Free
Retina  |   March 2014
Detection and Analysis of Hard Exudates by Polarization-Sensitive Optical Coherence Tomography in Patients With Diabetic Maculopathy
Author Affiliations & Notes
  • Jan Lammer
    Department of Ophthalmology, Medical University of Vienna, Vienna, Austria
  • Matthias Bolz
    Department of Ophthalmology, Medical University of Vienna, Vienna, Austria
  • Bernhard Baumann
    Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria
  • Michael Pircher
    Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria
  • Bianca Gerendas
    Department of Ophthalmology, Medical University of Vienna, Vienna, Austria
  • Ferdinand Schlanitz
    Department of Ophthalmology, Medical University of Vienna, Vienna, Austria
  • Christoph K. Hitzenberger
    Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria
  • Ursula Schmidt-Erfurth
    Department of Ophthalmology, Medical University of Vienna, Vienna, Austria
  • Correspondence: Jan Lammer, Department of Ophthalmology, Medical University of Vienna, Vienna, Austria; jan.lammer@meduniwien.ac.at
Investigative Ophthalmology & Visual Science March 2014, Vol.55, 1564-1571. doi:10.1167/iovs.13-13539
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Jan Lammer, Matthias Bolz, Bernhard Baumann, Michael Pircher, Bianca Gerendas, Ferdinand Schlanitz, Christoph K. Hitzenberger, Ursula Schmidt-Erfurth; Detection and Analysis of Hard Exudates by Polarization-Sensitive Optical Coherence Tomography in Patients With Diabetic Maculopathy. Invest. Ophthalmol. Vis. Sci. 2014;55(3):1564-1571. doi: 10.1167/iovs.13-13539.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose.: To image and analyze hard exudates (HEs) and their precursors in patients with diabetic macular edema (DME) by using polarization-sensitive optical coherence tomography (PS-OCT).

Methods.: Twenty-two eyes of 16 patients with DME were imaged by using color fundus photography (CF) and PS-OCT. In PS-OCT, HEs were automatically detected by their distinct polarization-scrambling qualities. Color fundus images were manually graded for the presence of HEs by two masked graders and correlated with the corresponding PS-OCT HE maps: corresponding images were overlaid and an identical grid of 128 × 128 fields was used for correlation of detected HEs.

Results.: In all eyes, HEs were present owing to DME. Agreement of a pixel-to-pixel analysis of HEs in CF images was 0.72 (Cohen's κ) between graders and 0.44 between graders and automated detection by PS-OCT. Mean ± SD detection of HEs was significantly higher in PS-OCT than in manual grading (1180.5 ± 1009.8 fields versus 828.8 ± 695.0 fields; P = 0.02). The higher detection rate of PS-OCT was confirmed by a linear regression analysis with a slope of β = 1.18 (r = 0.81).

Conclusions.: PS-OCT enables not only two-dimensional imaging of the extent of HEs, as in CF, but also allows tissue-specific, three-dimensional imaging of HEs throughout retinal layers, based on their distinct polarization-scrambling characteristics. The higher detection rate in PS-OCT images indicates an increased sensitivity of PS-OCT imaging over conventional CF.

