Вы находитесь на странице: 1из 10
Quanti fi cation of Diabetic Macular Ischemia Using Optical Coherence Tomography Angiography and Its Relationship

Quanti cation of Diabetic Macular Ischemia Using Optical Coherence Tomography Angiography and Its Relationship with Visual Acuity

Angiography and Its Relationship with Visual Acuity Wasim A. Samara, MD, Abtin Shahlaee, MD, Murtaza K.

Wasim A. Samara, MD, Abtin Shahlaee, MD, Murtaza K. Adam, MD, M. Ali Khan, MD, Allen Chiang, MD, Joseph I. Maguire, MD, Jason Hsu, MD, Allen C. Ho, MD

Purpose: To quantify foveal avascular zone (FAZ) area and macular vascular density objectively using optical coherence tomography angiography (OCTA) and to examine correlations with visual acuity in eyes with diabetic retinopathy (DR) in the absence of diabetic macular edema. Design: Retrospective observational case series. Participants: Eighty-four eyes from 55 patients with DR and 34 control eyes from 27 age-matched healthy participants. Methods: All eyes underwent OCTA (RTVue-XR Avanti; Optovue, Inc, Fremont, CA). Integrated automated algorithms were used to quantify FAZ area and macular vascular density. Main Outcome Measures: FAZ area, vessel area density (VAD), vessel length density (VLD), and visual acuity. Results: In each study eye, DR was classied as mild nonproliferative DR (NPDR; n ¼ 32 [38%]), moderate-

to-severe NPDR (n ¼ 31 [37%]), or proliferative DR (n ¼ 21 [25%]). Mean FAZ area was greater in diabetic eyes

compared with control eyes both in the super cial (0.427 mm

vs. 0.372 mm 2 ; P < 0.001) vascular networks. Mean VAD was lower in diabetic eyes compared with control eyes

in both the super cial (49.44% vs. 55.09%; P < 0.001) and deep (56.65% vs. 61.32%; P < 0.001) networks. Mean

mm 1 vs. 21.55

mm 1 ; P < 0.001) and deep (21.19 mm

vs. 24.38 mm 1 ; P < 0.001) networks. In all eyes, there was a sta-

VLD was also lower in diabetic eyes compared with control eyes in both the super cial (17.68

2

vs. 0.275 mm 2 ; P < 0.001) and deep (0.616 mm

2

1

tistically signi cant negative correlation between the logarithm of the minimum angle of resolution (logMAR) visual acuity and the vascular density in both the super cial (VAD, r ¼ 0.52; VLD, r ¼ 0.54; P < 0.001) and deep (VAD, r ¼ 0.50; VLD, r ¼ 0.50; P < 0.001) networks. A positive correlation was found between logMAR visual acuity and FAZ area in both the super cial ( r ¼ 0.29; P < 0.01) and deep ( r ¼ 0.48; P < 0.001) networks. Conclusions: Automated quantitative algorithms allow for objective assessment of retinal vascular changes in eyes with DR that are correlated to visual acuity. These methods may prove useful in monitoring disease progression and identifying parameters that affect visual function. Ophthalmology 2017;124:235-244 ª 2016 by the American Academy of Ophthalmology

Diabetic macular ischemia (DMI) is important in the path- ogenesis of diabetic retinopathy (DR) and is characterized by narrowing or occlusion of retinal capillaries. 1 Resulting tissue hypoxia leads to an increase in vascular endothelial growth factor levels, resulting in diabetic macular edema (DME). 2 Although DME is the most common cause of vision loss in patients with DR, 3,4 DMI also has been shown to result in vision loss, regardless of the presence or absence of DME. 5 Diabetic macular ischemia has been well characterized using uorescein angiography (FA), where there is enlargement of the foveal avascular zone (FAZ) and retinal capillary loss. 6,7 Fluorescein angiography remains an important method for imaging retinal vasculature in DR with the ability to detect microaneurysms, venous beading,

ª 2016 by the American Academy of Ophthalmology Published by Elsevier Inc.

capillary nonperfusion, and leakage. Nevertheless, it is an invasive, time-consuming test that requires dye injection, which can lead to several adverse effects that rarely may be life threatening. 8,9 Moreover, because of light scatter by inner retinal layers, FA fails to image the deep capillary network and provides information only about super cial capillary network perfusion. 10 The recent development of optical coherence tomogra- phy angiography (OCTA) has allowed for acquisition of high-resolution depth-resolved images of the retinal vascular layers in a rapid, noninvasive manner. 11 Optical coherence tomography angiography has been used to describe the retinal vasculature in several diseases and in healthy eyes with high reliability and reproducibility. 12 e 16 More recently, different automated quanti cation algorithms have

