[Skip to Content]
[Skip to Content Landing]
Figure 1.
Vitreous Hyperreflective Foci (VHRF) Algorithm
Vitreous Hyperreflective Foci (VHRF) Algorithm

A, An optical coherence tomographic (OCT) B-scan of a patient with diabetic macular edema with select VHRF circled. B, Initial 3-dimensional (3-D) rendering of contrast-enhanced OCT scans. C, The VHRF detection algorithm applied and cropped to focus on the vitreous. Spheres represent VHRF, and the retina appears as a solid 3-D object.

Figure 2.
Vitreous Hyperreflective Foci (VHRF) Scores for Comparison Groups
Vitreous Hyperreflective Foci (VHRF) Scores for Comparison Groups

DME indicates diabetic macular edema; DR, diabetic retinopathy. Horizontal bars represent mean values; error bars, SD.

aP < .05 compared with the DME group.

bP < .005 compared with the DME group.

Figure 3.
Vitreous Hyperreflective Foci Scores Subgrouped by Vitreoretinal Separation (VRS)
Vitreous Hyperreflective Foci Scores Subgrouped by Vitreoretinal Separation (VRS)

DME indicates diabetic macular edema; DR, diabetic retinopathy. Horizontal bars represent mean values; error bars, SD.

aDME with VRS vs control without VRS and vs diabetes without retinopathy or VRS; P < .0001 with Bonferroni analysis.

bDME with VRS vs control with VRS; P = .002 with Bonferroni analysis.

cDME with VRS vs DME without VRS; P = .001 with Bonferroni analysis.

Figure 4.
Vitreous Hyperreflective Foci (VHRF) Repeatability
Vitreous Hyperreflective Foci (VHRF) Repeatability

Variance in the VHRF score of successive optical coherence tomographic scans. Each participant had 2 scans per day on 3 days during a 2-week period.

Table.  
Patient Characteristics
Patient Characteristics
Video. Vitreous Hyperreflective Foci Detection Algorithm.

Three-dimensional rendering of optical coherence tomographic scan of the macula in a patient with diabetic macular edema. Rendering has been cropped to include the vitreous and retina. The algorithm has been applied and shows detected vitreous hyperreflective foci in green and retinal mapping in yellow.

1.
Wild  S, Roglic  G, Green  A, Sicree  R, King  H.  Global prevalence of diabetes: estimates for the year 2000 and projections for 2030.  Diabetes Care. 2004;27(5):1047-1053.PubMedArticle
2.
Ferris  FL  III, Patz  A.  Macular edema: a complication of diabetic retinopathy.  Surv Ophthalmol. 1984;28(suppl):452-461.PubMedArticle
3.
Antcliff  RJ, Marshall  J.  The pathogenesis of edema in diabetic maculopathy.  Semin Ophthalmol. 1999;14(4):223-232.PubMedArticle
4.
Antonetti  DA, Klein  R, Gardner  TW.  Diabetic retinopathy.  N Engl J Med. 2012;366(13):1227-1239.PubMedArticle
5.
Nguyen  QD, Shah  SM, Khwaja  AA,  et al; READ-2 Study Group.  Two-year outcomes of the Ranibizumab for Edema of the Macula in Diabetes (READ-2) study.  Ophthalmology. 2010;117(11):2146-2151.PubMedArticle
6.
Tang  J, Kern  TS.  Inflammation in diabetic retinopathy.  Prog Retin Eye Res. 2011;30(5):343-358.PubMedArticle
7.
Funatsu  H, Noma  H, Mimura  T, Eguchi  S, Hori  S.  Association of vitreous inflammatory factors with diabetic macular edema.  Ophthalmology. 2009;116(1):73-79.PubMedArticle
8.
Cantón  A, Martinez-Cáceres  EM, Hernández  C, Espejo  C, García-Arumí  J, Simó  R.  CD4-CD8 and CD28 expression in T cells infiltrating the vitreous fluid in patients with proliferative diabetic retinopathy: a flow cytometric analysis.  Arch Ophthalmol. 2004;122(5):743-749.PubMedArticle
9.
Urbančič  M, Kloboves Prevodnik  V, Petrovič  D, Globočnik Petrovič  M.  A flow cytometric analysis of vitreous inflammatory cells in patients with proliferative diabetic retinopathy.  Biomed Res Int. 2013;2013:251528.Article
10.
Pakzad-Vaezi  K, Or  C, Yeh  S, Forooghian  F.  Optical coherence tomography in the diagnosis and management of uveitis.  Can J Ophthalmol. 2014;49(1):18-29.PubMedArticle
11.
Saito  M, Barbazetto  IA, Spaide  RF.  Intravitreal cellular infiltrate imaged as punctate spots by spectral-domain optical coherence tomography in eyes with posterior segment inflammatory disease.  Retina. 2013;33(3):559-565.PubMedArticle
12.
Gallagher  MJ, Yilmaz  T, Cervantes-Castañeda  RA, Foster  CS.  The characteristic features of optical coherence tomography in posterior uveitis.  Br J Ophthalmol. 2007;91(12):1680-1685.PubMedArticle
13.
Agarwal  A, Ashokkumar  D, Jacob  S, Agarwal  A, Saravanan  Y.  High-speed optical coherence tomography for imaging anterior chamber inflammatory reaction in uveitis: clinical correlation and grading.  Am J Ophthalmol. 2009;147(3):413-416.e3.PubMedArticle
14.
Igbre  AO, Rico  MC, Garg  SJ.  High-speed optical coherence tomography as a reliable adjuvant tool to grade ocular anterior chamber inflammation.  Retina. 2014;34(3):504-508.PubMedArticle
15.
Keane  PA, Karampelas  M, Sim  DA,  et al.  Objective measurement of vitreous inflammation using optical coherence tomography.  Ophthalmology. 2014;121(9):1706-1714.PubMedArticle
16.
Knott  EJ, Sheets  KG, Zhou  Y, Gordon  WC, Bazan  NG.  Spatial correlation of mouse photoreceptor-RPE thickness between SD-OCT and histology.  Exp Eye Res. 2011;92(2):155-160.PubMedArticle
17.
National Institutes of Health.  ImageJ: image processing and analysis in Java. http://imagej.nih.gov/ij/. Accessed September 16, 2015.
Views 789
Citations 0
Original Investigation
January 2016

