Volume 20, Issue 2 p. 941-953
RESEARCH ARTICLE
Open Access

Association of fractal dimension and other retinal vascular network parameters with cognitive performance and neuroimaging biomarkers: The Multi-Ethnic Study of Atherosclerosis (MESA)

Sally S. Ong

Corresponding Author

Sally S. Ong

Department of Ophthalmology, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Correspondence

Sally S. Ong, MD, Department of Ophthalmology, Wake Forest University School of Medicine, Medical Center Boulevard, Winston-Salem, 27157 NC, USA.

Email: [email protected]

Search for more papers by this author
Jeremy J. Peavey

Jeremy J. Peavey

Department of Internal Medicine, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Kevin D. Hiatt

Kevin D. Hiatt

Department of Radiology, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Christopher T. Whitlow

Christopher T. Whitlow

Department of Radiology, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Rebecca M. Sappington

Rebecca M. Sappington

Department of Ophthalmology, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Department of Biochemistry, Wake Forest School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Atalie C. Thompson

Atalie C. Thompson

Department of Ophthalmology, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Samuel N. Lockhart

Samuel N. Lockhart

Department of Internal Medicine, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Haiying Chen

Haiying Chen

Department of Psychiatry and Behavioral Medicine, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Suzanne Craft

Suzanne Craft

Department of Internal Medicine, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Stephen R. Rapp

Stephen R. Rapp

Biostatistics and Data Science, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
Annette L. Fitzpatrick

Annette L. Fitzpatrick

Department of Epidemiology, School of Public Health, University of Washington, Seattle, Washington, USA

Search for more papers by this author
Susan R. Heckbert

Susan R. Heckbert

Department of Epidemiology, School of Public Health, University of Washington, Seattle, Washington, USA

Search for more papers by this author
José A. Luchsinger

José A. Luchsinger

Departments of Medicine and Epidemiology, Columbia University Irving Medical Center, New York, New York, USA

Search for more papers by this author
Barbara E. K. Klein

Barbara E. K. Klein

Department of Ophthalmology and Visual Sciences, University of Wisconsin-Madison, Madison, Wisconsin, USA

Search for more papers by this author
Stacy M Meuer

Stacy M Meuer

Department of Ophthalmology and Visual Sciences, University of Wisconsin-Madison, Madison, Wisconsin, USA

Search for more papers by this author
Mary Frances Cotch

Mary Frances Cotch

National Eye Institute, National Institute of Health, Bethesda, Maryland, USA

Search for more papers by this author
Tien Y. Wong

Tien Y. Wong

Singapore Eye Research Institute, Singapore National Eye Center, Ophthalmology and Visual Sciences Academic Clinical Program, Duke-NUS Medical School, Singapore

Tsinghua Medicine, Tsinghua University, Beijing, China

Search for more papers by this author
Timothy M. Hughes

Timothy M. Hughes

Department of Internal Medicine, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Search for more papers by this author
First published: 12 October 2023
Citations: 1

Abstract

INTRODUCTION

Retinal vascular network changes may reflect the integrity of the cerebral microcirculation, and may be associated with cognitive impairment.

METHODS

Associations of retinal vascular measures with cognitive function and MRI biomarkers were examined amongst Multi-Ethnic Study of Atherosclerosis (MESA) participants in North Carolina who had gradable retinal photographs at Exams 2 (2002 to 2004, n = 313) and 5 (2010 to 2012, n = 306), and detailed cognitive testing and MRI at Exam 6 (2016 to 2018).

RESULTS

After adjustment for covariates and multiple comparisons, greater arteriolar fractal dimension (FD) at Exam 2 was associated with less isotropic free water of gray matter regions (β = −0.0005, SE = 0.0024, p = 0.01) at Exam 6, while greater arteriolar FD at Exam 5 was associated with greater gray matter cortical volume (in mm3, β = 5458, SE = 20.17, p = 0.04) at Exam 6.

CONCLUSION

Greater arteriolar FD, reflecting greater complexity of the branching pattern of the retinal arteries, is associated with MRI biomarkers indicative of less neuroinflammation and neurodegeneration.

1 BACKGROUND

Vascular contributions are increasingly recognized as important risk factors for the development of dementia, with β-amyloid and cerebrovascular disease constituting the most prevalent combination of pathologies leading to dementia.1 There is increasing recognition that, in addition to the traditional impact of large artery vascular disease (eg, stroke), cerebral small vessel disease (eg, microinfarcts, leukoaraiosis, microbleeds, and loss of white matter microstructure) also contributes to cognitive decline and dementia. While a variety of cerebral biomarkers of microvascular disease have been examined in relation to their contributions to cognitive decline, Alzheimer's disease (AD) and related dementias, these biomarkers have significant limitations, and the search for other microvascular biomarkers remains important.1

The retinal vasculature is the only visible microvascular network that can be easily imaged and measured repeatedly and non-invasively. Embryologically, the retina and optic nerve extend from the diencephalon, and vascular changes in the retina and the central nervous system may be similarly affected by neurodegenerative processes. In contrast to magnetic resonance imaging (MRI), retinal imaging techniques like digital fundus photography are inexpensive and can provide better detail about microvascular structure. Notably, a spectrum of retinal vascular changes have been described and shown to be associated with signs of vascular disease in the brain, specifically the presence of infarcts, white matter lesions, cerebral microbleeds, and atrophy.2 These retinal vascular changes have shown promising associations with dementia and cognitive impairment.2-4

Semi-automated computer software has now allowed measurement of different retinal vascular parameters,5 such as retinal vessel caliber, central retinal arteriolar equivalent (CRAE) and central retinal venular equivalent (CRVE), and global vessel network measurements like fractal dimension (FD), tortuosity, and branching angle. CRAE and CRVE are measurements that quantify generalized retinal vessel narrowing or widening, and reductions in these parameters have been found in dementia in some studies6, 7 but not others.8 Global vascular network parameters, such as FD, tortuosity and branching angle, reflect the optimization of the branching pattern of the retinal vascular tree5 and may also reflect the integrity of the cerebral microcirculation.9

The goal of this study is to examine cross-sectional and longitudinal associations of retinal vascular network parameters with cognitive performance and neuroimaging biomarkers of small vessel disease and neurodegeneration among Multi-Ethnic Study of Atherosclerosis (MESA) participants from the Wake Forest field center. We hypothesized that changes in vascular network geometric parameters would be associated with poor cognitive performance, cognitive decline, and evidence of cerebral small vessel disease and neurodegeneration.

