Comparison of Geographic Atrophy Measurements from the OCT Fundus Image and the Sub-RPE Slab Image

Transcription

1 CLINICAL SCIENCE Comparison of Geographic Atrophy Measurements from the OCT Fundus Image and the Sub-RPE Slab Image Zohar Yehoshua, MD, MHA; Carlos Alexandre A. Garcia Filho, MD; Fernando M. Penha, MD, PhD; Giovanni Gregori, PhD; Paul F. Stetson, PhD; William J. Feuer, MS; Philip J. Rosenfeld, MD, PhD PURPOSE: To compare two different approaches to measuring areas of geographic atrophy (GA) using spectral-domain optical coherence tomography (SD-OCT). METHODS: Fifty eyes with GA were imaged with an SD-OCT instrument. OCT fundus images and sub retinal pigment epithelium (RPE) slab images were generated. Three graders manually drew the GA boundaries on both en face images. An automated algorithm was used to segment the GA boundaries from the sub-rpe slabs. RESULTS: The agreement between the three manual measurements on both OCT fundus images (ICC =.998) and sub-rpe slabs (ICC =.999) was excellent. Area measurements from OCT fundus images and sub-rpe slabs were highly correlated. The agreement between manual and automated measurements on the sub-rpe slabs was very good (ICC =.795). CONCLUSION: Both OCT fundus images and sub- RPE slab images proved useful for measuring GA in age-related macular degeneration. The automated algorithm typically provided useful measurements of GA area from the sub-rpe slabs. [Ophthalmic Surg Lasers Imaging Retina. 2013;44: ] INTRODUCTION Age-related macular degeneration (AMD) is the leading cause of severe vision loss in developed countries. 1 Geographic atrophy (GA) is a significant cause of both moderate and severe central vision loss and is bilateral in most patients with advanced AMD. 2-5 GA represents the loss of photoreceptors, retinal pigment epithelium (RPE), and choriocapillaris within the macula. The natural history of GA has been described as a progressive condition that evolves through stages, with loss of vision occurring over years. 5-7 In the early stages of GA, the atrophic areas tend to involve the parafoveal retina; however, over time these areas may coalesce, and new atrophic patches can emerge. This can result in a horseshoe pattern of atrophy that progressively encircles the fovea. 6,8 Once the fovea becomes involved, a significant loss of vision occurs. In the United States, GA is present in 3.5% of people aged 75 years and older, 9,10 and its prevalence rises to 22% or more in those older than 90 years. 11 Spectral-domain optical coherence tomography (SD-OCT) can be used to image and measure GA. 12,13 Several studies have used SD-OCT to describe a wide spectrum of morphologic alterations that appear within the atrophic area, as well as within the surrounding retinal tissue A common finding within the atrophic lesions is the absence of the outer nuclear layer, as well as the absence of SD-OCT bands 1, 2, 3, and 4, which correspond to anatomical layers including the external limiting membrane, the inner-segment/outer- From the Department of Ophthalmology, Bascom Palmer Eye Institute, University of Miami Miller School of Medicine, Miami, Florida (ZY, CAAGF, FMP, GG, WJF, PJR); and Carl Zeiss Meditec Inc., Dublin, California (PFS). Originally submitted May 9, Accepted for publication December 31, The research was supported by a grant from Carl Zeiss Meditec, an unrestricted grant from Research to Prevent Blindness, NEI core center grant P30 EY014801, and Department of Defense Grant W81XWH to the University of Miami, the Macula Vision Research Foundation, the Jerome A. Yavitz Charitable Foundation, Feig Family Foundation, and the Emma Clyde Hodge Memorial Foundation. Drs. Yehoshua, Penha, Rosenfeld, and Gregori received research support from Carl Zeiss Meditec. Dr. Gregori and the University of Miami co-own a patent that is licensed to Carl Zeiss Meditec. Dr. Rosenfeld has received honoraria for lectures from Carl Zeiss Meditec. Dr. Feuer has no financial or proprietary interest in the materials presented herein. Dr. Stetson is an employee of Carl Zeiss Meditec. Address correspondence to Philip J. Rosenfeld, MD, PhD, Bascom Palmer Eye Institute, 900 NW 17th St., Miami, FL 33136; ; Fax: ; prosenfeld@med.miami.edu. doi: / March/April 2013 Vol. 44, No