Algorithm for the Measure of Vitreous Hyperreflective Foci in Optical Coherence Tomographic Scans of Patients With Diabetic Macular Edema

Author Affiliations
  • 1Beaumont Eye Institute, Royal Oak, Michigan
  • 2Department of Ophthalmology and Visual Sciences, Kellogg Eye Center, University of Michigan, Ann Arbor
JAMA Ophthalmol. 2016;134(1):15-20. doi:10.1001/jamaophthalmol.2015.3949
Abstract

Importance  Developing a noninvasive measure of diabetic retinopathy disease progression may provide physicians with information needed for patient-specific intervention.

Objective  To develop an algorithm to measure vitreous hyperreflective foci (VHRF) from standard, 3-dimensional optical coherence tomographic (OCT) images in an unbiased manner.

Design, Setting, and Participants  We retrospectively analyzed OCT scans from 97 patients who were evaluated at the Kellogg Eye Center, University of Michigan. Patients with diabetes mellitus without signs of retinopathy (n = 29) and patients with diabetic macular edema (DME) (n = 31) were compared with healthy control participants (n = 37). The algorithm was used to determine whether the VHRF score is associated with DME and may serve as a noninvasive measure of inflammation. The study was conducted from November 14, 2011, to August 5, 2015. Data analysis was performed from May 15, 2014, to August 13, 2015.

Main Outcomes and Measures  An algorithm was developed to enhance the vitreous imaging from OCT to allow automated quantification of VHRF and calculation of a VHRF score. This score was compared between the healthy control, diabetes without retinopathy, and DME groups.

Results  In the 97 scans evaluated, VHRF scores, reported as mean (SD), were increased in patients with DME by 2.95-fold (5.60 [8.65]) compared with healthy controls (1.90 [3.42]; 95% CI, 0.75-7.45; P = .012) and by 6.83-fold compared with patients with diabetes without retinopathy (0.82 [1.26]; 95% CI, 1.46-8.82; P = .005).

Conclusions and Relevance  Scores obtained using the VHRF algorithm may be obtained from OCT images that include the vitreous and could provide a rapid, noninvasive clinical correlate for ocular inflammation. Higher VHRF scores in patients with DME compared with controls and diabetic patients without retinopathy warrant further population-based and longitudinal studies to help determine the value of the VHRF score in selecting therapeutic intervention.