Introduction
Diabetic retinopathy (DR) and diabetic macular edema (DME), common complications of diabetes mellitus, are among the leading causes of visual impairment in the Western world and will dramatically increase in other regions such as Asia and South America. 1,2 The overall prevalence of DME in patients with diabetes is reported to be approximately 7% (6.8%–7.5%), with a five-fold increased risk for subjects with type 1 diabetes (duration ≥ 20 years) compared to those with type 2 diabetes (duration < 10 years). 3,4 The Wisconsin Epidemiologic Study of Diabetic Retinopathy has reported a 10-year incidence of DME of approximately 20% in individuals with type 1 diabetes and 14% to 25% in those with type 2 diabetes, and a 25-year incidence of 29% for DME and 17% for clinically significant DME in individuals with type 1 diabetes. 5,6  
Early detection of DR and DME is crucial for prevention of vision loss. Among other retinal changes, such as micro-aneurysms and intraretinal hemorrhages, hard exudates (HEs) are striking features of an impairment of the blood–retinal barrier, leading to retinal swelling and edema. 7 Extravasation of lipids, proteinaceous material, and inflammatory cells lead to deposits primarily located in the outer plexiform and outer nuclear layer of the neural retina. 8,9 Increasing numbers of HEs seem to be associated with an increased risk of visual loss. 7 Furthermore, eyes with severe HEs also have an increased risk of developing subretinal fibrosis, an additional complication of macular edema. 10  
Retinal imaging is widely used by ophthalmologists and primary care physicians to screen for epidemic eye diseases such as DR and DME. Color fundus photography (CF) was previously the gold standard. Optical coherence tomography (OCT), however, has recently entered the field of retinal imaging in clinical daily routine. While CF provides a high sensitivity for a wide range of diabetic retinal changes two-dimensionally, 11,12 OCT raster scanning offers important cross-sectional information about the retinal layers. 13,14  
Automatic detection of DR lesions, especially of HEs, is a difficult task in both CF and OCT imaging. Several attempts have been made to segment HEs in CF photographs with improving sensitivity and specificity. 1517 However, CF only provides two-dimensional information about the extent of HEs. Owing to their density and opacity, HEs are also distinguishable in OCT B-scans as hyperreflective lesions. 9,18 OCT can additionally provide three-dimensional information about HE extent and distribution throughout the retinal layers. However, conventional spectral-domain (SD) OCT imaging and quantification of HEs are limited by the lack of automated segmentation algorithms: owing to the intensity-based image generation in SD-OCT, differentiation of HEs from other hyperreflective structures within the retinal layers remains unfeasible. 
Polarization-sensitive OCT (PS-OCT) is a functional extension of OCT that enables high-resolution three-dimensional imaging of biological samples, based on their polarization properties, in addition to providing conventional OCT images based on the intensity of backscattered light. 19,20 In PS-OCT, the sample is illuminated with one or more well-defined polarization states, and the OCT signal is detected in a polarization-sensitive fashion. By analyzing the detected polarization states, PS-OCT can distinguish tissues by the polarization of backscattered light and thus provides additional, tissue-specific contrast. In the human eye, PS-OCT can differentiate between birefringent (e.g., sclera, retinal nerve fiber layer [RNFL]), polarization-preserving (e.g., photoreceptor layer), and depolarizing tissues (e.g., retinal pigment epithelium [RPE]). 2129 A comprehensive review of PS-OCT and its application to ocular imaging can be found in a recent review. 30 Hard exudates appear depolarizing in PS-OCT images in exudative diseases such as wet AMD, vein occlusion, and diabetic retinopathy with DME. This characteristic can be used for specific segmentation of HEs. 
In this article, we systematically analyzed this phenomenon and compared this new imaging modality to the existing gold standard, namely color fundus photography. The aim of the study was to assess the capability of PS-OCT to selectively image HEs on a proof-of-principle basis and to compare the results with the current gold standard. 
Methods
Patients and Inclusion
In this prospective, cross-sectional, observational study, patients with DME secondary to DR were enrolled at the Department of Ophthalmology, Medical University of Vienna, Austria. The study protocol adhered to the tenets of the Declaration of Helsinki and was approved by the institutional ethics committee. All participants signed an informed consent form after a detailed explanation of the study design, associated investigations for scientific purposes, and adjuvant imaging procedures. 
Inclusion criteria for the study were DR due to type 2 diabetes mellitus, with presence of HEs in the macula and perimacular regions. Patients with media opacities (cornea, lens, vitreous) or macular alterations due to other diseases were excluded from the study. 
Image Assessment
Standardized examination procedures were performed according to a study protocol: During a same-day examination, patients underwent a complete evaluation, including standardized best corrected Early Treatment Diabetic Retinopathy Study Research Group (ETDRS) visual acuity testing, slit-lamp examination, fundoscopy, 30° and 50° macular CF (FF450 Fundus Camera; Carl Zeiss Meditec, Inc., Dublin, CA), SD OCT (Spectralis HRA+OCT; Heidelberg Engineering, Inc., Heidelberg, Germany) and PS-OCT imaging. 
PS-OCT was conducted by using a prototype developed at the Center for Medical Physics and Biomedical Engineering, Medical University Vienna, Austria. A detailed description of the instrument can be found elsewhere. 31 In brief, the system was based on SD PS-OCT 21 and used a detection unit with two identical spectrometer units. The system operated at a center wavelength of 839 nm and provided an axial resolution of <5 μm in tissue. The imaging speed was 20,000 A-scans per second. PS-OCT data sets of 512 × 128 A-scans were acquired in 3.3 seconds and covered an area of 6.2 mm (x-axis) × 6.7 mm (y-axis) on the fundus. 
Degree of polarization uniformity (DOPU) images were computed to assess depolarization information. 32 DOPU is related to the degree of polarization (DOP) known from classical polarization optics, which, however, cannot be directly measured by a coherent imaging technique such as OCT. DOPU was calculated from the Stokes vectors, which are derived from the raw PS-OCT data and indicate the polarization states corresponding to the image pixels. However, unlike DOP calculations, for DOPU calculation the Stokes vector elements are averaged in a sliding evaluation window in order to provide a measure for the local variance of polarization states. 32 The value of DOPU ranges from close to 1 for uniform polarization (as in most retinal layers such as, for instance, the polarization-preserving photoreceptors) to lower values in depolarizing structures (such as the RPE or hard exudates). To segment depolarizing image features, DOPU images were binarized with respect to a user-defined threshold (e.g., DOPUthresh = 0.8). All depolarizing pixels had values of DOPU < DOPUthresh. The locations of such depolarizing pixels could then be overlapped on intensity images (Figs. 1B, 1C, 1D, 1G). 
Figure 1
 
Segmentation of hard exudates by PS-OCT. (A) Color fundus photography of a DME patient with HEs. The dotted box indicates the location of the PS-OCT cube. (B) Fundus projection image generated from the PS-OCT data set. The yellow line indicates the location of the B-scan images on the right. (C) Intensity B-scan image. (D) Overlay of depolarizing pixels (DOPU < 0.8, red) on the intensity image. (E) En face image showing the summation of all depolarizing (red) pixels in every A-scan. The summated pixels include both the RPE and HEs. (F) Depolarizing pixels (red) within the HE segmentation band (light grey). (G) Segmented HEs (red) overlaid on intensity image. (H) HE thickness map generated by summing pixels within the HE segmentation band.
Figure 1
 
Segmentation of hard exudates by PS-OCT. (A) Color fundus photography of a DME patient with HEs. The dotted box indicates the location of the PS-OCT cube. (B) Fundus projection image generated from the PS-OCT data set. The yellow line indicates the location of the B-scan images on the right. (C) Intensity B-scan image. (D) Overlay of depolarizing pixels (DOPU < 0.8, red) on the intensity image. (E) En face image showing the summation of all depolarizing (red) pixels in every A-scan. The summated pixels include both the RPE and HEs. (F) Depolarizing pixels (red) within the HE segmentation band (light grey). (G) Segmented HEs (red) overlaid on intensity image. (H) HE thickness map generated by summing pixels within the HE segmentation band.
In retinal PS-OCT images of DME patients, both HEs and the RPE appeared depolarizing. To segment solely the HEs, we developed a dedicated software tool (Hard Exudate Analysis Software Tool, HEAST), which first segmented both RPE and HEs, and subsequently isolated a band-shaped retinal portion including HEs. The basic concept of the segmentation algorithm relied on the ideas for segmentation of drusen and geographic atrophy described by Baumann et al. 31 The internal limiting membrane (ILM) was segmented from the gradient in the intensity image. Depolarizing structures (i.e., RPE and HEs) were segmented by thresholding DOPU images (red structures in Fig. 1D, en face map in Fig. 1E). As described by Baumann et al., 31 a smooth function was automatically adapted posteriorly to the RPE by using Savitzky-Golay filtering, approximating the position of Bruch's membrane (BM). The posterior border of the HE segmentation band (light grey in Fig. 1F) was then defined by advancing the BM segmentation line a certain number of pixels anteriorly from the BM in order to exclude the RPE band, inner/outer segments of photoreceptors, and external limiting membrane. For each data set, this value was set to 60 pixels by default but could be adjusted manually if segmentation errors occurred. Analogously, the anterior border of the HE segmentation band was defined by the ILM segmentation line. Finally, the depolarizing pixels within the segmentation band, which were attributed to the HEs, were summed for each A-scan in order to generate en face HE thickness maps that could be compared to fundus photographs (Fig. 1H). 
Image Analysis
To correlate the extent of HEs in both CF and polarization-sensitive HE maps, overlays of the two imaging types were generated. Corresponding images were imported into Adobe Photoshop (CS6 version 13.0; Adobe Systems, Inc., San Jose, CA). Using retinal vessels as landmarks, the dimensions of the CF images were adjusted to the size and orientation of the respective en face images of the PS-OCT and cropped for further use (Figs. 2A–2E). Two experienced and certified graders (JL, BG) from the Vienna Reading Center manually graded the cropped CF images independently and in a masked fashion: Areas of HEs (as described by the ETDRS 11 : small white or yellowish-white deposits with sharp margins, often slightly waxy or glistening appearance) in the CF image were marked and exported as black-and-white (b/w) images. 
Figure 2
 