2 METHODS

2.1 Study population

The MESA is a prospective longitudinal study that consists of a diverse population sample from six field centers (Forsyth County, NC; Baltimore, MD; Chicago, IL; Los Angeles, CA; New York, NY; St. Paul, MN). Individuals free of clinically-recognized cardiovascular disease (CVD; n = 6814) were recruited and have been followed since Exam 1 (July 2000 to July 2002) to identify risk factors for CVD and development of clinical events. The present analysis included only participants from the Wake Forest field center in Forsyth County, NC. MESA was approved by the Institutional Review Board of each participating field center and written informed consent was collected from all participants.

2.2 Measurements

At baseline and each follow-up clinical examination, data were collected using standardized protocols for the following: anthropometry (including height and weight for body mass index [BMI]); blood pressure; blood draw; and questionnaires to assess self-reported demographics (age, gender, race/ethnicity, level of education, smoking status) and medication usage for high blood pressure, high cholesterol, or diabetes. Blood sample collection included assays for fasting plasma glucose and serum creatinine. Impaired fasting glucose and diabetes mellitus (DM) was defined by 2003 American Diabetes Association criteria.10 Serum creatinine was used to estimate glomerular filtration rate (eGFR) according to the Modification of Diet in Renal Disease study equation.11

2.3 Retinal photography and grading

Fundus photographs were taken at Exams 2 (2002 to 2004) and 5 (2010 to 2012)12 using a standardized protocol that has been previously published.13 Both eyes from each participant were photographed with a 45-degree 6.3-megapixel digital nonmydriatic camera (Canon, Lake Success, New York, USA) and images were sent to the Ocular Epidemiology Reading Center at the University of Wisconsin (Madison, Wisconsin) for analysis. Optic disc-centered photographs were used. Measurements from the right eye of each participant were used for each of the retinal vascular network parameters studied. If an image from the right eye was not available, or considered ungradable, then the left eye was used as the study eye. Images from eight participants were deemed to have poor quality and were excluded from the analyses (four at Exam 2 and four at Exam 5).

The Singapore I Vessel Assessment (SIVA) is a semi-automated program which identifies arteries and veins on fundus photographs and computes retinal vascular parameters.5 SIVA (version 2.0) was used to quantitatively measure several retinal vascular network parameters from the fundus photographs taken at Exams 2 and 5. Trained graders, masked to participants’ characteristics, extracted measurements from the images with a standardized protocol using the SIVA program. The graders were masked to measurements made at prior visits, and regular quality control measures to check their grading were performed to maintain reproducibility.14 Retinal vascular network measurements were calculated based on the analysis of vessels at 0.5 to 2.0 disc diameters (zone C) from the optic disc margin.

RESEARCH IN CONTEXT

  1. Systematic review. We completed a literature review using publicly available sources (eg, PubMed). Previously, associations between retinal vascular network parameters with cognitive performance and neuroimaging findings have been reported but results have been inconsistent.

  2. Interpretation. We found longitudinal associations between greater arteriolar fractal dimensions with less isotropic free water of gray matter regions and greater gray matter cortical volume. A more complex branching pattern of the retinal arteries is associated with MRI biomarkers of less neuroinflammation and neurodegeneration. However, most of the measured retinal vascular network parameters were not independently correlated with MRI markers of cognitive decline and dementia.

  3. Future directions. This study provides a framework for the conduct of future studies exploring the (1) relationships between arteriolar versus venular retinal vascular parameters with cerebral microvasculature, (2) contributions of confounders to the associations between retinal and cerebral biomarkers, and (3) promise of artificial intelligence to analyze data from a larger, prospective study.

The three retinal vascular network measurements analyzed included branching angle, curvature tortuosity, and FD (Figure 1). Branching angle is calculated from the first angle subtended between two daughter vessels at each bifurcation.15 A larger branching angle may reflect abnormal changes in retinal vascular geometry. Curvature tortuosity is derived from the integral of the curvature square along the path of the vessel, normalized by the total path length.16 Curvature tortuosity distinguishes increased length due to bowing and points of inflection, unlike simple tortuosity which does not make this distinction.16 The tortuosity value is lower when the vessel is straighter. FD reflects complexity of the branching pattern of the retinal vascular tree using line tracings of the retinal vessels from the box-counting method.17 A lower value reflects a sparser/less complex branching pattern. Rarefaction of retinal vasculature is thought to reflect suboptimal retinal vascular branching complexity, which may reflect poor ocular blood flow in disease.18

Details are in the caption following the image
Highest (A,C,E) and lowest (B,D,F) quartile representative images of retinal vascular network parameters that were graded using the Singapore “I” Vessel Assessment (SIVA) software. (A,B) Branching angle is calculated from the first angle formed between two daughter vessels of one parent vessel at its bifurcation. (C,D) Curvature tortuosity is derived from the integral of curvature square along the vessel path, normalized by its total path length. (E,F) Fractal dimension is a measurement of the complexity of the whole branching pattern of the retinal vascular tree and is calculated by plotting the logarithm of the number of boxes against the logarithm of the size of the boxes.