ISSN 0161-6420/16

235

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

Ophthalmology Volume 124, Number 2, February 2017

been used to extract angiographic data, including FAZ area, vascular density, areas of non ow, and vessel length from OCTA scans. 15,17,18 These algorithms lately have been used to quantify several angiographic parameters in DR eyes as markers for DMI. 12,15 However, these studies have not evaluated the association of these parameters with visual acuity. In this study, we performed a detailed quantitative analysis of FAZ area and macular vascular density measured by vessel area density (VAD) and vessel length density (VLD). In addition, we investigated the correlation of these parameters with visual acuity in a cohort of patients with DR without DME.

Methods

Participants

This retrospective observational cases series was approved by the Institutional Review Board of Wills Eye Hospital and complied with the Health Insurance Portability and Accountability Act. Patients with a diagnosis of diabetes mellitus with DR who underwent OCTA of one or both eyes at the Retina Service of Wills Eye Hospital between January 1, 2015, and August 30, 2015, were evaluated. All diabetic participants underwent standard ophthalmic examination and spectral-domain optical coherence tomography (SD OCT; Spectralis; Heidelberg Engineering, Heidelberg, Germany), and the stage of retinopathy was determined with fundus examination based on the International Clinical Diabetic Retinopathy and Diabetic Macular Edema Severity scale before the acquisition of the OCTA scans. 19 Patients with DME on SD OCT, concomitant retinal disease (age-related macular degeneration or retinal vascular occlusion), or both were excluded. Age-matched healthy controls with no history of systemic disorders were selected from a normative database collected by our service. The healthy participants underwent dilated ophthalmic examination before obtaining OCTA scans to ensure a healthy macular status. Best-corrected visual acuity was measured based on current spectacle correction with pinhole.

Optical Coherence Tomography Angiography Image Acquisition

Optical coherence tomography angiography images were obtained with the AngioVue OCTA system on the commercially available Avanti SD OCT device (Optovue, Inc, Fremont, CA). This system uses split-spectrum amplitude decorrelation angiography algorithm (version 2015.100.0.35) and operates at 70 000 A-scans per second to acquire OCTA volumes consisting of 304 304 A-scans. Orthogonal registration and merging of 2 consecutive scans were used to obtain OCTA volume scans over a central 3 3-mm area. Optical coherence tomography angiography images of the super- cial and deep capillary networks were generated separately using the automated software algorithm. Based on these default settings, the boundaries of supercial network extended from 3 mm below the internal limiting membrane to 15 mm below the inner plexiform layer. The deep capillary network extended from 15 to 70 mm below the inner plexiform layer. Two experienced independent graders (W.A.S., A.S.) reviewed the OCTA images. Patients with poor image quality were excluded based on the presence of one or more of the following criteria: low signal strength index (SSI; < 50), presence of 1 or more blink artifacts, poor xation leading to motion or doubling artifacts, media opacity obscuring view of the vasculature, and presence of

236

cystoid macular changes causing disrupted retinal anatomic features and segmentation errors. Of 108 eyes with DR without DME, 24 eyes (22%) were excluded based on the aforementioned image quality criteria.

Vascular Density and Foveal Avascular Zone Measurement

Using the acquired images, measurements of the FAZ area were calculated using the nonow function on the OCTA software at the level of the supercial and deep vascular networks ( Fig 1). Vascular density values were calculated for the entire en face scan for super cial and deep networks excluding the FAZ area. To allow quanti cation of vascular density, the OCTA software converts the obtained scans into 2 forms of binary images using an automated thresholding algorithm. One form represents binarization of the original scan obtained, allowing for measurement of VAD ( Fig 1C). Vessel area density is calculated as the percentage area occupied by blood vessels, with the blood vessels being de ned as pixels having decorrelation values above the threshold level. Vessel area density is calculated using the following formula:

VAD ¼ area occupied by vasculature ð pixels Þ =

ð total scan area FAZ area Þðpixels Þ :