Introduction

The incidence of diabetes mellitus is rising, with the total number of people affected worldwide projected to increase to 366 million by 2030.1 The most common cause of vision impairment in patients with diabetic retinopathy is macular edema.2 Macular edema is believed to result from a loss of the normal blood-retinal barrier with an increase in vascular permeability. It may also include impaired removal of proteins and fluid, with subsequent accumulation of fluid and cystoid formation.3,4 Vascular endothelial growth factor contributes to this process, and therapies targeting vascular endothelial growth factor have proved effective in improving visual acuity.5

In addition to vascular endothelial growth factor, mounting evidence suggests that neuroinflammation contributes substantially to the development and progression of diabetic retinopathy.6 Inflammatory mediators found in the vitreous fluid are associated with the severity of diabetic macular edema (DME).7 Evidence8,9 suggests that inflammatory cells are present in patients with diabetic retinopathy requiring vitrectomy and in those with vitreous hemorrhage. However, these findings were limited to advanced stages of diabetic retinopathy with confounding factors that necessitated the vitrectomy procedure. The contribution of neuroinflammation to the induction and progression of diabetic retinopathy remains an area of active investigation.

Optical coherence tomographic (OCT) scans are used in the diagnosis of ocular disease. Some investigators10,11 have noted the presence of hyperreflective foci in OCT scans of the vitreous of patients with inflammatory conditions, including uveitis. These foci have been proposed to represent inflammatory cells. Saito et al11 characterized hyperreflective foci as dots that were larger and denser than the “usual background speckle” and noted increased spot density closer to areas of retinitis. In addition, the authors noted a decreased number of spots as vitritis resolved with treatment. Gallagher et al12 similarly described OCT findings, characterized vitreous spots in patients with uveitis, and suggested that the spots or foci represented inflammatory cells migrating into the vitreous. Furthermore, other authors13,14 have proposed using OCT to analyze anterior chamber inflammation by quantifying these hyperreflective spots and have compared them with clinical grades of inflammation. Keane et al15 quantified the total OCT vitreous signal and compared it with clinical markers of inflammation. However, to our knowledge, no method has been developed to quantify the vitreous hyperreflective foci (VHRF), which may represent a more specific and accurate method to assess infiltrating cells.

To objectively characterize these VHRF, we have created an algorithm to analyze and quantify OCT scans, thereby providing a VHRF score. This algorithm has the benefits of analyzing the total volume of a scan in 3 dimensions (3-D) and controlling for variability in the vitreous volume scanned by use of a spots-per-volume metric. This algorithm was applied to OCT scans from patients with diabetes to determine whether the VHRF score may serve as a noninvasive measure of diabetic retinopathy disease progression and potential ocular inflammation. This algorithm was used to compare the incidence of VHRF in healthy controls, diabetic patients without retinopathy, and patients with DME. We observed increased VHRF scores in patients with DME compared with both healthy controls and diabetic patients without retinopathy. Collectively, this study demonstrates an algorithm to analyze OCT scans and rapidly quantify VHRF. The VHRF scores increased with DME, and quantification of a VHRF score may act as a noninvasive measure of the progression of diabetic retinopathy.

Box Section Ref ID

At a Glance

  • Vitreous hyperreflective foci (VHRF) observed in optical coherence tomographic (OCT) scans have been associated with inflammation. This article describes an algorithm to quantify VHRF from OCT.

  • Vitreous hyperreflective foci increase in patients with diabetic macular edema compared with patients with diabetes but no diabetic macular edema or patients without diabetes.

  • Longitudinal studies may help determine the value of a VHRF score in selecting therapeutic intervention for patients with diabetes.

Methods
Patients

We retrospectively analyzed the OCT scans from 97 patients who were evaluated at the Kellogg Eye Center, University of Michigan. Patient characteristics, including demographics, duration of disease, retina volume, and hemoglobin A1c level closest to the date of the scan, were recorded. The OCT scans had been obtained from 37 healthy control individuals, 29 diabetic patients without retinopathy, and 31 patients with DME. The OCT scans were obtained as dense, 20 × 20°, high-speed, automatic, real-time spectral-domain images (Spectralis HRA+OCT; Heidelberg Engineering). Optical coherence tomographic cube scans (512 A-scans in each B-scan, and 3.87-μm axial resolution, automatic real-time noise reduction of 16 scans per line over the macula with 97 sections) were obtained using the standard OCT acquisition window. The scans were deidentified before being exported from the OCT machines, and the image analysis protocol described below was conducted for each scan. Each study was approved by the University of Michigan institutional review board. Participants had provided written informed consent. The investigation was conducted from November 14, 2011, to August 5, 2015; data analysis was performed from May 15, 2014, to August 13, 2015.