2.4 Cognitive function

Cognitive performance was evaluated at Exam 5 (2010 to 2012) and Exam 6 (2016 to 2018) using three standardized and validated tests—the Cognitive Abilities Screening Instrument (CASI, version 2), a measure of global cognitive functioning; the Digit Symbol Coding (DSC), a test of processing speed; and the Digit Span (DS, forward and backward combined), a test of working memory. Each of these tests had been previously described in detail.19 At Exam 6, MESA participants at the Wake Forest field center were also administered the Uniform Data Set version 3 (UDSv3) cognitive assessment protocol.20 The National Alzheimer's Coordinating Center's UDSv3 cognitive assessment protocol includes detailed cognitive testing; information about family history of AD, medications, and health history; clinician-assessed medical conditions and judgment of symptoms; and a neurological examination. All of these data were used in a standardized adjudication process to determine cognitive impairment. The adjudication panel consisted of neuropsychologists, geriatricians, neurologists, and other aging experts. Consensus adjudicated cognitive status, according to published criteria,21, 22 included cognitively normal, mild cognitive impairment (MCI), and dementia, including subtypes. A diagnostic category of “Other” was used when cognition could not be classified due to insufficient data relating to functional impairment or indeterminant cognitive impairment.

2.5 MRI acquisition and processing

At Exam 6, brain MRI was acquired for all participants at Wake Forest on a 3T Siemens Skyra scanner using a high-resolution 32-channel head coil. MRI sequences included T1-weighted 3D volumetric MPRAGE (to quantify gray matter volume in mm3 and cortical thickness in mm), T2-weighted FLAIR (to quantify volume of white matter hyperintensities [WMH], in mm3), and Quantitative Susceptibility Mapping (QSM)/Susceptibility Weighted Imaging (SWI) to identify cerebral microbleeds. Sequence details can be found in a prior publication.23 Total intracranial volume and cortical thickness were calculated on T1 MRI using FreeSurfer v5.3 (https://surfer.nmr.mgh.harvard.edu). Cortical thickness of a temporal lobe meta region of interest (meta-ROI) was calculated by averaging surface area-weighted cortical thickness of bilateral entorhinal, inferior/middle temporal, and fusiform regions24; this meta-ROI has been shown to be a useful measure of neurodegeneration in regions characteristically impacted by AD and other age-related dementias. Diffusion tensor imaging (DTI) and neurite orientation dispersion and density imaging (NODDI)25 were acquired in a single run of diffusion scans. Free water of supratentorial white matter and gray matter, representing isotropic volume fraction of water infiltration and reduced microstructural integrity, were quantified from NODDI as described previously.23 White matter lesions were segmented by the lesion growth algorithm (LGA)26 implemented in the LST toolbox v2.0.15 (www.statistical-modelling.de/lst.html), in Matlab SPM12 (www.fil.ion.ucl.ac.uk/spm) using FLAIR images with T1 images as reference. WMH masks were manually edited by trained observers. Total WMH lesion volume was divided by total intracranial volume (to correct for head size) and log-normalized to generate a global measure of WMH volume.

Vascular lesions were read on MRI by two neuroradiologists (KH and CW) to quantify the presence and number of large and lacunar infarcts and microbleeds, applying established criteria and validated methods for ischemic lesion identification.27, 28 Representative MRI images of these vascular lesions have been previously published.27 Briefly, axial FLAIR and T1-weighted images were reviewed to identify lesions of presumed ischemic origin using a score derived from clinical stroke and MRI-pathological correlation studies based on the size, location, and imaging characteristics of each lesion. The number of microbleeds was assessed on SWI/QSM using the Microbleed Anatomical Rating Scale (MARS) and the Brain Observer MicroBleed Scale (BOMBS) rating scales to maximize intra- and inter-rater reliability and capture lesion locations.29, 30

2.6 Statistical analysis

Demographic characteristics of the analytic sample during retinal imaging were described using means and standard deviations for continuous measures and count and percent for categorical measures. Differences in retinal vascular network measurements at Exam 2 between different categories of cognitive statuses were tested using multinomial logistic regression (shown in Table S1). Multivariable linear models were used to test associations between retinal vascular network measurements at Exams 2 and 5 with cognitive performance at Exam 5 and continuous MRI biomarkers at Exam 6. In the adjusted Model 1, adjustments were made for demographics including age, sex, race/ethnicity, and level of education. In the adjusted Model 2, adjustments were made for variables in Model 1 and factors associated with microvascular disease including BMI, DM, hypertension medication, and current smoking status, which may independently increase risk of cognitive impairment. Effect moderation by age, APOE-ε4, eGFR, and cognitive status were assessed using interaction terms between these parameters and retinal imaging measures for each outcome. Results were reported as parameter estimates representing change in cognitive test score or MRI biomarker value per each SD in retinal vascular network measurement, and standard errors. All results shown are after applying the Benjamini-Hochberg procedure to correct for multiple comparisons. P values were considered significant at p < 0.05. Analyses were performed using SAS v9.4 (The SAS Institute, Cary, NC).

3 RESULTS

At the Wake Forest field site, 320 participants were recruited to undergo detailed neurocognitive testing, adjudication, and brain imaging during MESA Exam 6. Among these participants, 313 had gradable retinal photographs at Exam 2, and 306 had gradable retinal photographs at Exam 5 (Figure 2). Characteristics of the study population at the Wake Forest field site with gradable retinal photographs at Exams 2 and 5 are listed in Table 1. Three hundred and six of the 313 patients had retinal images of sufficient quality to quantify SIVA at both time points. At Exam 5, participants were 7 years older (p < 0.0001) and, on average, more likely to have lower diastolic blood pressure (p < 0.0001), treated diabetes (p < 0.0001), and lower arterial curvature tortuosity (p = 0.0006), venular curvature tortuosity (p = 0.0005), arteriolar fractal dimension (p = 0.0008), and venular fractal dimension (p = 0.0057).