Superposition of CF and PS-OCT images. (A) Color fundus photograph of a left eye. The yellow box indicates the area scanned by PS-OCT. (B) Using retinal vessels as landmarks, the PS-OCT raster scan is superposed on the CF image. (C) The corresponding CF area is then cropped and exported for manual and automated assessment of HEs. (D) Overlay of the manual grading (in black, by grader 1). (E) Overlay of the automated assessment by PS-OCT (false color thickness map).
Figure 2
 
Superposition of CF and PS-OCT images. (A) Color fundus photograph of a left eye. The yellow box indicates the area scanned by PS-OCT. (B) Using retinal vessels as landmarks, the PS-OCT raster scan is superposed on the CF image. (C) The corresponding CF area is then cropped and exported for manual and automated assessment of HEs. (D) Overlay of the manual grading (in black, by grader 1). (E) Overlay of the automated assessment by PS-OCT (false color thickness map).
For statistical analysis, image data were translated into data sheets: Using the free ImageJ plugin “Patch Detector Plus” (available in the public domain at http://microscopy.uni-graz.at/index.php?item=new1), a grid containing 128 × 128 fields was generated, based on the PS-OCT's resolution of 512 × 128 pixels, and marked areas in all graded b/w exports and corresponding polarization-sensitive HE maps were automatically assessed. This alignment allowed for a detailed comparative analysis of HEs in CF and depolarizing spots in PS-OCT maps in 16,384 fields of each data set. 
Statistical Analysis
Statistical analysis was performed with SPSS (SPSS for Windows, version 21; SPSS, Inc., Chicago, IL). For pairwise, interrater agreement, Cohen's κ 33 was calculated for results of grader 1 and 2 (both in CF), grader 1 (in CF) and the PS-OCT, and grader 2 (in CF) and the PS-OCT. Bland-Altman plots were generated for each pair. For all calculations, a maximum P value of 0.05 was considered as the level of significance. 
Results
Twenty-two eyes of 16 patients were analyzed in this study. Mean age ± SD was 61 ± 7.4 years, and 44% of patients were female. All eyes showed clinically detectable HEs due to DR at the posterior pole. 
From all 22 eyes 360,448 individual grid fields were assessed by each grader (grader 1, grader 2, PS-OCT), resulting in a total of 1,081,344 analyzed grid fields. For manual assessment of CF images, grader 1 (JL) graded a mean ± SD of 836.6 ± 700.3 fields per eye as “HE detected,” while grader 2 (BG) graded 820.9 ± 711.1 fields. For automated detection of depolarizing signals in PS-OCT, 1309.0 ± 1240.5 fields per eye were graded as “HE detected.” Means did not differ between grader 1 and grader 2 (P = 0.77) but did differ between PS-OCT and the graders (both P = 0.02). Correlation between PS-OCT detection and the mean of graders was good, with a regression line slope of β = 1.24 and a Pearson's r = 0.70 (P < 0.001) (Figs. 3A, 3B). Interrater reliability for the manual grading of HEs in the CF images of grader 1 and 2 was κ = 0.72. Agreement of results of the aligned PS-OCT images with the individual graders was κ = 0.44 for grader 1 and κ = 0.43 for grader 2. For agreement matrix, see the Table
Figure 3
 
Plots of detected fields containing hard exudates. (A) Agreement of manual grading between grader 1 and grader 2 was good with a mean difference of 15.7 fields. (B) Differences between mean manual detection of graders and automated detection of PS-OCT before correction of segmentation artifacts. (C) Linear correlation of detected fields after correction of segmentation artifacts (solid line = line of equality [slope β = 1], dashed line = trend line of scatter plot). (D) Differences between mean manual detection of graders and automated detection of PS-OCT after correction of segmentation artifacts. Both, the slope β = 1.18 of the linear regression line (C) as well as the mean difference of −351.7 fields (D) indicate a higher detection by PS-OCT.
Figure 3
 
Plots of detected fields containing hard exudates. (A) Agreement of manual grading between grader 1 and grader 2 was good with a mean difference of 15.7 fields. (B) Differences between mean manual detection of graders and automated detection of PS-OCT before correction of segmentation artifacts. (C) Linear correlation of detected fields after correction of segmentation artifacts (solid line = line of equality [slope β = 1], dashed line = trend line of scatter plot). (D) Differences between mean manual detection of graders and automated detection of PS-OCT after correction of segmentation artifacts. Both, the slope β = 1.18 of the linear regression line (C) as well as the mean difference of −351.7 fields (D) indicate a higher detection by PS-OCT.
Table.
 
Agreement Matrix for the Detection of HEs by the Three Different Raters in the Graded Fields
Table.
 