VHRF Algorithm

Complete algorithm details are available in the eAppendix in the Supplement. Each OCT scan was imported into ImageJ software16 as a raw .vol file via the Open Hyex Raw plugin.17 A median filter was then applied to all 97 sections to reduce signal noise, and the file was converted into individual 16-bit .tif format images. These images were imported into IMARIS software (Imaris X64; Bitplane Scientific Software) and reconstructed in 3-D (Figure 1B). The 3-D renderings were then cropped to include the area of interest of vitreous and retina, removing the subretinal tissues from analysis. Next, the retinal volume was automatically mapped by our algorithm and subtracted from the image analysis field, leaving only the vitreous for spot detection. The normalize layers function, which sets the contrast of each section equal to the mean contrast for the 3-D rendering, was used to automatically adjust layers with aberrantly high background signal that may affect analysis. We then applied an algorithm (eAppendix in the Supplement) to identify and quantify VHRF (Figure 1C). Our algorithm accounted for VHRF size (estimated diameter, 4.0 µm) and the IMARIS proprietary quality thresholds (quality >30), which selectively identify spots based on signal intensity with respect to the background, as seen in the 3-D–rendered algorithm (Video). The same algorithm was applied to all scans. The volume of the vitreous and retina was then calculated.

We calculated the ratio of spots per volume of vitreous by using the following formula:

(Total Spots/Total Vitreous Volume Imaged) × 105.We defined the resulting value as the VHRF score and designated it as our primary outcome. Owing to the limitations of OCT, including space in the z-axis between sections and limited resolution in the horizontal axis, “volume” is in fact rendered volume in IMARIS and should not be interpreted as the true value of vitreous volume analyzed.

Repeatability

We performed 2 studies to validate the repeatability of this algorithm in the identification of VHRF, one comprising 5 individuals (4 healthy controls and 1 diabetic patient without retinopathy) who received 2 successive OCT scans within a session and completed 3 sessions within a 2-week period. The second study included 12 patients with DME who received 2 successive OCT scans within a single session. In each session, OCT scans were conducted with acquisition variables identical to those of the main study and were performed by the same technician (T.S.). Our VHRF identification algorithm was subsequently applied to all scans.

Statistical Analysis

Statistical analysis was done using PRISM, version 6 (GraphPad Software Inc). One-way analysis of variance with posttest Bonferroni multiple comparison was used to compare means, and differences were considered statistically significant at P < .05. The Grubbs test (http://graphpad.com/quickcalcs/grubbs1/) for outliers was used in each group analysis.

A total of 97 OCT scans were analyzed from 37 healthy controls, 29 diabetic patients without retinopathy, and 31 patients with DME. Significant artifacts were identified in the OCT scans of 1 healthy control, 3 diabetic patients without retinopathy, and 1 patient with DME; these scans were excluded from analysis. After the algorithm was applied, 1 outlier in the VHRF score was identified in each of the control, diabetes without retinopathy, and DME groups through the use of a Grubbs test, and these scans were excluded from analysis.

Vitreoretinal separation (VRS) subgrouping was defined by OCT images according to grading by a retinal specialist (G.C.) as to whether the patients had full or partial posterior vitreous detachment or no posterior vitreous detachment and was completed in a masked fashion. After the algorithm was applied, 1 outlier in the VHRF score was identified in a control with no VRS, 1 in a control with VRS, and 1 in a patient with diabetes but without DME or VRS; these OCT scans were excluded from analysis.

Statistical analysis for the repeatability study was done using the variance components procedure of SAS, version 9.4 (SAS Institute Inc), which estimates the contribution of each effect to the variance of the VHRF score. The intraclass correlation coefficient (ICC) was then calculated for each source of variability. Nonparametric bootstrapping was used to calculate 95% CIs for the ICCs. The original sample of 12 participants with diabetes (each person with 2 VHRF measures on a single day) was resampled with replacement to obtain 1000 bootstrap samples, each with 12 patients. The unit of selection was the patient to preserve the nesting of measures. The ICCs were then calculated on each resample to obtain the distributions of the ICC statistics. The 2.5- and 97.5-percentiles were used to create 95% CIs. All analysis was performed with SAS, version 9.4).