The second binary form is obtained by skeletonizing the acquired scan into 1-pixel e wide vessels allowing for measurement of VLD ( Fig 1D). Vessel length density is calculated by measuring the total vessel length in the obtained scan using the following formula:

VLD ¼ length of skeletonized vasculature ð mmÞ =

ð total scan area FAZ area Þ mm 2 :

Statistical Analysis

Statistical analyses were performed with SPSS software version

20 (SPSS, Inc, Chicago, IL). Best-corrected visual acuity was

converted to the logarithm of the minimum angle of resolution (logMAR). Variable normality was inspected using histograms and the Shapiro eWilk test. The FAZ area, supercial VAD, and visual acuity data were skewed negatively with no simple transformation to redress the skewed data, and accordingly, the Manne Whitney U test or unpaired t test was used to compare FAZ area and vascular density between diabetics and controls based on variable normality. Analysis of variance with the post hoc Tukey (parametric) or Kruskal-Wallis (nonparametric) test was used to compare age, central macular thickness, FAZ area, and vascular density between different stages of DR and controls. Statistical test application was dependent on the normality of data for each variable. Spearman s correlation was used to assess the relationship between logMAR visual acuity and vascular density as well as FAZ area. A P value of less than 0.05 was considered statistically signi cant.

Results

Eighty-four eyes from 55 patients with DR and 34 control eyes from

27 age-matched healthy participants were included in this study.

Baseline characteristics are listed in Table 1. Diabetic retinopathy was classi ed as mild nonproliferative diabetic retinopathy (NPDR) in 32 eyes (38%), moderate-to-severe NPDR in 31 eyes (37%), and proliferative DR (PDR) in 21 eyes (25%). Thirty-two diabetic eyes (38%) had been treated previously. Prior treatments are listed in Table 2. There was no signi cant difference in age across all groups ( P ¼ 0.17). Mean visual acuity was 0.19 0.136 logMAR in diabetic eyes (Snellen equivalent, 20/31; range, 20/20 e 20/80) and 0.03 0.047 logMAR in control eyes

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

Samara et al Quantication of DMI with OCTA

Samara et al Quanti fi cation of DMI with OCTA Figure 1. Foveal avascular zone (FAZ)

Figure 1. Foveal avascular zone (FAZ) and macular vascular density measurement using optical coherence tomography angiography (OCTA) in an eye with mild nonproliferative diabetic retinopathy. The 3 3-mm OCTA scans are centered on the macula at the level of the ( A) supercial and (F) deep retinal vascular networks. An automated software algorithm integrated into the OCTA system graphically highlighted and calculated the FAZ area in the (B) supercial and (G) deep networks. Vascular density is measured in the entire image by converting the acquired OCTA scan to a binary image. Vessel area density (VAD) is calculated from a binary image representing the original scan in the ( C) supercial and (H) deep networks. Vessel length density (VLD) is calculated from a binary image after skeletonization in the (D) supercial and (I) deep networks. The respective cross-sectional reference lines are shown on the B-scans for the (E ) supercial and (J) deep networks.

(Snellen equivalent, 20/21; range 20/20 e20/25). Central macular thickness was similar between all groups ( P ¼ 0.20). Signal strength index was higher in control eyes compared with diabetic eyes ( P < 0.001, unpaired t test). On analysis of variance with post hoc Tukey analysis, SSI was statistically higher in control eyes compared with mild NPDR eyes ( P ¼ 0.04) and moderate- to-severe NPDR eyes ( P ¼ 0.001); however, it was similar in control eyes when compared with PDR eyes ( P ¼ 0.08). Examples of FAZ, VAD, and VLD measurements are shown in Figure 2. Table 3 discloses corresponding measurements in the supercial and deep vascular networks in eyes with DR and controls. Foveal avascular zone area was signicantly greater and vascular density (both VAD and VLD) was signicantly lower in all diabetic eyes compared with control eyes (all P < 0.001). Subgroup analysis was performed to compare FAZ area and vascular density between each DR stage and control eyes and between individual DR stages ( Table 4 ). Compared with controls,