Agreement Matrix for the Detection of HEs by the Three Different Raters in the Graded Fields
Rater Rater
JL BG PS-OCT
HEs No HEs Total HEs No HEs Total HEs No HEs Total
JL
 HEs 13,427 4,978 18,405 10,853 7,552 18,405
 No HEs 4,633 337,410 342,043 17,944 324,099 342,043
 Total 18,060 342,388 360,448 28,797 331,651 360,448
BG
 HEs 13,427 4,633 18,060 10,562 7,498 18,060
 No HEs 4,978 337,410 342,388 18,235 324,153 342,388
 Total 18,405 342,043 360,448 28,797 331,651 360,448
κ* = 0.72 κ* (JL) = 0.44
κ* (BG) = 0.43
In the combined analysis of PS-OCT B-scans and aligned CF, depolarizing formations were mainly found in the outer nuclear layer (ONL) and outer plexiform layer (OPL) but also in the inner nuclear layer (INL) and inner plexiform layer as seen in PS-OCT (Figs. 1D, 1G, 4C, 5C). On PS-OCT, individual, small depolarizing foci (≤15 pixels in diameter) were distributed throughout all retinal layers (Fig. 5C), while larger formations accumulated at the border of the INL and OPL (Figs. 1D, 1G, 4C, 5C). All depolarizing foci and formations showed a shadowing phenomenon in the projection of the scanning beam, which is known to be typical for HEs in conventional SD OCT imaging. In contrast, such small depolarizing foci could not be identified in the corresponding CF (Fig. 5A). 
Figure 4
 
Artifacts in PS-OCT segmentation. (A) Magnification of the graded area within a color fundus photograph. The arrow indicates a cotton-wool spot. (B) Overlay of the color fundus photograph and the HE thickness map generated by PS-OCT. Note how the vessels (arrowheads) and the cotton-wool spot (arrow) cause segmentation artifacts. The green lines in (B) indicate the location of the corresponding PS-OCT B-scans. (C) Depolarizing structures are segmented in red. Note the segmentation artifact at the location of the vessel (arrowhead) and the cotton-wool spot (arrow). Small arrows in (C) indicate nondepolarizing hyperreflective lesions.
Figure 4
 
Artifacts in PS-OCT segmentation. (A) Magnification of the graded area within a color fundus photograph. The arrow indicates a cotton-wool spot. (B) Overlay of the color fundus photograph and the HE thickness map generated by PS-OCT. Note how the vessels (arrowheads) and the cotton-wool spot (arrow) cause segmentation artifacts. The green lines in (B) indicate the location of the corresponding PS-OCT B-scans. (C) Depolarizing structures are segmented in red. Note the segmentation artifact at the location of the vessel (arrowhead) and the cotton-wool spot (arrow). Small arrows in (C) indicate nondepolarizing hyperreflective lesions.
Figure 5
 
Depolarizing signal of precursors of HEs. (A) Color fundus photograph of an eye with severe DR and clusters of HEs spread out over the posterior pole. (B) Overlay of the magnified color fundus photograph and the HE thickness map generated by PS-OCT. Note the numerous small depolarizing particles in the HE map (blue dots). (C) PS-OCT B-scans corresponding to the green lines in (A) and (B). The corresponding arrowheads (big versus small) point out examples of clinically invisible HE precursors (A) that are detected by PS-OCT (blue dots in [B], red dots in [C]). Small arrows in (C) indicate nondepolarizing hyperreflective lesions.
Figure 5
 
Depolarizing signal of precursors of HEs. (A) Color fundus photograph of an eye with severe DR and clusters of HEs spread out over the posterior pole. (B) Overlay of the magnified color fundus photograph and the HE thickness map generated by PS-OCT. Note the numerous small depolarizing particles in the HE map (blue dots). (C) PS-OCT B-scans corresponding to the green lines in (A) and (B). The corresponding arrowheads (big versus small) point out examples of clinically invisible HE precursors (A) that are detected by PS-OCT (blue dots in [B], red dots in [C]). Small arrows in (C) indicate nondepolarizing hyperreflective lesions.
Artifacts could be identified in 4 of 22 PS-OCT data sets (four lowest data points in Fig. 3B), resulting from depolarizing formations in the RNFL caused by retinal vessels (data sets No.1, No.3, and No.19), hemorrhage (data set No.17), cotton-wool spots (data set No.1; Fig. 4), as well as errors of the RPE segmentation algorithm (data set No.19; Figs. 6A, 6B). To compare uncorrected and corrected PS-OCT performance, artifacts were manually corrected in the corresponding B-scans and re-entered into the analysis (Figs. 6C, 6D). 
Figure 6
 
Correction of artifacts in PS-OCT segmentation. (A) Overlay of a color fundus photograph and the HE thickness map generated by PS-OCT before manual correction. Note how the vessels (arrowheads) and regions of RPE atrophy (dashed rectangle) cause segmentation artifacts. (B) PS-OCT B-scan corresponding to green line in (A): depolarizing structures are segmented in red. Note the segmentation artifacts at the location of the vessel (arrowhead) and the RPE atrophy (dashed rectangle). (C) PS-OCT B-scan after manual correction of segmentation artifacts (arrowhead, dashed rectangle). Corrected B-scans are then used to generate corrected HE thickness maps (D).
Figure 6
 