2 Figure 1. Bland-Altman plots of measurements made on en face spectral-domain OCT images by three graders. The agreement between average lesion area measurements made by three graders on the spectral-domain OCT fundus images and sub-rpe slabs is shown. OFI = OCT fundus image. segment layer, and the RPE Bruch s membrane complex. 15 Schmitz-Valckenberg et al 19 compared the fundus autofluorescence and SD-OCT appearance of eyes with GA and showed that the mean length of an atrophic lesion measured on the fundus autofluorescence image had the closest agreement with the appearance of choroidal hyperreflectivity on the SD-OCT B-scan, and the reduction of the fundus autofluorescence signal seen from GA was spatially correlated with the abrupt transition on the SD-OCT B-scan from a hyporeflective choroid to a hyperreflective choroid. This increased penetration of light below Bruch s membrane is presumably due to the loss of the RPE and choriocapillaris. 13,14 Sayegh et al 20 also showed that the area of choroidal signal enhancement on the SD-OCT B- scans correlated well with the hypofluorescent area measured on fundus autofluorescence. OCT imaging can also provide an en face image of GA that complements the B-scan cross-sectional images. This OCT fundus image (OFI) is generated by summing the signal of each of the A-scans and viewing their relative values en face. 14,21,22 It is this en face OFI that results in the visualization of GA as a bright area on the image due to the increased penetration of light into the choroid where atrophy has occurred in the macula. The increased OCT choroidal signal associated with GA results in the GA appearing brighter on the en face OFI compared with the surrounding areas because the absence of the RPE and choriocapillaris in GA results in more light penetrating the choroid and being reflected back to the detector compared with the light scattering that occurs with an intact RPE in the surrounding areas. 23 This OFI correlates well with the GA observed on clinical examination, fundus photography, and autofluorescence imaging. 12,13 The current study introduces a new approach for imaging GA, recently introduced in the Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA) advanced RPE analysis, known as the sub-rpe slab, which creates the en face image only from the light reflected from beneath the RPE. By only using the light that penetrates into the choroid, we found the sub- RPE slab had higher contrast at the borders of GA, and the GA appeared more distinct. To validate this new approach, we compared the same datasets and measured the areas of GA observed using the standard OFI and the areas measured using the sub-rpe slab. We then tested the performance of an automatic algorithm (available on the Cirrus HD-OCT) used to detect and quantitate GA areas on the sub-rpe slab, and we compared these automated measurements with measurements obtained manually. SUBJECTS AND METHODS Subjects Patients were recruited from an outpatient clinic at the Bascom Palmer Eye Institute. Approval for the collection and analysis of SD-OCT images was obtained from the institutional review board at the University of Miami Miller School of Medicine, and all patients signed informed consent documents. The study was performed in accordance with HIPAA regulations. Patients enrolled in the study were older than 50 years with a diagnosis of atrophic AMD in at least one eye. The GA could be unifocal or multifocal. Patients were excluded from the study if the study eye demonstrated areas of peripapillary atrophy communicating with the central area of GA. Patients were also excluded if the GA extended outside the central scanning area, which was a square centered on the fovea with dimensions of 6 6 mm. Additional exclusion criteria for the study eye included evidence of neovascular AMD, a history of retinal surgery (including laser treatment), any sign of diabetic retinopathy, a history of retinal vascular occlusion, and a history or evidence of an inherited retinal degeneration. OCT Fundus Image SD-OCT images were obtained using the Cirrus HD- OCT. This SD-OCT device has an axial resolution of Ophthalmic Surgery, Lasers & Imaging Retina Healio.com/OSLIRetina

3 µm and a scanning rate of 27,000 A-scans per second. The central macula of each study eye was imaged using a raster scan pattern. The first number in the scan pattern refers to the number of A-scans used to form each horizontal B-scan, whereas the second number is the total number of horizontal B-scans. The SD-OCT was calibrated so that the horizontal and vertical dimensions of the OFI generated from all 40,000 A-scans measured 6 6 mm on the retina. A single research technician performed all scans. All scans were performed between January 1, 2007, and June 1, 2010, using a single Cirrus SD-OCT instrument and performed after dilation of the pupil with one drop of 2.5% phenylephrine hydrochloride and 1% tropicamide. Attempts were made to perform five scans on each patient; however, low-quality scans with signal strength less than 7 and scans with motion artifacts were discarded in the analysis. The SD-OCT en face fundus image and the sub- RPE slab were obtained for each three-dimensional dataset using the commercially available software on the Cirrus HD-OCT (version 6.0). The OFI is created by summing all pixel intensity values along individual A-scans, while the sub-rpe slab is an en face visualization using only the light penetrating below the RPE into the choroid and sclera. In the healthy regions of