the FAZ area was signi cantly greater and vascular density (both VAD and VLD) was signi cantly lower in all stages of DR at the level of both the supercial and deep vascular networks. The mean supercial FAZ area trended upward and was 0.365 mm 2 in eyes with mild NPDR, 0.437 mm 2 in eyes with moderate-to- severe NPDR, and 0.508 mm 2 in eyes with PDR ( Fig 3A) without statistical difference between individual stages (all P > 0.05). A similar increasing trend was observed in mean deep FAZ area with worsening of DR stage. However, the difference in deep FAZ size was signicant only between mild NPDR and PDR stages (0.508 mm 2 vs. 766 mm 2 ; P ¼ 0.004). With regard to supercial vascular density (both VAD and VLD) values, there was no difference between individual DR stages (all P > 0.05). However, the deep vascular density (both VAD and VLD) decreased as the DR stage worsened ( Fig 3B). The mean deep VAD in the mild NPDR group was signi cantly greater than the moderate-to-severe NPDR group (58.09% vs. 55.83%; P ¼ 0.002)

237

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

Ophthalmology Volume 124, Number 2, February 2017

Table 1. Demographics of the Study Population

Signal Strength

No. of Participants No. of Eyes No. of Women (%) Age (yrs)* Visual Acuity Thickness (mm) y Index (Range) z

Central Macular

(logMAR)

Controls

27

34

14 (52)

62.3

11.2

0.03

0.047

259

20

75 7.9 (62 e92)

Diabetics

55

84

22 (40)

62.3

12.1

0.19

0.136

257

31

68 8.9 (50 e85)

Mild NPDR

22

32

7 (32)

61.2

13.6

0.15

0.111

249

29

69 9.1 (50 e85)

Moderate-to-severe NPDR

20

31

11 (55)

66.1

11.4

0.22

0.141

262

32

67 8.9 (50 e81)

PDR

13

21

4 (31)

58.5

9

0.21

0.154

264

29

69 8.7 (55 e85)

Total study population

82

118

36 (44)

62

11.7

0.15

0.136

260

27

70 9.1 (50 e92)

logMAR ¼ logarithm of the minimum angle of resolution; NPDR ¼ nonproliferative diabetic retinopathy; PDR ¼ proliferative diabetic retinopathy. Data are no., no. (%), or mean standard deviation. *Distribution was similar in controls and in each stage of diabetic retinopathy ( P ¼ 0.17, analysis of variance). y Similar in controls and in each stage of diabetic retinopathy (P ¼ 0.20, analysis of variance). z Statistically different between all groups ( P ¼ 0.001, analysis of variance).

and the PDR group (58.09% vs. 55.68%; P ¼ 0.004). The moderate- to-severe NPDR group deep VAD values were not statistically different from the PDR group ( Table 4; P ¼ 0.996). Similarly, mean deep VLD was signi cantly greater in mild NPDR eyes compared with moderate-to-severe NPDR eyes (22.12 mm 1 vs. 20.81 mm 1 ; P ¼ 0.015) and PDR eyes (22.12 mm 1 vs. 20.34 mm 1 ; P ¼ 0.002). No difference was found in deep VLD between moderate-to-severe NPDR eyes and PDR eyes ( P ¼ 0.761). Finally, Spearman s correlation test ( r) was used to study the association of FAZ area and vascular density with logMAR visual acuity in all study eyes ( Fig 4). The FAZ area had a weak positive correlation with logMAR visual acuity at the level of the super cial network ( r ¼ 0.29; P < 0.01) and a moderate positive correlation at the level of the deep network ( r ¼ 0.48; P < 0.001). The vascular density (both VAD and VLD) had a moderate negative correlation with logMAR visual acuity at the level of both the supercial (VAD, r ¼ 0.52; VLD, r ¼ 0.54; P < 0.001) and deep (VAD, r ¼ 0.50; VLD, r ¼ 0.50; P < 0.001) vascular networks.

Discussion

The current study used a commercially available OCTA system and integrated automated algorithms to examine macular vascular density and FAZ area in eyes with varying levels of DR severity compared with healthy controls. The FAZ area at the levels of the super cial and deep vascular networks of eyes with DR were found to be larger compared with those of age-matched healthy controls. As would be expected, control eyes had higher vascular den- sity in both networks compared with the diabetic eyes.