Details are in the caption following the image
Flow diagram of MESA participants at the Wake Forest Site with retinal imaging, SIVA, brain imaging, and cognitive testing. At MESA Exam 1 (2000 to 2002), 1077 participants were recruited at the Wake Forest site. Of these, 985, 813, and 360 returned for follow up evaluation at Exams 2 (2003 to 2004), 5 (2010 to 2012), and 6 (2016 to 2019), respectively. At Exams 2 and 5, retinal imaging was obtained for 973 and 672 participants, respectively. No retinal imaging was performed at Exam 6. At Exam 6, cognitive testing and adjudication were completed for 320 participants. Of these 320 participants, 317 and 310 also had retinal images taken at Exams 2 and 5, respectively. Brain MRI was performed for 256 participants at Exam 6 and none at Exams 2 and 5. Cognitive testing was also completed for 752 participants at Exam 5 and none at Exam 2. CASI, Cognitive Abilities Screening Instrument; DS, Digit Span; DSC, Digit Symbol Coding; GM, gray matter; MESA, Multi-Ethnic Study of Atherosclerosis; MRI, magnetic resonance imaging; QC, quality control; SIVA, Singapore I Vessel Assessment; WM, white matter.
TABLE 1. Demographics of MESA participants with retinal imaging at Exam 2 and Exam 5 and multimodal neuroimaging and cognitive testing at Exam 6.
Exam 2 (n = 313) Exam 5 (n = 306)
n/mean %/SD n/mean %/SD p-value
Age (years) 59 7.3 66 7.1 <0.0001*
Sex 0.9264
Male 140 44.7 138 45.1
Female 173 55.3 168 54.9
Race 0.959
White 154 49.2 147 48.1
Black 158 50.5 158 51.6
Hispanic 1 0.3 1 0.3
Education 0.9998
High school or less 57 18.2 55 18.0
Some college 97 31.0 95 31.0
Undergraduate degree 74 23.6 73 23.9
Graduate degree 85 27.2 83 27.1
Smoker status 0.8788
Never 150 47.9 152 49.7
Former 125 39.9 120 39.2
Current 38 12.1 34 11.1
Body mass index 29 5.2 29 5.3 0.9586
Systolic blood pressure (mmHg) 124 20.6 123 19.1 0.6628
Diastolic blood pressure (mmHg) 72 11.0 69 10.7 <0.0001*
Diabetes status <0.0001*
Normal 227 72.5 206 67.3
Pre-diabetes 54 17.3 40 13.1
Untreated diabetes 13 4.2 5 1.6
Treated diabetes 19 6.1 55 18.0
Exam 6 diagnosis 0.9989
Normal 206 65.8 202 66.0
Mild cognitive impairment 80 25.6 78 25.5
Dementia 14 4.5 14 4.6
Othera 13 4.2 12 3.9
Arteriolar branching angle 77 11.8 79 12.0 0.1795
Venular Branching angle 78 10.3 78 11.1 0.3731
Arterial curvature tortuosity 7.15E-05 1.56E-05 6.76E-05 1.92E-05 0.0006*
Venular curvature tortuosity 7.83E-05 1.65E-05 7.42E-05 1.98E-05 0.0005*
Arteriolar fractal dimensions 1.19 0.05 1.18 0.06 0.0008*
Venular fractal dimensions 1.18 0.05 1.17 0.05 0.0057*
  • Abbreviation: MESA, Multi-Ethnic Study of Atherosclerosis.
  • a Includes cognitive diagnostic categories where cognition could not be classified due to insufficient data relating to functional impairment or indeterminant cognitive impairment.
  • * Denotes statistical significance.

As shown in Table 2, in unadjusted analyses, venular curvature tortuosity and arteriolar FD at Exam 2 were associated with CASI (global cognitive functioning) at Exam 5, while arteriolar FD at Exams 2 and 5 were associated with DSC (processing speed) at Exam 5. After adjustment for covariates in both models 1 and 2, there were no significant associations between any of the retinal vascular network measurements with performance on CASI, digit symbol coding, digit span (working memory), or changes in CASI over time at either time point. Table S1 also shows that retinal vascular network measures at Exam 2 did not differ by adjudicated cognitive status at Exam 6.