RPE, the pigment in the RPE will block most of the light penetrating into the choroid, while in areas of GA, the underlying choroid will be illuminated. The sub-rpe slab is formed by axially projecting only the OCT image data from a region below the contour of a robust polynomial fit to the RPE segmentation. The region extends from 65 µm to 400 µm below the robust RPE fit. Image processing was used to reduce noise and suppress the appearance of the choroidal vasculature. These en face fundus images were then used to measure the area of GA. A proprietary algorithm included in the Cirrus OCT software was used to automatically quantitate the area of GA from the sub-rpe slab. Measurement of GA The OFI and the sub-rpe slab for each patient were exported as a bitmap file measuring pixels. Figure 2. Correlation between measurements made on two en face spectral-domain OCT images by three graders. The correlation between average lesion area measurements made by three graders on the spectral-domain OCT fundus images and sub-rpe slabs is shown. The area of GA was quantified by using a CintiQ WA- COM digitizing tablet (WACOM, Vancouver, WA) and image analysis software (Adobe Photoshop CS2; Adobe Systems, San Jose, CA). Three graders outlined the areas of GA identified on the OFI and the sub-rpe slab. In the cases in which the precise limits of GA were ambiguous, all three graders discussed and reached an agreement on the areas to be included before drawing the final outline. The encircled area was then quantified using software written as a MATLAB (MathWorks, Natick, MA) R2006a routine, which would convert the number of pixels into the area of GA. The pixel area was converted to square millimeters using the same fixed ratio established by the calibrated SD-OCT; the conversion factor per pixel was (36 mm 2 /40,000 pixels). Statistical Methods Multiple scans (more than three) were performed on each patient. Each scan was reviewed for quality, and low-quality scans with signal strength less than 7, as well as scans with motion artifacts, were discarded. The best quality scan was chosen to generate the OFI and the sub-rpe slab. In cases in which several scans had the same good quality, the scan to be used was chosen randomly. Analysis was performed to assess inter-grader agreement of area measurements made from each image. In addition to intraclass correlation coefficient calculations, Bland-Altman plots were constructed to determine limits of agreement March/April 2013 Vol. 44, No

4 Figure 3. Bland-Altman plots of measurements of geographic atrophy from sub-rpe slabs obtained manually and by an automatic algorithm. The agreements between lesion area measurements detected manually by three graders and by the automatic algorithm are shown. between measurements made on the sub-rpe slabs by two graders and between the manual segmentation and the automatic segmentation algorithm. 24 Statistical analysis to test the correlation between OFI and sub-rpe slab was done using the Pearson correlation coefficient. All statistical calculations were carried out with SPSS V17.0 software (SPSS, Chicago, IL). Statistical significance was defined as P <.05. RESULTS Fifty eyes of 37 patients with GA were analyzed. The mean total area measured manually on the OFI by the three graders was 6.43 mm 2 (SD: 4.78), with a range from 0.16 to mm 2. The mean total area measured manually on the sub-rpe slab was 6.41 mm 2 (SD: 4.88), with a range from 0.17 to mm 2. The agreement between the three graders on measurements made from the OFIs and sub-rpe slabs as assessed by intraclass correlation coefficients were and 0.999, respectively. The agreements between the two en face SD-OCT imaging strategies using the Bland-Altman test are shown in Figure 1. Bland-Altman graphs relate the measurement differences between two SD-OCT en face images versus the average measurement from two images, and overall there was little difference between the measurements. The mean difference between OFI and the sub-rpe slab was 0.06 (SD: 0.23), with a range of 0.5 to 0.89 (P =.062). Figure 2 shows the correlation between the measurements from the two en face SD-OCT images, showing a good correlation between these two strategies. To assess the ability of the automatic algorithm to detect and segment the area of GA, we compared the average measurements made by the three graders on the sub-rpe slabs with the measurements made by the automatic segmentation algorithm. Overall, there was a good correlation between the average manual grader measurements and the automatic segmentation algorithm made on the sub-rpe slab. The ICC for the automated algorithm and the averaged graders was (ICC for automated algorithm and graders 1, 2, and 3 were 0.794, 0.791, and 0.798, respectively). As shown in Figure 3, a Bland-Altman graph relates the difference between the automatic segmentations made on the sub-rpe slab and the