Results

Application of the algorithm to 97 OCT scans from healthy controls and patients with diabetes, with or without DME-identified VHRF, led to the development of a VHRF score. The Table provides relevant patient data for the OCT scans used. Figure 1 depicts a typical OCT scan from a patient with DME. Hyperreflective foci are present in the vitreous of the patient (Figure 1A). Image stacks were imported into IMARIS software and cropped to illustrate VHRF (Figure 1B), followed by contrast enhancement and application of the spot detection algorithm that accounted for signal in relation to the background and contained a defined minimum and maximum spot size. The VHRF were readily detected and quantified in an unbiased manner using IMARIS (Figure 1C).

The algorithm was applied to controls, diabetic patients without retinopathy, and patients with DME. Diabetic macular edema was associated with a clear increase in the VHRF score. Bonferroni analysis yielded a 2.95-fold difference in VHRF score between the DME and control groups (mean [SD], 5.60 [8.65] vs 1.90 [3.42]; 95% CI, 0.75-7.45; P = .012) as well as a 6.83-fold difference between the DME and diabetes without retinopathy groups (0.82 [1.26]) (95% CI, 1.46-8.82; P = .005) (Figure 2). No significant difference was found between the control and diabetes without retinopathy groups. Patients were further subgrouped according to the presence of VRS as determined by a masked retinal specialist (G.C.). The groups included control without VRS [n = 19], control with VRS (n = 15), diabetes without retinopathy or VRS (n = 22), diabetes without retinopathy with VRS (n = 3), DME without VRS (n = 15), and DME with VRS (n = 15). Bonferroni analysis of these subgroups yielded statistical significance in the VHRF score between the control without VRS and DME with VRS groups (Figure 3) (P < .0001), between the control with VRS and the DME with VRS groups (P = .002), between the diabetes without retinopathy or VRS and the DME with VRS groups (P < .0001), and between the DME without VRS and the DME with VRS groups (P = .001). No significant difference was observed for VRS as an independent factor for VHRF score by posttest analysis of variance between the control and VRS group. However, the clear increase in VHRF score between the DME without VRS and DME with VRS groups suggests that VRS, at a minimum, acts with DME as a risk factor for a high VHRF score. Collectively, these results demonstrate increased VHRF scores in patients with DME compared with controls or patients with diabetes without retinopathy and suggest increased inflammation in individuals with DME. Furthermore, increased VHRF scores were associated with patients with DME who had VRS, which may reflect increased neuroinflammation in these patients but may also represent increased cell debris.

To determine whether patient age affected the VHRF score, regression analysis was performed. In a model of age predicting VHRF score, age is a predictor: for every 1-year increase in age, the VHRF score increases by 0.12 units (P = .0028). In a model including age and both diabetes groups, after adjustment for age, there was still an effect of DME on the VHRF score (overall P = .0364). Specifically, the DME group showed larger VHRF scores than the diabetes without retinopathy group (P = .0340) and the control group (P = .0165). The control group did not show a significant difference in VHRF score from the diabetes without retinopathy group (P = .8236). There was no interaction between group and age (P = .6654).

The difference in VHRF score was not the result of variations in measures. To determine repeatability, the variance of 2 successive OCT scans and the VHRF scores for each scan was calculated within a session. There were 4 controls and 1 patient with diabetes without diabetic retinopathy, for which 3 sessions were completed during a 2-week period (Figure 4). The variation between the 2 scans in each session accounted for 13% of the variation in the data (ICC, 0.13), and the variation within a patient over different sessions accounted for 29% of the variation in the data (ICC, 0.29). The variation in the group as a whole over different days was approaching 0, suggesting that the variation in scans from day to day was nearly undetectable compared with differences among individuals. We performed a separate repeatability study of patients with DME. There were 12 patients who had 2 successive OCT scans within a single session. The variation between the 2 scans in each session accounted for 38% of the variation in the data (ICC, 0.38).