TABLE 2. Association between cognitive performance at Exam 5 with retinal vascular measurements at Exams 2 and 5.
Exam 2 Exam 5
Unadjusted Model 1 Model 2 Unadjusted Model 1 Model 2
Cognitive performance at Exam 5 Beta SE p-value Beta SE p-value Beta SE p-value Beta SE p-value Beta SE p-value Beta SE p-value
CASI (global cognitive function) (nexam2 = 303; nexam5 = 301)
Arteriolar BA 0.45 0.33 0.33 0.14 0.29 1.00 0.15 0.29 0.94 −0.54 0.33 0.29 −0.41 0.29 0.91 −0.36 0.29 1.00
Venular BA 0.38 0.33 0.39 0.12 0.29 1.00 0.14 0.30 1.00 −0.08 0.31 0.79 −0.18 0.28 0.62 −0.13 0.28 0.79
Arteriolar CT −0.22 0.37 0.67 0.02 0.32 0.94 0.06 0.33 0.90 0.23 0.34 0.75 0.16 0.30 0.59 0.16 0.30 0.69
Venular CT −1.24 0.36 0.004* −0.68 0.32 0.23 −0.75 0.33 0.18 −0.26 0.30 0.78 −0.23 0.27 1.00 −0.25 0.28 0.70
Arteriolar FD 0.99 0.34 0.01* 0.49 0.31 0.35 0.52 0.32 0.32 0.73 0.33 0.16 0.23 0.30 0.89 0.19 0.31 1.00
Venular FD −0.08 0.35 0.81 −0.09 0.32 0.94 −0.08 0.32 0.97 −0.15 0.33 0.78 −0.21 0.30 0.73 −0.18 0.30 0.92
Change in CASI between Exams 5 and 6 (nexam2 = 303; nexam5 = 301)
Arteriolar BA 0.33 0.40 0.50 0.20 0.40 0.74 0.33 0.40 0.62 0.27 0.40 0.61 0.20 0.39 0.93 0.17 0.40 1.00
Venular BA 0.44 0.41 0.42 0.47 0.41 0.75 0.47 0.41 0.78 0.31 0.39 0.63 0.35 0.38 1.00 0.35 0.39 1.00
Arteriolar CT 0.08 0.45 0.85 0.06 0.44 0.89 0.13 0.46 0.77 −0.60 0.42 0.90 −0.60 0.41 0.84 −0.58 0.42 0.99
Venular CT 0.65 0.44 0.29 0.55 0.45 1.00 0.64 0.46 1.00 0.07 0.37 0.85 −0.18 0.37 0.75 −0.14 0.39 0.87
Arteriolar FD 0.66 0.42 0.74 0.46 0.43 0.58 0.44 0.44 0.62 0.52 0.41 0.60 0.34 0.41 0.83 0.31 0.42 0.92
Venular FD 0.64 0.43 0.42 0.24 0.44 0.87 0.19 0.44 0.79 0.50 0.40 0.43 −0.01 0.41 0.97 −0.02 0.42 0.96
Digit Symbol Coding (speed of processing) (nexam2 = 300; nexam5 = 292)
Arteriolar BA 0.68 0.91 0.68 −0.12 0.78 0.88 0.14 0.77 1.00 −0.40 0.90 0.79 −0.03 0.78 0.97 −0.14 0.77 0.90
Venular BA 0.59 0.92 0.62 −0.32 0.80 1.00 −0.50 0.79 0.96 −0.33 0.87 0.71 −0.50 0.76 0.77 −0.55 0.75 0.92
Arteriolar CT −0.61 1.02 0.55 −0.28 0.88 0.90 −0.09 0.88 1.00 −1.04 0.95 0.41 −1.07 0.82 0.39 −0.81 0.81 0.46
Venular CT −2.00 1.00 0.14 −1.02 0.87 0.73 −0.81 0.88 1.00 −1.13 0.83 0.35 −1.63 0.73 0.15 −1.59 0.73 0.13
Arteriolar FD 3.38 0.94 0.002* 1.36 0.85 0.66 1.30 0.85 0.82 3.68 0.90 0.0004* 1.73 0.82 0.11 1.54 0.81 0.26
Venular FD 0.42 0.95 0.31 0.27 0.85 1.00 0.20 0.83 0.97 0.42 0.92 0.36 0.24 0.83 0.93 0.28 0.81 0.95
Digit Span total (working memory) (nexam2 = 301; nexam5 = 299)
Arteriolar BA 0.41 0.21 0.29 0.27 0.19 1.00 0.31 0.19 0.74 0.05 0.21 0.99 0.11 0.19 0.67 0.09 0.19 0.65
Venular BA 0.28 0.21 0.29 0.16 0.20 1.27 0.13 0.20 0.85 0.04 0.20 0.85 0.07 0.18 0.70 0.05 0.19 0.78
Arteriolar CT 0.04 0.24 1.00 0.15 0.21 0.73 0.19 0.22 1.00 0.34 0.22 0.34 0.33 0.19 0.52 0.39 0.20 0.40
Venular CT −0.36 0.23 0.26 −0.07 0.22 0.89 −0.02 0.22 0.98 0.08 0.19 1.00 0.14 0.18 0.88 0.16 0.18 0.85
Arteriolar FD 0.42 0.22 0.19 0.15 0.21 0.92 0.12 0.21 0.76 0.42 0.21 0.29 0.15 0.20 0.67 0.13 0.20 0.66
Venular FD 0.006 0.23 0.98 −0.01 0.21 0.95 −0.01 0.21 1.00 0.21 0.21 0.62 0.21 0.20 0.87 0.23 0.20 0.98
  • Note. Model 1 adjusted for age, sex, race/ethnicity, level of education. Model 2 adjusted for age, sex, race/ethnicity, level of education, systolic blood pressure, BMI, DM presence, and hypertension medication.
  • Abbreviations: BA, branching angle; CASI, Cognitive Abilities Screening Instrument; CT, curvature tortuosity; FD, fractal dimension.
  • * Denotes statistical significance after Benjamini-Hochberg correction for multiple comparisons.

Table 3 presents associations of retinal vascular network measurements at Exams 2 and 5 with MRI biomarkers at Exam 6. In unadjusted analyses, significant associations were noted between arteriolar FD at Exams 2 and 5 with gray matter cortical volume and isotropic free water of gray matter regions; venular FD at Exam 2 with isotropic free water of gray matter regions; arteriolar and venular FD at Exams 2 and 5 with cortical thickness meta-ROI; as well as venular FD at Exam 2 with isotropic free water of white matter regions. However, after adjusting for demographics in Model 1, only the association between greater arteriolar FD at Exam 5 with greater gray matter cortical volume; and greater arteriolar FD at Exam 2 with lower isotropic free water fraction of gray matter regions remained statistically significant. These associations remained significant after additional adjustment for vascular and metabolic risk factors in Model 2. The association between greater arteriolar FD at Exam 5 with lower isotropic free water of gray matter regions almost reached statistical significance (p = 0.07) in Model 1. There were no significant associations between branching angle or curvature tortuosity with isotropic free water fraction of gray matter regions, gray matter cortical volume, cortical thickness meta-ROI, WMH, or isotropic free water of white matter regions.