Correction of artifacts in PS-OCT segmentation. (A) Overlay of a color fundus photograph and the HE thickness map generated by PS-OCT before manual correction. Note how the vessels (arrowheads) and regions of RPE atrophy (dashed rectangle) cause segmentation artifacts. (B) PS-OCT B-scan corresponding to green line in (A): depolarizing structures are segmented in red. Note the segmentation artifacts at the location of the vessel (arrowhead) and the RPE atrophy (dashed rectangle). (C) PS-OCT B-scan after manual correction of segmentation artifacts (arrowhead, dashed rectangle). Corrected B-scans are then used to generate corrected HE thickness maps (D).
Mean ± SD fields per eye graded as “HE detected” changed from 1309.0 ± 1240.5 (uncorrected) to 1180.5 ± 1009.8 (corrected; P = 0.1) and still differed significantly from the results of both graders (both P ≤ 0.02). Agreement of the corrected results with the graders did not change significantly, with κ = 0.45 for grader 1 and κ = 0.44 for grader 2. However, Pearson's correlation coefficient for linear regression improved to r = 0.81 (slope β = 1.18; Figs. 3C, 3D). 
Discussion
The aim of this study was to evaluate the feasibility of a new imaging modality for HE assessment, based on their intrinsic polarization properties in PS-OCT imaging in patients with diabetic maculopathy. Images obtained by using PS-OCT were systematically compared to CF, the existing gold standard for imaging HEs. Thickness maps of intraretinal depolarizing structures of the posterior pole were generated by using three-dimensional PS-OCT data sets and correlated with corresponding CF. Results of statistical image analysis showed a marked correlation between HEs in CF and intraretinal depolarizing structures. The findings suggest that these depolarizing structures represent HEs, and that PS-OCT may provide a novel method for assessing and quantifying HEs. 
Automated assessment and segmentation of HEs is a difficult task in both CF and OCT. Automated HE detection in CF is aggravated by uneven illumination, changing contrast, and color variation of the retinal images. Several attempts have been made to segment HEs from the retinal background, based on grey level thresholding, 34 homogeneity of HE illumination, 35 and edge detection, 15 or mixture models and clustering algorithms. 16 Since these techniques are based on fundus photography assessment, quantifying the extent of HEs is limited to two-dimensional information. However, as known from histology and SD OCT, HEs are found throughout multiple retinal layers and tend to cluster and overlay each other. Recent studies have presented detailed analyses of the presence and distribution of HEs, based on high-resolution SD-OCT imaging. Bolz et al. 9 first described HEs secondary to DME as hyperreflective lesions in SD-OCT mainly located at the border of the ONL and OPL. Additionally, clinically invisible precursors of HEs, microfoci, were detected in all retinal layers before condensing to larger, clinically visible aggregates over the course of time. Other studies have correlated the presence of such microfoci with decreased visual acuity 36 or manually quantified and described the presence and course of HEs and precursors following treatments of DME. 37,38 All of these studies underline the importance of OCT imaging of HEs and precursors for determining the extent of lipid extravasation resulting from blood–retina barrier breakdown. Further, the need for software-based demarcations and calculations is emphasized by these authors owing to the difficulties and limitations of the currently required manual grading. 38  
The findings of our study suggest that PS-OCT overcomes the limitations of two-dimensional CF and intensity-based SD-OCT imaging for assessment of HEs. Based on their intrinsic property to change the state of polarized light, HEs could be automatically detected and segmented. By raster scanning the macular region of interest, thickness maps of HEs were generated from 128 individual B-scans. Good agreement with clinical findings of HEs (in this study represented by CF) was achieved with the new method. 
The remaining discrepancies between the results obtained by CF and PS-OCT can be attributed to the following reasons: First, accuracy and precision of OCT imaging in general requires clear optical media, as well as sufficient compliance of the patient, to avoid imaging artifacts due to poor image quality or eye motion. This also affects PS-OCT. In this study, images of insufficient quality were eliminated by defining media opacities as exclusion criteria, and scans with evident motion artifacts were repeated until acceptable quality was achieved. Nonetheless, superimposing CF and PS-OCT images often did not reach perfect accuracy of retinal landmarks (e.g., vessels). This is likely caused by residual motion artifacts and can lead to a loss of exact matching between the compared imaging modalities. Such a slight mismatch has a large influence on the pixel-to-pixel correlation of small features such as tiny HEs but should not influence the total count of HEs within the images, which is the relevant diagnostic quantity. Nevertheless, further improvement of PS-OCT image acquisition speed, 3942 as well as using a retinal tracker, 43 could likely overcome this limitation in future generations of PS-OCT instruments. 
Furthermore, errors in the segmentation algorithm may be caused by lesions such as cotton-wool spots, profound RPE atrophy, or blood vessels. In our cohort, four data sets contained noticeable segmentation artifacts. After manual correction of these data sets, linear correlation between the two modalities improved. However, neither means of detected fields nor interrater agreement of pixelwise HE detection differed significantly from the uncorrected data. This indicates a rather negligible effect of segmentation errors on the overall performance of HE detection in PS-OCT imaging. 
Another explanation for some of the observed discrepancies might be found in a detailed breakdown of individual PS-OCT B-scans: In an analysis of all acquired B-scans, we found that HEs showed depolarizing properties, as did the previously described clinically invisible precursors of HEs. 9,3638 However, the depolarization behavior of these precursors (defined as small hyperreflective lesions in SD-OCT) was heterogeneous. While some hyperreflective lesions showed depolarizing signals, others did not. This was observed to be independent of size, degree of reflectivity in the intensity image, and location within the retinal layers (Figs. 1D, 1G, 4C, 5C). There are two possible explanations: First, the size of these precursors might be too small to provide a sufficient number of independent data points within the DOPU evaluation window to be picked up by the algorithm, which measures the variance of polarization states among neighboring image pixels. Second, different types of hyperreflective microfoci might exist: Using immunohistology, Cusick et al. 8 have found and imaged not only a dense concentration of apolipoprotein B deposits surrounding retinal vessels (as in hyperreflective lesions), but also cellular components such as foam cells and leukocytes. Further studies will be required to describe and categorize these hyperreflective microfoci in more detail. 
Since PS-OCT detects at least part of these precursors, it would be expected to have an increased sensitivity for HE detection, as compared to CF imaging. This was in accordance with the significantly higher rates of detected HE-containing fields in PS-OCT in 77.3% of cases as compared to the individual graders as well as the means of the graders (P = 0.02) (Table; Figs. 3, 5). Therefore, the total sensitivity of PS-OCT to detect any HE is indeed higher than that of CF. 
Another advantage of PS-OCT for HE assessment is the three-dimensional component of the detection. Hard exudates were automatically segmented throughout all retinal layers. Therefore, their distribution within individual retinal layers as well as their volume can be obtained, quantities that are inaccessible to CF imaging. This might add valuable information to the understanding of DME and its underlying pathomechanisms. 
Given the prototype state of our PS-OCT instrument, the preliminary segmentation algorithms, and the limited study population, our data provide satisfactory accuracy to prove the presented principle. Our results strongly suggest that the presented pattern of depolarizing structures in the retina represent HEs. Further studies are underway to assess the reliability and reproducibility of volumetric assessment of HEs by PS-OCT in order to provide a new tool in the analysis of DME and treatment responses. 
Acknowledgments
The authors thank Philipp Roberts, Christopher Schütze, and Karin Kofler for helping with recording and processing the PS-OCT images; Erich Goetzinger for discussions on technical background; and Skyler L. Jackman and Michael M. Lin for English proofreading. 
Supported by the Austrian Science Fund (CKH; FWF Grant No. P19624-B02, Vienna, Austria) and the European Union (CKH; projectFUNOCT, FP7 HEALTH, Contract No. 201880). 