Furthermore, eyes with worse DR stage showed larger FAZ area and lower vascular density values. This study also showed that angiographic parameters, namely FAZ area,

VAD, and VLD, correlated with visual acuity. Baseline characteristics were similar between participants with DR and controls. Because prior studies have shown that vascular density as measured with OCTA tends to decrease

healthy control participants were age matched

with age,

to the diabetic cohort. Eyes with DME were excluded to ensure correlations of vision with vascular density and FAZ area were not confounded by the presence of macular edema. Additionally, in eyes with increased central macular thick- ness, the disrupted retinal anatomic features may lead to segmentation errors 21 and to false vascular density, FAZ measurement, or both. Indeed, central macular thickness values were statistically equivalent across all DR stages and healthy eyes. However, it should be noted that SSI was statistically higher in control eyes when compared with diabetic eyes. This difference was expected because SSI is a measure of tissue re ectance, which is more likely to be reduced in diabetic eyes because they are more prone to cataract development and other media opacities. We believe that this had a bigger effect on VAD

measurements rather than on VLD measurements. Alterations in the FAZ in DR have been well characterized using FA. 5,6,22 Sim et al 5 showed that FAZ size increases as DMI severity progresses and is most profound in PDR. When using FA, however, only the super cial vascular network is visualized and it is not possible to characterize FAZ

changes in the deep network.

With OCTA, we are now

18,20

10

Table 2. Prior Treatments in 84 Diabetic Eyes

 

Mild Nonproliferative

Moderate-to-Severe Nonproliferative Diabetic Retinopathy

Proliferative Diabetic

All Diabetic

Treatment

Diabetic Retinopathy

Retinopathy

Eyes

Treatment naïve Anti-VEGF injections only Focal/PRP laser only Anti-VEGF injections and focal/PRP laser

32 (100)

17 (55)

3 (14)

52 (62)

0 (0)

3 (10)

0 (0)

3 (4)

0 (0)

5 (16)

18 (86)

23 (27)

0 (0)

6 (19)

0 (0)

6 (7)

PRP ¼ panretinal photocoagulation; VEGF ¼ vascular endothelial growth factor. Data are no. (%).

238

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

Samara et al Quantication of DMI with OCTA

Samara et al Quanti fi cation of DMI with OCTA Figure 2. Foveal avascular zone (FAZ),

Figure 2. Foveal avascular zone (FAZ), vessel area density (VAD), and vessel length density (VLD) measurement using optical coherence tomography angiography (OCTA) in a healthy eye and in eyes with different stages of diabetic retinopathy (DR). The 3 3-mm OCTA scans show the supercial network in a healthy eye and in different stages of DR (mild nonproliferative DR [NPDR], moderate-to-severe NPDR, and proliferative DR [PDR]). First column, Supercial FAZ area has been recorded under each scan. Second column, Vessel area density is measured in the supercial network from the binary image representing the original scan with values recorded underneath. Third column, vessel length density is measured in the supercial network from the skeletonized binary image with values recorded underneath. Fourth column, the 3 3-mm OCTA scans of the deep network in a healthy eye and in different stages of DR, with deep FAZ area recorded underneath. Under each image, ( fth column) deep VAD and (sixth column) VLD values are recorded.

able to examine the deep network, and FAZ measurements using OCTA in both the super cial and deep networks have been shown to be repeatable and reproducible biomarkers for DR. 12 As was found in the current study, several studies using OCTA have shown that FAZ area is larger at both the level of super cial and deep vascular

networks in eyes with DR compared with healthy eyes. 12,23 These FAZ changes seem to be more pronounced at the

level of the deep vascular network and in eyes with

Interestingly, one study showed that even in diabetic patients

without clinically detectable DR, FAZ area was enlarged compared with healthy eyes. 25 In the current study, the super cial FAZ area was larger in diabetics compared with controls. However, subgroup analysis did not show any signi cant differences in super- cial FAZ area between DR stages. Previous FA studies have shown that super cial FAZ area enlarges with wors- ening of DR stage. This difference from our ndings may be

PDR. 23,24

the result of the patient population and possible inclusion of eyes with poorer xation and worse DMI in the FA studies, because imaging by FA does not require steady xation. In contrast, FAZ area was signi cantly enlarged in the deep network compared with controls, and post hoc analysis did show a signi cant difference between PDR and mild NPDR. This is in line with prior observations showing more pronounced changes at the level of the deep vascular

network. 23

Further studies exploring the relative sensitivity

of different vascular parameters in the super cial and deep

networks would be of potential interest in this regard. Previous studies also have correlated FAZ area with visual function using various imaging methods in patients with DR. One study using FA found a negative correlation between FAZ area and best-corrected visual acuity after adjusting for the presence of DME. 5 Recently, Freiberg

et al

demonstrated that the maximum FAZ diameter

measured by OCTA in both the super cial and deep

23

239

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

240

For personal

Ophthalmology Volume 124, Number 2, February 2017

Downloaded

use only. No

from

other

ClinicalKey.com

uses without permission.

at ClinicalKey

Copyright

Global ©2017.