TABLE 3. Association between MRI biomarkers at Exam 6 with retinal vascular measurements at Exams 2 and 5.
Exam 2 Exam 5
Unadjusted Model 1 Model 2 Unadjusted Model 1 Model 2
MRI biomarkers at Exam 6 Beta SE p-value Beta SE p-value Beta SE p-value Beta SE p-value Beta SE p-value Beta SE p-value
Gray matter cortical volume (nexam2 = 243; nexam5 = 242)
Arteriolar BA 817 1995 0.82 −704 1875 0.85 −358 1915 1.02 452 2002 0.99 −875 1862 0.77 −850 1898 0.65
Venular BA 837 2025 1.02 787 1911 1.02 1336 1920 0.97 91 1862 0.96 201 1725 0.91 938 1776 0.72
Arteriolar CT −879 2359 0.71 16 2205 0.99 177 2254 0.94 −1422 2029 0.73 −1849 1862. 0.64 −1624 1913 0.79
Venular CT −4154 2637 0.23 −1182 2542 1.29 −1007 2616 1.05 −1665 2140 0.87 −1392 2040. 0.74 −1476 2074 0.72
Arteriolar FD 6990 2175 0.01* 4352 2083 0.23 3390 2129 0.68 8416 2046 0.0003* 5302 1984 0.049* 5458 2017 0.04*
Venular FD 3518 2217 0.34 1772 2114 1.21 1545 2127 1.41 4002 2052 0.16 2552 2002 0.61 1945 2032 1.02
Isotropic free water of gray matter regions (nexam2 = 230; nexam5 = 230)
Arteriolar BA 0.0047 0.0029 0.62 0.0008 0.0024 0.89 0.0016 0.0028 1.00 0.0050 0.0024 0.95 0.0039 0.0024 0.17 0.0032 0.0023 0.22
Venular BA 0.0058 0.0033 0.53 0.0006 0.0026 0.82 0.0035 0.0032 0.85 0.0045 0.0027 0.94 −0.0006 0.0022 0.94 0.0045 0.0026 0.60
Arteriolar CT −0.0102 0.0027 0.17 0.0026 0.0028 0.69 −0.0080 0.0026 1.13 −0.0084 0.0025 0.12 0.0042 0.0023 0.14 −0.0062 0.0025 0.25
Venular CT −0.0074 0.0027 0.16 0.0045 0.0032 0.48 −0.0009 0.0027 0.84 −0.0032 0.0025 0.18 0.0052 0.0026 0.15 0.0011 0.0025 0.27
Arteriolar FD −0.0012 0.0025 0.001* −0.0089 0.0026 0.004* −0.0005 0.0024 0.01* −0.0084 0.0025 0.01* −0.0064 0.0025 0.07 −0.0059 0.0024 0.10
Venular FD −0.0020 0.0026 0.02* −0.0025 0.0027 0.52 −0.0005 0.0025 1.11 −0.0006 0.0023 0.31 −0.0006 0.0025 0.81 −0.0015 0.0022 0.68
Cortical thickness meta-ROIa (nexam2 = 243; nexam5 = 242)
Arteriolar BA 0.0138 0.0078 0.15 0.0096 0.0079 0.43 0.094 0.0080 0.48 0.0066 0.0079 0.49 −0.0006 0.0079 0.94 −0.0003 0.0080 0.97
Venular BA −0.0001 0.0080 0.99 −0.0041 0.0080 0.90 −0.0030 0.0081 1.06 −0.0112 0.0073 0.25 −0.0144 0.0072 0.14 −0.0118 0.0075 0.23
Arteriolar CT −0.0046 0.0093 0.75 −0.0015 0.0093 0.87 −0.0006 0.0095 0.95 −0.0063 0.0080 0.43 −0.0089 0.0079 0.39 −0.0078 0.0081 0.40
Venular CT −0.0086 0.0105 0.62 0.0022 0.0107 1.00 0.0026 0.0111 0.98 0.0092 0.0083 0.40 0.0086 0.0085 0.37 0.0098 0.0086 0.39
Arteriolar FD 0.0265 0.0086 0.01* 0.0209 0.0087 0.10 0.0185 0.0090 0.24 0.0253 0.0082 0.01* 0.0154 0.0085 0.14 0.0170 0.0086 0.15
Venular FD 0.0254 0.0087 0.01* 0.0149 0.0089 0.27 0.0125 0.0090 0.50 0.0283 0.0079 0.003* 0.0210 0.0083 0.08 0.0187 0.0085 0.18
White matter hyperintensities∼ (nexam2 = 242; nexam5 = 241)
Arteriolar BA −0.1019 0.0923 0.53 0.0006 0.0889 0.99 −0.0886 0.0848 1.00 −0.0994 0.0931 0.57 −0.0154 0.0885 1.00 −0.0165 0.0884 1.00
Venular BA −0.0430 0.0939 0.77 −0.0224 0.0904 0.96 0.0843 0.0849 0.97 0.0453 0.0868 0.72 0.0870 0.0823 0.58 0.0549 0.0832 1.00
Arteriolar CT 0.0338 0.1095 0.76 0.0820 0.1043 1.00 0.0076 0.1000 0.94 0.0472 0.0946 0.62 0.0956 0.0887 0.85 0.00698 0.0892 1.00
Venular CT 0.0610 0.1230 0.93 0.1809 0.1200 0.79 −0.1095 0.1162 0.69 −0.1146 0.0982 0.73 0.0029 0.0958 0.98 −0.0151 0.0955 0.87
Arteriolar FD −0.1505 0.1029 0.87 −0.0645 0.1007 1.00 −0.0105 0.0962 1.00 −0.0755 0.0986 0.67 −0.0055 0.0963 1.00 0.0165 0.0961 1.00
Venular FD −0.1368 0.1037 0.56 0.0279 0.1011 1.00 −0.0681 0.0953 0.71 −0.2262 0.0952 0.11 −0.1121 0.0951 1.00 −0.0679 0.0947 1.00
Isotropic free water of white matter regions (nexam2 = 235; nexam5 = 234)
Arteriolar BA −0.0022 0.0017 0.29 −0.0014 0.0017 0.49 −0.0027 0.0017 0.21 −0.0012 0.0017 0.75 0.0012 0.0018 0.73 0.0017 0.0017 0.68
Venular BA −0.0028 0.0017 0.22 −0.0018 0.0018 0.48 −0.0022 0.0017 0.32 −0.0007 0.0016 0.78 −0.0007 0.0016 0.67 −0.0016 0.0016 0.99
Arteriolar CT −0.0019 0.0020 0.40 −0.0023 0.0020 0.50 −0.0035 0.0019 0.43 0.0022 0.0016 0.39 0.0014 0.0017 0.83 0.0006 0.0017 0.85
Venular CT −0.0008 0.0022 0.72 0.0001 0.0023 0.96 0.0008 0.0023 0.74 −0.0001 0.0017 0.97 0.0009 0.0018 0.76 0.0002 0.0018 0.93
Arteriolar FD −0.0045 0.0018 0.05 −0.0042 0.0019 0.15 −0.0031 0.0018 0.27 −0.0027 0.0018 0.39 −0.0028 0.0018 0.79 −0.0025 0.0018 1.00
Venular FD −0.0060 0.0018 0.01* −0.0023 0.0019 0.67 −0.0013 0.0018 0.60 −0.0042 0.0017 0.09 −0.0026 0.0018 0.45 −0.0014 0.0018 0.63
  • Note. Model 1 adjusted for age, sex, race/ethnicity, level of education. Model 2 adjusted for age, sex, race/ethnicity, level of education, systolic blood pressure, BMI, DM presence, and hypertension medication.
  • Abbreviations: BA, branching angle; CT, curvature tortuosity; FD, fractal dimension.
  • a Weighted average of thickness in entorhinal cortex, inferior temporal, mid-temporal, inferior parietal, fusiform, and precuneus regions.
  • * Denotes statistical significance after Benjamini-Hochberg correction for multiple comparisons.