Disclosure: J. Lammer, None; M. Bolz, P; B. Baumann, P; M. Pircher, P; Canon (F); B. Gerendas, None; F. Schlanitz, None; C.K. Hitzenberger, Canon (F), P; U. Schmidt-Erfurth, P 
References
Chen L Magliano DJ Zimmet PZ. The worldwide epidemiology of type 2 diabetes mellitus—present and future perspectives. Nat Rev Endocrinol . 2012; 8: 228–236. [CrossRef]
Kempen JH O'Colmain BJ Leske MC The prevalence of diabetic retinopathy among adults in the United States. Arch Ophthalmol . 2004; 122: 552–563. [CrossRef] [PubMed]
Yau JWY Rogers SL Kawasaki R Global prevalence and major risk factors of diabetic retinopathy. Diabetes Care . 2012; 35: 556–564. [CrossRef] [PubMed]
Ding J Wong T. Current epidemiology of diabetic retinopathy and diabetic macular edema. Curr Diab Rep . 2012; 12: 346–354. [CrossRef] [PubMed]
Klein R Klein BE Moss SE Cruickshanks KJ. The Wisconsin Epidemiologic Study of Diabetic Retinopathy, XV: the long-term incidence of macular edema. Ophthalmology . 1995; 102: 7–16. [CrossRef] [PubMed]
Klein R Knudtson MD Lee KE The Wisconsin Epidemiologic Study of Diabetic Retinopathy XXIII: the twenty-five-year incidence of macular edema in persons with type 1 diabetes. Ophthalmology . 2009; 116: 497–503. [CrossRef] [PubMed]
Chew EY Klein ML Ferris FL Association of elevated serum lipid levels with retinal hard exudate in diabetic retinopathy: Early Treatment Diabetic Retinopathy Study (ETDRS) Report 22. Arch Ophthalmol . 1996; 114: 1079–1084. [CrossRef] [PubMed]
Cusick M Chew EY Chan CC Histopathology and regression of retinal hard exudates in diabetic retinopathy after reduction of elevated serum lipid levels. Ophthalmology . 2003; 110: 2126–2133. [CrossRef] [PubMed]
Bolz M Schmidt-Erfurth U Deak G Optical coherence tomographic hyperreflective foci: a morphologic sign of lipid extravasation in diabetic macular edema. Ophthalmology . 2009; 116: 914–920. [CrossRef] [PubMed]
Fong DS Segal PP Myers F Subretinal fibrosis in diabetic macular edema: ETDRS report 23. Arch Ophthalmol . 1997; 115: 873–877. [CrossRef] [PubMed]
Early Treatment Diabetic Retinopathy Study Research Group. Grading diabetic retinopathy from stereoscopic color fundus photographs—an extension of the modified Airlie House classification: ETDRS report number 10. Ophthalmology . 1991; 98: 786–806. [CrossRef] [PubMed]
Silva PS Cavallerano JD Sun JK Nonmydriatic ultrawide field retinal imaging compared with dilated standard 7-field 35-mm photography and retinal specialist examination for evaluation of diabetic retinopathy. Am J Ophthalmol . 2012; 154: 549–559.e2. [CrossRef] [PubMed]
Geitzenauer W Hitzenberger CK Schmidt-Erfurth UM. Retinal optical coherence tomography: past, present and future perspectives. Br J Ophthalmol . 2011; 95: 171–177. [CrossRef] [PubMed]
Chalam KV Bressler SB Edwards AR Retinal thickness in people with diabetes and minimal or no diabetic retinopathy: Heidelberg Spectralis optical coherence tomography. Invest Ophthalmol Vis Sci . 2012; 53: 8154–8161. [CrossRef] [PubMed]
Li H Chutatape O. Automated feature extraction in color retinal images by a model based approach. IEEE Trans Biomed Eng . 2004; 51: 246–254. [CrossRef] [PubMed]
Sánchez CI García M Mayo A Retinal image analysis based on mixture models to detect hard exudates. Med Image Anal . 2009; 13: 650–658. [CrossRef] [PubMed]
Akram UM Khan SA. Automated detection of dark and bright lesions in retinal images for early detection of diabetic retinopathy. J Med Syst . 2012; 36: 3151–3162. [CrossRef] [PubMed]
Otani T Kishi S. Tomographic findings of foveal hard exudates in diabetic macular edema. Am J Ophthalmol . 2001; 131: 50–54. [CrossRef] [PubMed]
Hee MR Huang D Swanson EA Fujimoto JG. Polarization-sensitive low-coherence reflectometer for birefringence characterization and ranging. J Opt Soc Am B . 1992; 9: 903–908. [CrossRef]
De Boer JF Milner TE Nelson JS. Determination of the depth-resolved Stokes parameters of light backscattered from turbid media by use of polarization-sensitive optical coherence tomography. Opt Lett . 1999; 24: 300–302. [CrossRef] [PubMed]
Götzinger E Pircher M Hitzenberger CK. High speed spectral domain polarization sensitive optical coherence tomography of the human retina. Opt Express . 2005; 13: 10217–10229. [CrossRef] [PubMed]
Cense B Chen TC Park BH Thickness and birefringence of healthy retinal nerve fiber layer tissue measured with polarization-sensitive optical coherence tomography. Invest Ophthalmol Vis Sci . 2004; 45: 2606–2612. [CrossRef] [PubMed]
Cense B Mujat M Chen TC Polarization-sensitive spectral-domain optical coherence tomography using a single line scan camera. Opt Express . 2007; 15: 2421–2431. [CrossRef] [PubMed]
Miyazawa A Yamanari M Makita S Tissue discrimination in anterior eye using three optical parameters obtained by polarization sensitive optical coherence tomography. Opt Express . 2009; 17: 17426–17440. [CrossRef] [PubMed]
Lim Y Yamanari M Fukuda S Birefringence measurement of cornea and anterior segment by office-based polarization-sensitive optical coherence tomography. Biomed Opt Express . 2011; 2: 2392–2402. [CrossRef] [PubMed]
Pircher M Götzinger E Findl O Human macula investigated in vivo with polarization-sensitive optical coherence tomography. Invest Ophthalmol Vis Sci . 2006; 47: 5487–5494. [CrossRef] [PubMed]
Michels S Pircher M Geitzenauer W Value of polarisation-sensitive optical coherence tomography in diseases affecting the retinal pigment epithelium. Br J Ophthalmol . 2008; 92: 204–209. [CrossRef] [PubMed]
Ahlers C Götzinger E Pircher M Imaging of the retinal pigment epithelium in age-related macular degeneration using polarization-sensitive optical coherence tomography. Invest Ophthalmol Vis Sci . 2010; 51: 2149–2157. [CrossRef] [PubMed]
Lammer J Bolz M Baumann B Imaging retinal pigment epithelial proliferation secondary to PASCAL photocoagulation in vivo by polarization-sensitive optical coherence tomography. Am J Ophthalmol . 2013; 155: 1058–1067. [CrossRef] [PubMed]
Pircher M Hitzenberger CK Schmidt-Erfurth U. Polarization sensitive optical coherence tomography in the human eye. Prog Retin Eye Res . 2011; 30: 431–4 51. [CrossRef] [PubMed]
Baumann B Götzinger E Pircher M Segmentation and quantification of retinal lesions in age-related macular degeneration using polarization-sensitive optical coherence tomography. J Biomed Opt . 2010; 15: 061704. [CrossRef] [PubMed]
Götzinger E Pircher M Geitzenauer W Retinal pigment epithelium segmentation by polarization sensitive optical coherence tomography. Opt Express . 2008; 16: 16410–16422. [CrossRef] [PubMed]
Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas . 1960; 20: 37–46. [CrossRef]
Walter T Klein J-C Massin P Erginay A. A contribution of image processing to the diagnosis of diabetic retinopathy—detection of exudates in color fundus images of the human retina. IEEE Trans Med Imaging . 2002; 21: 1236–1243. [CrossRef] [PubMed]
Sinthanayothin C Boyce JF Williamson TH Automated detection of diabetic retinopathy on digital fundus images. Diabet Med . 2002; 19: 105–112. [CrossRef] [PubMed]
Uji A Murakami T Nishijima K Association between hyperreflective foci in the outer retina, status of photoreceptor layer, and visual acuity in diabetic macular edema. Am J Ophthalmol . 2012; 153: 710–717.e1. [CrossRef] [PubMed]
Deák GG Bolz M Kriechbaum K Effect of retinal photocoagulation on intraretinal lipid exudates in diabetic macular edema documented by optical coherence tomography. Ophthalmology . 2010; 117: 773–779. [CrossRef] [PubMed]
Framme C Schweizer P Imesch M Behavior of SD-OCT-detected hyperreflective foci in the retina of anti-VEGF-treated patients with diabetic macular edema. Invest Ophthalmol Vis Sci . 2012; 53: 5814–5818. [CrossRef] [PubMed]
Baumann B Choi W Potsaid B Swept source/Fourier domain polarization sensitive optical coherence tomography with a passive polarization delay unit. Opt Express . 2012; 20: 10218–10230. [CrossRef] [PubMed]
Zotter S Pircher M Torzicky T Large-field high-speed polarization sensitive spectral domain OCT and its applications in ophthalmology. Biomed Opt Express . 2012; 3: 2720–2732. [CrossRef] [PubMed]
Zotter S Pircher M Götzinger E Measuring retinal nerve fiber layer birefringence, retardation, and thickness using wide-field, high-speed polarization sensitive spectral domain OCT. Invest Ophthalmol Vis Sci . 2013; 54: 72–84. [CrossRef] [PubMed]
Torzicky T Pircher M Zotter S High-speed retinal imaging with polarization-sensitive OCT at 1040 nm. Optom Vis Sci . 2012; 89: 585–592. [CrossRef] [PubMed]
Sugita M Zotter S Pircher M Motion artifact and speckle noise reduction in polarization sensitive optical coherence tomography by retinal tracking. Biomed Opt Express . 2013; 5: 106. [CrossRef] [PubMed]
Footnotes
 For the Diabetic Retinopathy Research Group (DRRG) Vienna.
Footnotes
 See the appendix for the members of the Diabetic Retinopathy Research Group (DRRG) Vienna.
Appendix
Diabetic Retinopathy Research Group (DRRG) Vienna
Sonja Prager 
Matthias Bolz 
Jan Lammer 
Gabor Deak 
Katharina Kriechbaum 
Georgios Mylonas 
Christoph Mitsch 
Andreas Pollreisz 
Berthold Pemp 
Christoph Scholda 
Katharina Kefer 
Figure 1
 