average manual segmentation performed by the three graders. The mean (SD [range]) of the averaged grader slab measurements was 6.41 (4.88 [0.17 to 18.79), and the automated algorithm mean was 5.27 (3.68 [0.1 to 13.42]). The mean difference was 1.13 (2.60) [2.02 to 12.81] with a 95% confidence interval between 1.9 and 0.4 (P =.003). Plots of the data suggest that there is a moderately strong, statistically significant negative correlation (r = 0.48, P <.001), which demonstrated larger differences with increasing lesion size. The automated algorithm tended to underestimate the size of the large lesions when compared with the manual grading. Figures 4 and 5 show examples of unifocal and multifocal GA with comparisons between two SD-OCT en face images and the results of the manual and automatic segmentation made on the sub-rpe slab images. DISCUSSION Both the OFI and sub-rpe slab images can be used to quantify areas of GA. The measurements obtained from the OFIs and sub-rpe slabs were highly reproducible, and there was excellent agreement between graders for both types of fundus images. There was excellent correlation between the manual measurements of atrophy on the OFI and the sub-rpe slab with a Pearson correlation of Up until recently, there has not been an automatic algorithm capable of detecting and measuring the area of GA imaged using SD-OCT. Schutze et al 25 reported on the use of the retinal thickness map as a way of detecting GA, but there was not a good topographic cor- 130 Ophthalmic Surgery, Lasers & Imaging Retina Healio.com/OSLIRetina

5 Figure 4. Right eye of an 83-year-old man with unifocal geographic atrophy (GA) imaged using different modalities and measured using both manual and automated techniques. The unifocal area of GA is shown on color fundus imaging (A), camera-based fundus autofluorescence imaging (B), OCT fundus imaging with the OCT fundus image superimposed on the autofluorescence image (C), OCT fundus image (area = mm2) (D), and sub-rpe slab image (E). The boundaries of GA were manually outlined on the sub-rpe slab image with the area of GA measured to be mm2 (F). The automatic algorithm measured the area of GA as mm2 (G). Figure 5. Left eye of a 75-year-old woman with multifocal geographic atrophy (GA) imaged using different modalities and measured using both manual and automated techniques. The multifocal area of GA is shown on color fundus imaging (A), camera-based fundus autofluorescence imaging (B), OCT fundus imaging with the OCT fundus image superimposed on the autofluorescence image (C), OCT fundus image (area = mm2) (D), and sub-rpe slab image (E). The boundaries of GA were manually outlined on the sub-rpe slab image with the area of GA measured to be mm2 (F). The automatic algorithm measured the area of GA to be mm2 (G). relation between areas of retinal thinning and actual GA identified using scanning laser ophthalmoscopy and fundus photography. Our study is, to the best of our knowledge, the first to report the use of an au- March/April 2013 Vol. 44, No. 2 tomatic algorithm for measuring GA on SD-OCT images. While we have demonstrated a good correlation with measurements made manually, the automated algorithm for measuring the area of GA tends to un- 131

6 derestimate the size of larger lesions. One explanation for this discrepancy in size between automated and manual measurements may be the result of the degenerating RPE along the border of the GA. The amount of light at the borders may vary depending on the amount of melanin present and the integrity of the choriocapillaris. As a result, the algorithm may miss some parts of the lesion with irregular brightness and/or poor contrast. Even with this caveat in mind, the sub-rpe slab algorithm and an automated algorithm for area detection equipped with an editor tool for modifying the boundaries of GA should prove very useful for the detection and measurement of GA in clinical practice and in clinical trials. These algorithms and the editing software are now available in version 6.0 of the Cirrus operating system. For larger areas of GA that extend beyond the borders of the 6 6 mm scan area, algorithms have been developed that are capable of assembling multiple overlapping SD-OCT fundus images into a larger montage of the macula 26 so that any area of GA can be imaged and measured using this SD-OCT approach. The limitations to this study include the use of lesions in which the area of the GA was wholly contained within the raster scan area measuring 6 6 mm. In addition, we used SD-OCT as the sole imaging modality and did not compare the SD-OCT images to other modalities such as fundus autofluorescence or color fundus images. In this study, we have shown that GA can be reproducibly identified and measured using the two types of en face SD-OCT fundus images. Advantages of using the SD-OCT fundus images compared with other imaging strategies for GA include the convenience and assurance of using only one type of imaging technique for documenting both en face and cross-sectional images of the macula and following patients with both wet and dry AMD. This imaging approach ensures that the area of perceived GA actually corresponds to the loss of photoreceptors and retinal pigment epithelium, which correlates with the loss of visual function. REFERENCES 1. Van Leeuwen R, Klaver CC, Vingerling JR, Hofman A, de Jong PT. Epidemiology of age-related maculopathy: A review. Eur J Epidemiol. 