Supplemental table 2 compares the relationships between retinal vascular network measurements at Exams 2 and 5 with microbleeds and lacunar infarcts on brain MRI at Exam 6. With or without adjustment, there was no association between any of the retinal vascular network measurements studied with presence of cerebral microbleeds or lacunar infarcts. In the associations between cognitive performance/MRI biomarkers and retinal vascular network measures, there were no significant interactions found with age, APOE-ε4 status, eGFR, or adjudicated cognitive status.

4 DISCUSSION

This study examines the associations between a range of retinal vascular network parameters and cognitive performance, cognitive decline, cognitive impairment, and brain structure on MRI in a population-based cohort of predominantly healthy persons. Few consistent associations were seen after controlling for demographics and other risk factors of dementia and cognitive decline. However, we demonstrated an association between retinal arteriolar FD and MRI biomarkers that were assessed up to 8 or 16 years after fundus imaging. To the best of our knowledge, this is the first study to report such findings, which may be important as they demonstrate the potential relationship of retinal arteriolar FD with brain MRI biomarkers of gray matter macrostructure and microstructure but not cognitive performance or cerebral small vessel disease. There was no effect modification of these relationships by age, APOE-ε4, eGFR, and cognitive status, suggesting that the observed relationships were generalizable to various subgroups of interest.

FD of the retinal vascular network was first described by Mandelbrot and Wheeler31 and reflects a computational method to integrate the density and overall complexity of the arborization of the retinal vascular network into a single number. Murray's Law of Minimal Work dictates that arborization of the retinal vascular tree follows a spatial pattern that minimizes energy consumption and shear stress across any vascular network, and this law influences the relationship between the radii of mother and daughter branches in a network and the angle at which vessels bifurcate.32

Our analysis in MESA did not detect a significant association of FD with cognitive status, and reflects conflicting results in prior literature.7, 15, 33-36 FD is also inconsistently associated with cognitive performance in prior literature. For example, a significant association between FD and cognitive performance was reported for the Abbreviated Mental Test,15 but not the Mini-Mental State Examination35 or the Montreal Cognitive Assessment.34 Our unadjusted data indicated an association between FD and CASI, and FD and DSC but these relationships became not significant after adjustment for confounders. Our data support the likelihood that the relationship between retinal vascular network parameters and cognitive performance is complex, is attenuated by demographics such as age, sex, race/ethnicity, and level of education, and is dependent, in part, upon the cognitive task itself.

Our work examined retinal vascular network parameters and related them to brain structure on MRI measured later in life. In unadjusted models correcting for multiple comparisons, we found significant associations between arteriolar FD and gray matter cortical volume, isotropic free water of gray matter regions, and cortical thickness meta-ROI. We also found significant associations between venular FD and isotropic free water of gray matter regions, cortical thickness meta-ROI, and free water of white matter regions. Adjustments for demographics fully attenuated most associations with MRI measures except for greater arteriolar FD with greater gray matter cortical volume and lower isotropic free water of gray matter regions, and these associations remained significant after further adjustments for common vascular and metabolic disorders, which are in the causal pathway linking retinal abnormalities with cognitive and brain structural abnormalities.