Discussion

The institution of spectral-domain OCT in most ophthalmology clinics provides a direct, noninvasive means to assess retinal structure associated with disease processes. Previous studies1012,15 have observed VHRF associated with inflammation; in the present study, we developed an algorithm to quantify, in an unbiased manner, the VHRF and provide a score for use in patients with diabetic retinopathy.

We observed an increased VHRF score in spectral-domain OCT images from patients with DME compared with both healthy control individuals and diabetic patients without retinopathy. This finding suggests an increase in VHRF at some point in the progression of the disease to DME. If these VHRF represent inflammatory cells, additional questions are raised regarding the contribution of vitreous inflammatory cells in the pathogenesis and progression of DME that warrant further investigation.

The repeatability study demonstrated the fidelity of our algorithm as observed over successive OCT scans within the same day, as well as additional scans during a short period (2 weeks). The mean VHRF score variance in healthy controls of 0.13 between successive scans is minor compared with the mean VHRF values for our main study groups. A nontrivial amount of variance in the VHRF score was seen within participants in our healthy repeatability study between different days as well as in patients with DME who underwent successive scans on the same day. The vitreous is a fluid structure with likely nonhomogeneous distribution of VHRF, and our scan window is restricted to the posterior portion; therefore, some variation may be expected. Thus, a mean of successive VHRF scores in a session may be more accurate when used in a clinical setting. Still, the variation of the control group as a whole on different days was approaching 0, suggesting that the variation in scans from day to day had little effect on the VHRF score compared with individual differences. Together, these outcomes suggest acceptable reliability and repeatability of our algorithm for use as a clinical diagnostic tool.

Certain limitations exist within this proof-of-concept study, which we look to address in our continuing work through more expansive prospective studies. First, the nature of VHRF is not known. As described in the Introduction, VHRF are associated with inflammation and may represent inflammatory cells, but whether they also represent cell debris or other factors is unknown. The VHRF score may become an important biomarker, but it is unknown whether the foci precede DME, contribute a causative role, or resolve with treatment. Finally, the role of VRS and age as contributors to higher VHRF scores needs to be explored further.

Studies using larger populations and longitudinal treatment response analysis may determine whether the use of the VHRF score will provide a means of characterizing disease progression for patients with diabetic retinopathy. Furthermore, the application of the algorithm to existing OCT images may allow larger cross-sectional and longitudinal studies to identify independent risk factors for the VHRF score. When patient OCT images were subgrouped for VRS in the present study, differences between the DME and DME with VRS groups were observed, suggesting that VRS was associated with neuroinflammation and might help explain the development of epiretinal membranes in this setting. We suspect that VRS is an independent risk factor for a higher VHRF score. With larger groups and longitudinal studies, we hope to further elucidate the risk factors for a high VHRF score and determine the precise stages of disease progression that may be associated with increasing scores.

Conclusions

The present studies demonstrate the development and use of an algorithm to quickly and easily quantify VHRF in OCT scans. Even in these limited studies, the VHRF score demonstrated a correlation with the presence of DME and likely represents an inflammatory component of the disease process. Applying this algorithm to OCT imaging may help to quantify disease progression and could identify individual patients with an inflammatory process associated with diabetic retinopathy.

Back to top
Article Information

Corresponding Author: David A. Antonetti, PhD, Department of Ophthalmology and Visual Sciences, Room 7317, University of Michigan Kellogg Eye Center, 1000 Wall St, Ann Arbor, MI 48105 (dantonet@med.umich.edu).

Submitted for Publication: April 22, 2015; final revision received August 18, 2015; accepted August 24, 2015.

Published Online: October 22, 2015. doi:10.1001/jamaophthalmol.2015.3949.

Author Contributions: Drs Korot and Antonetti had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Korot, Comer, Antonetti.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Korot, Antonetti.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Korot.

Obtained funding: Comer, Antonetti.

Administrative, technical, or material support: Steffens.

Study supervision: Antonetti.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.

Funding/Support: This study was supported by grant EY012021 from the National Institutes of Health (NIH) for study design, analysis, conduct, and management (Dr Antonetti); the Jules and Doris Stein Professorship from Research to Prevent Blindness for study design, analysis, conduct, and management (Dr Antonetti); and core grants EY007003 (Core Center for Vision Research at the Kellogg Eye Center) and DK020572 (imaging core of the Michigan Diabetes Research and Training Center) from the NIH for data analysis and interpretation. The JDRF (formerly Juvenile Diabetes Research Foundation) provided funding and support for data collection and analysis.