Guest Users

Elsevier

March

Inc. 13,

All 2017.

rights reserved.

Table 3.

Comparison of Foveal Avascular Zone and Vascular Density Measurements

at the Level

of

the

nopathy and Healthy Controls

Super cial

and

the

Deep Vascular Networks

in

Eyes with Diabetic

Reti-

Mean Foveal Avascular Zone Area (Range), mm 2

Supercial

Deep

Mean Vascular Density (Range)

Supercial

Vessel Area Density (%)

Vessel Length Density (mm 1 )

Deep

Vessel Area Density (%)

Vessel Length Density (mm 1 )

Controls

Diabetics

0.275 0.1070 (0.035e0.505) 0.372 0.1129 (0.113e0.613) 55.09 2.584 (47.74e 59.13) 21.55 1.705 (17.15e 24.08) 61.32 1.939 (56.55e 64.48) 24.83 1.262 (22.25 e26.76)

0.427 0.2198* (0.101e1.245) 0.616 0.2715* (0.196e1.573) 49.44 3.837* (40.14e 57.19) 17.68 2.312 y (11.69e 23.06) 56.65 2.832 y (49.82e 62.29) 21.19 1.976 y (16.68 e25.39)

Mild

0.365 0.1301 (0.154e0.774)

0.508 0.1853 (0.196e0.887)

49.69 3.955 (42.48e 57.19)

17.96 2.488 (13.46e 23.06)

58.09 2.715 (53.24e 62.12)

22.12 1.918 (18.17 e25.39)

NPDR

Moderate-

0.437 0.2030 (0.131e1.040)

0.625 0.2514 (0.272e1.344)

49.09 3.907 (40.14e 55.29)

17.54 2.162 (11.69e 21.08)

55.83 2.990 (49.82e 62.29)

20.81 1.885 (16.68 e23.95)

to-

severe

NPDR

PDR

0.508 0.3149 (0.101e1.245)

0.766 0.3402 (0.338e1.573)

49.55 3.693 (44.24e 55.51)

NPDR ¼ nonproliferative diabetic retinopathy; PDR ¼ proliferative diabetic retinopathy. Data are mean standard deviation unless otherwise indicated. *All P < 0.001 compared with controls, ManneWhitney U test. y P < 0.001 compared with controls, unpaired t test.

17.45 2.318 (13.77e 20.89)

55.68 1.763 (52.66e 58.42)

20.34 1.677 (17.76 e23.98)

Table 4. Pairwise Comparisons of Foveal Avascular Zone Area and Macular Vascular Density between Diabetic Subgroups and Healthy Controls

Foveal Avascular Zone Area

Vascular Density

Supercial

Deep

 

Super cial*

Deep *

Vessel Area Density*

Vessel Length Density y

Vessel Area Density y

Vessel Length Density y

Controls vs. mild NPDR Controls vs. moderate-to-severe NPDR Controls vs. PDR Mild NPDR vs. moderate-to-severe NPDR Mild NPDR vs. PDR Moderate-to-severe NPDR vs. PDR

0.013

0.002

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

<0.001

0.204

0.088

0.619

0.865

0.002

0.015

0.164

0.004

0.850

0.837

0.004

0.002

0.803

0.184

0.799

0.999

0.996

0.761

NPDR ¼ nonproliferative diabetic retinopathy; PDR ¼ proliferative diabetic retinopathy. All data are P values. *P < 0.05 threshold for signicance, nonparametric Kruskal-Wallis test.

y P <

0.05 threshold

for

signi cance,

parametric analysis

of variance. Statistically

signi cant

differences

are highlighted in boldface.