Segmentation of hard exudates by PS-OCT. (A) Color fundus photography of a DME patient with HEs. The dotted box indicates the location of the PS-OCT cube. (B) Fundus projection image generated from the PS-OCT data set. The yellow line indicates the location of the B-scan images on the right. (C) Intensity B-scan image. (D) Overlay of depolarizing pixels (DOPU < 0.8, red) on the intensity image. (E) En face image showing the summation of all depolarizing (red) pixels in every A-scan. The summated pixels include both the RPE and HEs. (F) Depolarizing pixels (red) within the HE segmentation band (light grey). (G) Segmented HEs (red) overlaid on intensity image. (H) HE thickness map generated by summing pixels within the HE segmentation band.
Figure 1
 
Segmentation of hard exudates by PS-OCT. (A) Color fundus photography of a DME patient with HEs. The dotted box indicates the location of the PS-OCT cube. (B) Fundus projection image generated from the PS-OCT data set. The yellow line indicates the location of the B-scan images on the right. (C) Intensity B-scan image. (D) Overlay of depolarizing pixels (DOPU < 0.8, red) on the intensity image. (E) En face image showing the summation of all depolarizing (red) pixels in every A-scan. The summated pixels include both the RPE and HEs. (F) Depolarizing pixels (red) within the HE segmentation band (light grey). (G) Segmented HEs (red) overlaid on intensity image. (H) HE thickness map generated by summing pixels within the HE segmentation band.
Figure 2
 
Superposition of CF and PS-OCT images. (A) Color fundus photograph of a left eye. The yellow box indicates the area scanned by PS-OCT. (B) Using retinal vessels as landmarks, the PS-OCT raster scan is superposed on the CF image. (C) The corresponding CF area is then cropped and exported for manual and automated assessment of HEs. (D) Overlay of the manual grading (in black, by grader 1). (E) Overlay of the automated assessment by PS-OCT (false color thickness map).
Figure 2
 