Role of the Funder/Sponsor: The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Steve Lentz, PhD, assisted with image analysis and David C. Musch, PhD, and Leslie M. Niziol, MS, assisted with statistical analysis (University of Michigan). There was no financial compensation.

References
1.
Wild  S, Roglic  G, Green  A, Sicree  R, King  H.  Global prevalence of diabetes: estimates for the year 2000 and projections for 2030.  Diabetes Care. 2004;27(5):1047-1053.PubMedArticle
2.
Ferris  FL  III, Patz  A.  Macular edema: a complication of diabetic retinopathy.  Surv Ophthalmol. 1984;28(suppl):452-461.PubMedArticle
3.
Antcliff  RJ, Marshall  J.  The pathogenesis of edema in diabetic maculopathy.  Semin Ophthalmol. 1999;14(4):223-232.PubMedArticle
4.
Antonetti  DA, Klein  R, Gardner  TW.  Diabetic retinopathy.  N Engl J Med. 2012;366(13):1227-1239.PubMedArticle
5.
Nguyen  QD, Shah  SM, Khwaja  AA,  et al; READ-2 Study Group.  Two-year outcomes of the Ranibizumab for Edema of the Macula in Diabetes (READ-2) study.  Ophthalmology. 2010;117(11):2146-2151.PubMedArticle
6.
Tang  J, Kern  TS.  Inflammation in diabetic retinopathy.  Prog Retin Eye Res. 2011;30(5):343-358.PubMedArticle
7.
Funatsu  H, Noma  H, Mimura  T, Eguchi  S, Hori  S.  Association of vitreous inflammatory factors with diabetic macular edema.  Ophthalmology. 2009;116(1):73-79.PubMedArticle
8.
Cantón  A, Martinez-Cáceres  EM, Hernández  C, Espejo  C, García-Arumí  J, Simó  R.  CD4-CD8 and CD28 expression in T cells infiltrating the vitreous fluid in patients with proliferative diabetic retinopathy: a flow cytometric analysis.  Arch Ophthalmol. 2004;122(5):743-749.PubMedArticle
9.
Urbančič  M, Kloboves Prevodnik  V, Petrovič  D, Globočnik Petrovič  M.  A flow cytometric analysis of vitreous inflammatory cells in patients with proliferative diabetic retinopathy.  Biomed Res Int. 2013;2013:251528.Article
10.
Pakzad-Vaezi  K, Or  C, Yeh  S, Forooghian  F.  Optical coherence tomography in the diagnosis and management of uveitis.  Can J Ophthalmol. 2014;49(1):18-29.PubMedArticle
11.
Saito  M, Barbazetto  IA, Spaide  RF.  Intravitreal cellular infiltrate imaged as punctate spots by spectral-domain optical coherence tomography in eyes with posterior segment inflammatory disease.  Retina. 2013;33(3):559-565.PubMedArticle
12.
Gallagher  MJ, Yilmaz  T, Cervantes-Castañeda  RA, Foster  CS.  The characteristic features of optical coherence tomography in posterior uveitis.  Br J Ophthalmol. 2007;91(12):1680-1685.PubMedArticle
13.
Agarwal  A, Ashokkumar  D, Jacob  S, Agarwal  A, Saravanan  Y.  High-speed optical coherence tomography for imaging anterior chamber inflammatory reaction in uveitis: clinical correlation and grading.  Am J Ophthalmol. 2009;147(3):413-416.e3.PubMedArticle
14.
Igbre  AO, Rico  MC, Garg  SJ.  High-speed optical coherence tomography as a reliable adjuvant tool to grade ocular anterior chamber inflammation.  Retina. 2014;34(3):504-508.PubMedArticle
15.
Keane  PA, Karampelas  M, Sim  DA,  et al.  Objective measurement of vitreous inflammation using optical coherence tomography.  Ophthalmology. 2014;121(9):1706-1714.PubMedArticle
16.
Knott  EJ, Sheets  KG, Zhou  Y, Gordon  WC, Bazan  NG.  Spatial correlation of mouse photoreceptor-RPE thickness between SD-OCT and histology.  Exp Eye Res. 2011;92(2):155-160.PubMedArticle
17.
National Institutes of Health.  ImageJ: image processing and analysis in Java. http://imagej.nih.gov/ij/. Accessed September 16, 2015.
×