Samara et al Quantication of DMI with OCTA

Samara et al Quanti fi cation of DMI with OCTA Figure 3. Boxplot graphs demonstrating (

Figure 3. Boxplot graphs demonstrating (A) the foveal avascular zone (FAZ) area, (B) vessel area density, and (C) vessel length density in the supercial and deep retinal vascular networks in healthy eyes and in eyes with different stages of diabetic retinopathy (DR): mild nonproliferative DR (NPDR), moderate-to-severe NPDR, and proliferative DR (PDR).

vascular networks correlated positively with best-corrected visual acuity. Our results showed a correlation between FAZ area and visual acuity, where eyes with worse vision had larger FAZ areas. It should be noted, however, that there

are interindividual differences in FAZ area and morphologic features, 13,14 which may limit its use as a tool for predicting visual acuity in DR. This also holds true in other bilateral conditions in which the FAZ area of the other eye cannot be used as a control. Nevertheless, longitudinal monitoring of FAZ area in these patients may be useful in tracking disease progression and predicting visual function. Fluorescein angiography has been the gold standard to detect capillary dropout and is the basis for the current

grading of DMI.

invasive, time consuming, and poor at identifying changes in the deep retinal vasculature. Most recently, OCTA has been used to quantify the macular vascular density in both

networks. Using total vessel length as a measure of

7

However, it remains participative,

vascular density, Agemy et al 15 also showed that the vascular density is lower in diabetic eyes when compared with healthy eyes in both 3 3-mm and 6 6-mm scans. Similar to our study, which used only a 3 3-mm scan area, these authors found a trend toward decreasing vascular density in both super cial and deep networks as DR worsened from mild NPDR to PDR. We further explored the correlation of vascular density with visual acuity and found a positive correlation at the level of both the super cial and deep vascular networks (Fig 4 ). This correlation held true for both VAD and VLD. Of note, vascular density measurements (both VAD and VLD) were based on the entire 3 3-mm scan density excluding the FAZ area, which ensured that these correla- tions are independent of FAZ area. Recent work demon- strated that the absence of perfusion in the deep capillary plexus in addition to FAZ enlargement seen on OCTA may play an important role in photoreceptor compromise and

241

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

Ophthalmology Volume 124, Number 2, February 2017

Ophthalmology Volume 124, Number 2, February 2017 Figure 4. Scatterplots showing the correlation between the logarithm

Figure 4. Scatterplots showing the correlation between the logarithm of the minimum angle of resolution (logMAR) visual acuity (best-corrected visual acuity) and macular vascular density in addition to foveal avascular zone (FAZ) area using Spearman correlation ( r) in all eyes. A signicant positive correlation between logMAR visual acuity and FAZ area is present in both the ( A) supercial and (B) deep networks. In the supercial network, a signicant negative correlation is present between logMAR visual acuity and both ( C) VAD and (D) VLD. Similarly, in the deep network, a signicant negative correlation is present between logMAR visual acuity and both ( E ) VAD and (F) VLD. NPDR ¼ mild nonproliferative diabetic retinopathy; PDR ¼ proliferative diabetic retinopathy.

242

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

Samara et al Quantication of DMI with OCTA

outer retinal changes seen on SD OCT of patients with diabetic retinopathy without DME. 26,27 More importantly, the choroidal ow seemed unaffected in these areas, which

points toward deep capillary non ow as a possible contributor. 27 Although it is well established that most of the metabolic supply to the outer retina is provided by diffusion from the choroidal circulation, 28 a recent experimental study showed that in the setting of hypoxia, the inner retinal vascular contribution to the metabolic demands of the outer retina becomes more pronounced. 29 This could be explained by failure of the autoregulatory mechanism of the choroidal vasculature in the presence of hypoxia. 29 Interestingly, the current study con rms that decreasing vascular density and enlargement of FAZ of both super cial and deep networks are associated with worse visual acuity. We also have noted that deep FAZ area correlation with visual acuity was stronger when compared with super cial FAZ area ( r ¼ 0.48 vs. r ¼ 0.29, respectively). However, further studies are needed to correlate OCTA changes in both networks with structural outer retinal changes seen on SD OCT. In addition, whether these OCTA ndings could predict visual function independent of structural SD OCT changes is yet to be elucidated. We recognize several limitations to our study. First, despite attempting to minimize the effect of artifacts by excluding poor-quality scans, image quality could still affect vascular density measurements. In the deep network, projection artifacts have been shown to increase measured ow in the deep vascular network articially. 30 With the development of artifact removal algorithms, the measurements in the deep network might more precisely

