurca-tions in the human retina. J Gen Physiol 1979;74(4):537–548.

[ 20] Cogan DG, Kuwabara T, Friedman E. Retinal vasculature. Microvasc Res 1968;1:115–132.

[ 21] Murray CD. The physiological principle of minimum work: II. Oxygen exchange in capillaries. Proc Nat Acad Sci USA 1926;12(5):299–304.

Oph

thal

mic

Epi

dem

iol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.