2003;18(9): Schatz H, McDonald HR. Atrophic macular degeneration. Rate of spread of geographic atrophy and visual loss. Ophthalmology. 1989;96(10): Potter JW, Thallemer JM. Geographic atrophy of the retinal pigment epithelium: diagnosis and vision rehabilitation. J Am Optom Assoc. 1981;52(6): Sunness JS, Bressler NM, Tian Y, Alexander J, Applegate CA. Measuring geographic atrophy in advanced age-related macular degeneration. Invest Ophthalmol Vis Sci. 1999;40(8): Maguire P, Vine AK. Geographic atrophy of the retinal pigment epithelium. Am J Ophthalmol. 1986;102(5): Sarks JP, Sarks SH, Killingsworth MC. Evolution of geographic atrophy of the retinal pigment epithelium. Eye (Lond). 1988;2 (Pt 5): Sunness JS, Gonzalez-Baron J, Applegate CA, et al. Enlargement of atrophy and visual acuity loss in the geographic atrophy form of agerelated macular degeneration. Ophthalmology. 1999;106(9): Sunness JS, Bressler NM, Maguire MG. Scanning laser ophthalmoscopic analysis of the pattern of visual loss in age-related geographic atrophy of the macula. Am J Ophthalmol. 1995;119(2): Klein R, Klein BE, Franke T. The relationship of cardiovascular disease and its risk factors to age-related maculopathy. The Beaver Dam Eye Study. Ophthalmology. 1993;100(3): Vingerling JR, Dielemans I, Hofman A, et al. The prevalence of age-related maculopathy in the Rotterdam Study. Ophthalmology. 1995;102(2): Hirvelä H, Luukinen H, Läärä E, Sc L, Laatikainen L. Risk factors of age-related maculopathy in a population 70 years of age or older. Ophthalmology. 1996;103(6): Yehoshua Z, Rosenfeld PJ, Gregori G, et al. Progression of geographic atrophy in age related macular degeneration imaged with spectral domain optical coherence tomography. Ophthalmology. 2011;118(4): Lujan BJ, Rosenfeld PJ, Gregori G, et al. Spectral domain optical coherence tomographic imaging of geographic atrophy. Ophthalmic Surg Lasers Imaging. 2009;40(2): Bearelly S, Chau FY, Koreishi A, Stinnett SS, Izatt JA, Toth CA. Spectral domain optical coherence tomography imaging of geographic atrophy margins. Ophthalmology. 2009;116(9): Fleckenstein M, Charbel Issa P, Helb HM, et al. High-resolution spectral domain-oct imaging in geographic atrophy associated with age-related macular degeneration. Invest Ophthalmol Vis Sci. 2008;49(9): Fleckenstein M, Schmitz-Valckenberg S, Adrion C, et al. Tracking progression with spectral-domain optical coherence tomography in geographic atrophy caused by age-related macular degeneration. Invest Ophthalmol Vis Sci. 2010;51(8): Helb HM, Charbel Issa P, Fleckenstein M, et al. Clinical evaluation of simultaneous confocal scanning laser ophthalmoscopy imaging combined with high-resolution, spectral-domain optical coherence tomography. Acta Ophthalmol. 2010;88(8): Fleckenstein M, Schmitz-Valckenberg S, Martens C, et al. Fundus autofluorescence and spectral-domain optical coherence tomography characteristics in a rapidly progressing form of geographic atrophy. Invest Ophthalmol Vis Sci. 2011;52(6): Schmitz-Valckenberg S, Fleckenstein M, Göbel AP, Hohman TC, Holz FG. Optical coherence tomography and autofluorescence findings in areas with geographic atrophy due to age-related macular degeneration. Invest Ophthalmol Vis Sci. 2011;52(1): Sayegh RG, Simader C, Scheschy U, et al. A systematic comparison of spectral-domain optical coherence tomography and fundus autofluorescence in patients with geographic atrophy. Ophthalmology. 2011;118(9): Jiao S, Knighton RW, Huang X, Gregori G, Puliafito C. Simultaneous acquisition of sectional and fundus ophthalmic images with spectraldomain optical coherence tomography. Opt Express. 2005;13(2): Wojtkowski M, Srinivasan V, Fujimoto JG, et al. Three-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence tomography. Ophthalmology. 2005;112(10): Drexler W, Sattmann H, Hermann B, et al. Enhanced visualization of macular pathology with the use of ultrahigh-resolution optical coherence tomography. Arch Ophthalmol. 2003;121(5): Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1(8476): Schutze C, Ahlers C, Sacu S, et al. Performance of OCT segmentation procedures to assess morphology and extension in geographic atrophy. Acta Ophthalmol. 2011;89(3): Li Y, Gregori G, Lam BL, Rosenfeld PJ. Automatic montage of SD- OCT data sets. Opt Express. 2011;19(27): Ophthalmic Surgery, Lasers & Imaging Retina Healio.com/OSLIRetina