represent the actual ow in deep network. 30

variations in SSI between diabetics and controls may have impacted vascular density measurements. New algorithms are currently being developed to decrease the effect of variation in tissue reectance on vascular density measurements. 31 Third, eyes with poorer vision were more likely to be excluded because of inadequate scan quality resulting from poor xation during OCTA acquisition, which explains the narrow range of visual acuity in this study (20/20e 20/80). In addition, the visual acuity was measured based on current spectacle correction with pinhole rather than Early Treatment Diabetic Retinopathy Study protocol refraction. Fourth, the quantitative OCTA ndings from the current study may not be generalizable to other OCTA systems, which may use different scanning protocols and image processing methods. Fifth, because of the retrospective nature of the study, the participants were evaluated at different points in time after the diagnosis of DR. Additionally, prior therapeutic interventions (e.g., antie vascular endothelial growth factor, laser, or both) may have affected the study outcomes in some previously treated patients. Finally, only 3 3-mm scans were used to quantify vascular density. Although this choice limits the macular area scanned, it ensures quantication of vascular density from highest quality scans because there is a tradeoff between scan size and image resolution using the current device settings. Nevertheless, our study has several advantages, including the quantitative

Second,

analysis of angiographic indices using OCTA and the use of automated algorithms for measurement of vascular density and FAZ area, which diminishes the interobserver variability seen with manual measurement methods. In conclusion, the current study demonstrates that vascular density and FAZ area can be quanti ed in DR eyes in a rapid, automated, and noninvasive manner with OCTA. Foveal avascular zone area and vascular density as measures of DMI correlate with visual acuity in patients with DR without DME. Quantitative OCTA metrics may prove increasingly useful as image acquisition and processing improves. The impact of these novel vascular biomarkers in the clinical setting remains unclear, but there is potential for metrics like vascular density and FAZ area to supplement and even replace more well-established, participative markers of ocular diabetic disease activity. As OCTA technology evolves, inclusion of this imaging tool in future prospective studies will be essential in determining its potential impact on clinical care.

References

1.

1310.

2.

3.

2069.

4.

5.

6.

7.

8.

9.

10.

.

11.

243

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

Ophthalmology Volume 124, Number 2, February 2017

Footnotes and Financial Disclosures

Originally received: May 28, 2016. Final revision: September 17, 2016. Accepted: October 10, 2016. Available online: November 22, 2016.

Retina Service, Wills Eye Hospital, Mid Atlantic Retina, Thomas Jefferson University, Philadelphia, Pennsylvania.

Manuscript no. 2016-1140.

Financial Disclosure(s):

The author(s) have made the following disclosure(s): M.A.K.: Financial support e Allergan (Dublin, Ireland)

J.I.M.: Consultant e Genentech (San Francisco, CA); Lecturer e Regen- eron (Tarrytown, NY)

J.H.: Consultant e Optovue, Inc (Fremont, CA); UCB Pharma (Slough, UK); Financial support e Santen (Osaka, Japan); Ophthotech (New York, NY); Genentech (San Francisco, CA)

A.C.H.: Consultant, Scientic advisor, and Lecturer e Optovue, Inc, Fre- mont, CA

Supported in part by the Philadelphia Retina Endowment Fund, Philadel- phia, Pennsylvania (A.S.).

244

Author Contributions:

Conception and design: Samara, Shahlaee, Chiang, Maguire, Hsu, Ho

Analysis and interpretation: Samara, Shahlaee, Adam, Hsu, Ho

Data collection: Samara, Shahlaee, Adam, Khan, Hsu, Ho

Obtained funding: none

Overall responsibility: Samara, Shahlaee, Adam, Khan, Chiang, Maguire, Hsu, Ho

Abbreviations and Acronyms:

DME ¼ diabetic macular edema; DMI ¼ diabetic macular ischemia; DR ¼ diabetic retinopathy; FA ¼ uorescein angiography; FAZ ¼ foveal avascular zone; logMAR ¼ logarithm of the minimum angle of resolution; NPDR ¼ nonproliferative diabetic retinopathy; OCTA ¼ optical coherence tomography angiography; PDR ¼ proliferative diabetic retinopathy; SD OCT ¼ spectral-domain optical coherence tomography; SSI ¼ signal strength index; VAD ¼ vessel area density; VLD ¼ vessel length density.

Correspondence:

Allen C. Ho, MD, Retina Service, Wills Eye Hospital, 840 Walnut Street, Suite 1020, Philadelphia, PA 19107. E-mail: acho@midatlanticretina.com.

Downloaded from ClinicalKey.com at ClinicalKey Global Guest Users March 13, 2017. For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.

Оценить