Although we detected associations between FD and MRI biomarkers in both the gray and white matter, only associations with gray matter biomarkers remained significant after adjustment. Of note, gray matter cortical volume, but not cortical thickness meta-ROI, remained significant after adjustment. Cortical thickness meta-ROI incorporates brain regions most affected in AD and other age-related dementias (ie, entorhinal cortex, inferior temporal, middle temporal, and fusiform regions). However, cortical volume may be a more reliable indicator of neurodegenerative pathology as volume measurements are generally more reliable than thickness measurements.24

White matter abnormalities are also widely reported in patients with MCI and AD.37 On MRI, white matter pathology in AD is observed as white matter atrophy and WMH.38, 39 Previous studies suggest that FD is associated with white matter abnormalities on MRI in older adults.40 In contrast, after adjustment, our study did not find an association between FD and MRI biomarkers in the white matter. We hypothesize this is because microvascular density in the gray matter of the brain is three times greater than that in the white matter.41 Thus, the vascular branching network of the retina may reflect pathologic changes in the cerebral gray matter microvasculature earlier in disease than that of white matter microvasculature. Further longitudinal study in this cohort is warranted to assess the impact of disease stage on associations between FD and white matter biomarkers.

Free water is a more recent diffusion-weighted imaging-based biomarker of neurodegeneration. Free water accumulates in the extracellular space of brain parenchyma as a result of pathologic processes such as inflammation, infection, or trauma.38 Free water in the white matter has been associated with cognitive decline, disease severity,42 MCI, or AD.38, 43 In our study, associations between arteriolar FD with free water in the gray matter persisted after covariate adjustment and correction for multiple comparisons, while association in the white matter were attenuated.

This work adds to the literature that describes inconsistent relationships between retinal vascular network parameters and brain MRI findings. Hilal and coauthors found that venular caliber, arteriolar FD, and arteriolar tortuosity were significantly associated with cerebral microbleeds, but not lacunar infarcts or white matter lesions volume in an elderly Chinese Singaporean population of 261 participants.44 Meanwhile, Crystal and colleagues found in a group of 34 HIV-infected and 21 HIV-uninfected women that higher venous FD was associated with smaller cortical volumes and lower fractional anisotropy, while higher arterial FD was associated with the opposite patterns.45 Most recently, Nadal et al. showed in 26 participants from the Cognitive REServe and Clinical ENDOphenotype (CRESCENDO) project that cerebral blood flow was associated with venular FD, arteriolar tortuosity, and arteriolar branching angle, but there were no significant associations with arteriolar FD, venular tortuosity, or venular branching angle.9

The strengths of this study include its longitudinal design over decades. The study is limited by the smaller number of participants, as SIVA parameters were only obtained for participants from the Wake Forest field site as an exploratory study of a subsection of the overall MESA cohort. A large number of multiple comparisons were performed in this study, but the likelihood of type 1 error was minimized through the use of the Benjamini-Hochberg procedure with a false discovery rate of 0.05 and adjustments for possible confounding variables. The Benjamini-Hochberg procedure is a standard method of controlling for multiple comparisons, as it is scalable, adaptive, and maintains a higher statistical power than other methods.46 These confounding variables included social, environmental, and demographic characteristics. However, the possibility of unadjusted unmeasured confounding is still possible. Another limitation is that ophthalmic diagnoses were not included as potential confounders in this study; however, the most common conditions that could influence retinal vascular network changes (hypertension and diabetes) were included in the list of confounders and adjusted for in the statistical models. Moreover, SIVA, like any semi-automated software, requires significant human input, which is subjective in nature. This could lead to variability in retinal vessel caliber measurements,47 and further contributes to the limitations of the study. The use of artificial intelligence deep-learning algorithms48 may alleviate this limitation in the future. Recent publications have shown promising results in the use of deep learning models to identify patients with cognitive decline and dementia utilizing retinal photographs alone.48, 49

The detection of differences in arteriolar FD on retinal imaging were apparent years before associated abnormalities in gray matter cortical volume and isotropic free water of gray matter reflective of neuroinflammation and neurodegeneration were detected on brain MRI. Our data also identified significant, unadjusted associations that underscore the influence of one or more confounders that may be relevant for future study, that is, synergistic or combinatorial association. These findings should be validated in a larger population of participants in a prospective study.

ACKNOWLEDGMENTS

The authors gratefully thank Jordan E. Tanley for her assistance with statistical analysis, as well as Haslina Binte Hamzah and Nishal Banu Binte Makdoom for their assistance with the preparation of Figure 1. This research was supported by contracts 75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, N01-HC-95163, 75N92020D00004, N01-HC-95164, 75N92020D00007, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168 and N01-HC-95169 and by grant R01HL127659 from the National Heart, Lung, and Blood Institute; by grants UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420 from the National Center for Advancing Translational Sciences (NCATS); by grants P30AG049638, R01AG054069 and R01AG058969 from the National Institute on Aging (NIA), and research award ZIAEY000403 from the NIH Intramural Research Program. The authors thank the other investigators, the staff, and the participants of the MESA study for their valuable contributions. A full list of participating MESA investigators and institutions can be found at http://www.mesa-nhlbi.org. The funding sources had no involvement in study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the article for publication.

    CONFLICT OF INTEREST STATEMENT

    The authors declare that no conflicts of interest exist. Author disclosures are available in the supporting information.

    CONSENT STATEMENT

    Written informed consent was obtained from all participants prior to their inclusion in the study.