Superposition of CF and PS-OCT images. (A) Color fundus photograph of a left eye. The yellow box indicates the area scanned by PS-OCT. (B) Using retinal vessels as landmarks, the PS-OCT raster scan is superposed on the CF image. (C) The corresponding CF area is then cropped and exported for manual and automated assessment of HEs. (D) Overlay of the manual grading (in black, by grader 1). (E) Overlay of the automated assessment by PS-OCT (false color thickness map).
Figure 3
 
Plots of detected fields containing hard exudates. (A) Agreement of manual grading between grader 1 and grader 2 was good with a mean difference of 15.7 fields. (B) Differences between mean manual detection of graders and automated detection of PS-OCT before correction of segmentation artifacts. (C) Linear correlation of detected fields after correction of segmentation artifacts (solid line = line of equality [slope β = 1], dashed line = trend line of scatter plot). (D) Differences between mean manual detection of graders and automated detection of PS-OCT after correction of segmentation artifacts. Both, the slope β = 1.18 of the linear regression line (C) as well as the mean difference of −351.7 fields (D) indicate a higher detection by PS-OCT.
Figure 3
 
Plots of detected fields containing hard exudates. (A) Agreement of manual grading between grader 1 and grader 2 was good with a mean difference of 15.7 fields. (B) Differences between mean manual detection of graders and automated detection of PS-OCT before correction of segmentation artifacts. (C) Linear correlation of detected fields after correction of segmentation artifacts (solid line = line of equality [slope β = 1], dashed line = trend line of scatter plot). (D) Differences between mean manual detection of graders and automated detection of PS-OCT after correction of segmentation artifacts. Both, the slope β = 1.18 of the linear regression line (C) as well as the mean difference of −351.7 fields (D) indicate a higher detection by PS-OCT.
Figure 4
 
Artifacts in PS-OCT segmentation. (A) Magnification of the graded area within a color fundus photograph. The arrow indicates a cotton-wool spot. (B) Overlay of the color fundus photograph and the HE thickness map generated by PS-OCT. Note how the vessels (arrowheads) and the cotton-wool spot (arrow) cause segmentation artifacts. The green lines in (B) indicate the location of the corresponding PS-OCT B-scans. (C) Depolarizing structures are segmented in red. Note the segmentation artifact at the location of the vessel (arrowhead) and the cotton-wool spot (arrow). Small arrows in (C) indicate nondepolarizing hyperreflective lesions.
Figure 4
 
Artifacts in PS-OCT segmentation. (A) Magnification of the graded area within a color fundus photograph. The arrow indicates a cotton-wool spot. (B) Overlay of the color fundus photograph and the HE thickness map generated by PS-OCT. Note how the vessels (arrowheads) and the cotton-wool spot (arrow) cause segmentation artifacts. The green lines in (B) indicate the location of the corresponding PS-OCT B-scans. (C) Depolarizing structures are segmented in red. Note the segmentation artifact at the location of the vessel (arrowhead) and the cotton-wool spot (arrow). Small arrows in (C) indicate nondepolarizing hyperreflective lesions.
Figure 5
 
Depolarizing signal of precursors of HEs. (A) Color fundus photograph of an eye with severe DR and clusters of HEs spread out over the posterior pole. (B) Overlay of the magnified color fundus photograph and the HE thickness map generated by PS-OCT. Note the numerous small depolarizing particles in the HE map (blue dots). (C) PS-OCT B-scans corresponding to the green lines in (A) and (B). The corresponding arrowheads (big versus small) point out examples of clinically invisible HE precursors (A) that are detected by PS-OCT (blue dots in [B], red dots in [C]). Small arrows in (C) indicate nondepolarizing hyperreflective lesions.
Figure 5
 
Depolarizing signal of precursors of HEs. (A) Color fundus photograph of an eye with severe DR and clusters of HEs spread out over the posterior pole. (B) Overlay of the magnified color fundus photograph and the HE thickness map generated by PS-OCT. Note the numerous small depolarizing particles in the HE map (blue dots). (C) PS-OCT B-scans corresponding to the green lines in (A) and (B). The corresponding arrowheads (big versus small) point out examples of clinically invisible HE precursors (A) that are detected by PS-OCT (blue dots in [B], red dots in [C]). Small arrows in (C) indicate nondepolarizing hyperreflective lesions.
Figure 6
 
Correction of artifacts in PS-OCT segmentation. (A) Overlay of a color fundus photograph and the HE thickness map generated by PS-OCT before manual correction. Note how the vessels (arrowheads) and regions of RPE atrophy (dashed rectangle) cause segmentation artifacts. (B) PS-OCT B-scan corresponding to green line in (A): depolarizing structures are segmented in red. Note the segmentation artifacts at the location of the vessel (arrowhead) and the RPE atrophy (dashed rectangle). (C) PS-OCT B-scan after manual correction of segmentation artifacts (arrowhead, dashed rectangle). Corrected B-scans are then used to generate corrected HE thickness maps (D).
Figure 6
 
Correction of artifacts in PS-OCT segmentation. (A) Overlay of a color fundus photograph and the HE thickness map generated by PS-OCT before manual correction. Note how the vessels (arrowheads) and regions of RPE atrophy (dashed rectangle) cause segmentation artifacts. (B) PS-OCT B-scan corresponding to green line in (A): depolarizing structures are segmented in red. Note the segmentation artifacts at the location of the vessel (arrowhead) and the RPE atrophy (dashed rectangle). (C) PS-OCT B-scan after manual correction of segmentation artifacts (arrowhead, dashed rectangle). Corrected B-scans are then used to generate corrected HE thickness maps (D).
Table.
 
Agreement Matrix for the Detection of HEs by the Three Different Raters in the Graded Fields
Table.
 
Agreement Matrix for the Detection of HEs by the Three Different Raters in the Graded Fields
Rater Rater
JL BG PS-OCT
HEs No HEs Total HEs No HEs Total HEs No HEs Total
JL
 HEs 13,427 4,978 18,405 10,853 7,552 18,405
 No HEs 4,633 337,410 342,043 17,944 324,099 342,043
 Total 18,060 342,388 360,448 28,797 331,651 360,448
BG
 HEs 13,427 4,633 18,060 10,562 7,498 18,060
 No HEs 4,978 337,410 342,388 18,235 324,153 342,388
 Total 18,405 342,043 360,448 28,797 331,651 360,448
κ* = 0.72 κ* (JL) = 0.44
κ* (BG) = 0.43
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×