Portal de Periódicos da CAPES

Drusen detection in retro-mode imaging by a scanning laser ophthalmoscope

2011; Wiley; Volume: 89; Issue: 5 Linguagem: Inglês

10.1111/j.1755-3768.2011.02123.x

1755-3768

Jennifer H. Acton, Robert P. Cubbidge, Hélen Dean King, Paul Galsworthy, Jonathan Gibson,

Glaucoma and retinal disorders

Purpose: The Nidek F-10 is a scanning laser ophthalmoscope that is capable of a novel fundus imaging technique, so-called 'retro-mode' imaging. The standard method of imaging drusen in age-related macular degeneration (AMD) is by fundus photography. The aim of the study was to assess drusen quantification using retro-mode imaging. Methods: Stereoscopic fundus photographs and retro-mode images were captured in 31 eyes of 20 patients with varying stages of AMD. Two experienced masked retinal graders independently assessed images for the number and size of drusen, using purpose-designed software. Drusen were further assessed in a subset of eight patients using optical coherence tomography (OCT) imaging. Results: Drusen observed by fundus photography (mean 33.5) were significantly fewer in number than subretinal deposits seen in retro-mode (mean 81.6; p < 0.001). The predominant deposit diameter was on average 5 μm smaller in retro-mode imaging than in fundus photography (p = 0.004). Agreement between graders for both types of imaging was substantial for number of deposits (weighted κ = 0.69) and moderate for size of deposits (weighted κ = 0.42). Retro-mode deposits corresponded to drusen on OCT imaging in all eight patients. Conclusion: The subretinal deposits detected by retro-mode imaging were consistent with the appearance of drusen on OCT imaging; however, a larger longitudinal study would be required to confirm this finding. Retro-mode imaging detected significantly more deposits than conventional colour fundus photography. Retro-mode imaging provides a rapid non-invasive technique, useful in monitoring subtle changes and progression of AMD, which may be useful in monitoring the response of drusen to future therapeutic interventions. Drusen are early fundus changes characteristic of age-related maculopathy, and early detection is likely to be increasingly important in view of future possible interventions. Conventional fundus photography is widely used in imaging and detecting drusen, and although nonmydriatic cameras exist, pupillary dilation is often necessary in elderly patients with small pupils and lenticular opacities. Optical coherence tomography (OCT) is another commonly used non-invasive imaging technique in which retinal structures may be viewed in cross section. Analogous to B-scan ultrasonography, where light rather than sound measurements are made from tissue boundaries, drusen are visible in OCT images as elevations of the retinal pigment epithelium (RPE) (Freeman et al. 2010). RPE atrophy (Pieroni et al. 2006), intraretinal fluid, pigment epithelial detachment and neovascular membranes (Ahlers et al. 2006) may also be detected by OCT. The use of infrared imaging is advantageous in elderly patients with lens opacities, since light is minimally scattered in the presence of media opacities (Elsner et al. 1996). The deeper penetration of infrared light compared to visible wavelengths allows for easier observation of subretinal structures such as drusen and choroidal neovascularization (Hartnett & Elsner 1996). Retro-mode imaging, as implemented by the F-10 (Nidek, Gamagori, Japan) confocal scanning laser ophthalmoscope (SLO), images the retina with an infrared laser (Fig. 1). It was based on and is very similar to the indirect mode imaging (Elsner et al. 1996) of a SLO, because scattered light is collected in both. Both imaging methods employ an aperture with a central stop, where indirect mode uses a large annular aperture and retro-mode only uses part of the annular aperture and thus deviates laterally from the confocal light path (Fig. 2). Directly reflected light from the fundus is blocked by the central stop, and only the scattered light passes through the aperture (Webb et al. 1987). In this way, more laterally scattered light is sampled than in the direct mode (Elsner et al. 1996). Only retinal structures that scatter incident light laterally are detected, thus highlighting features such as drusen which act as an efficient source of light scatter (Wormington 2003). Indirect mode has demonstrated enhanced imaging of macular drusen in a small number of patients (Manivannan et al. 1994; Hartnett & Elsner 1996), an area of elevation in choroidal neovascularization (Hartnett & Elsner 1996), easy detection of detachment of the neuroretina in central serous retinopathy (Remky et al. 1998) and the emphasized appearance of borders in cystoid macular oedema (Remky et al. 1999). Retro-mode imaging, which only employs laterally scattered light from one direction, has been used to investigate macular pathologies. In cystoid macular oedema secondary to polypoidal choroidal vasculopathy, cystoid spaces detected by OCT imaging were detected by retro-mode but were not visible on fundus photography (Yamamoto et al. 2009). In myopic eyes with macular retinoschisis, a 'fingerprint' pattern was demonstrated by retro-mode imaging (Tanaka et al. 2010). Optical diagram of F-10, a confocal scanning laser ophthalmoscope. Scanning laser ophthalmoscope (SLO) Apertures (top). The confocal aperture width can be varied from 1.5 to 7 mm. The smallest aperture is pinhole-like and produces a high-contrast image, whereas the wider aperture collects light from a larger area, but produces a lower contrast image. Scattered light is collected by the annular aperture, and reflected light is blocked by the central stop. The aperture in retro-mode incorporates the central stop in indirect mode, but also blocks light from either the left or right side, which creates the shadow to one side of the abnormal feature, such as drusen. Retro-Mode Imaging (bottom). Directly reflected light and scattered light from one side of the drusen are blocked. The right side of the drusen appears lighter in the image, when the aperture is in this position. The aim of this study was to compare images captured using the Nidek F-10 SLO in retro-mode with standard fundus photography to quantify macular drusen. The nature of the retro-mode images was demonstrated previously in a small pilot study (Gibson et al. 2009). Additionally, in eight subjects, OCT examination of the subretinal deposits was performed with the purpose of a structural observation of corresponding deposits on fundus photography and retro-mode imaging. Patients were recruited at various stages of age-related maculopathy and age-related macular degeneration (AMD). Based on pilot drusen counts in retro-mode images and fundus photography images, a sample of 19 would give 95% power in detecting a difference of 40 drusen. Exclusions were made where there was any history of other ocular disease or diabetes. Eyes with ocular trauma or eyes which had been treated for AMD were excluded. Only eyes with observable drusen and with minimal or no lenticular opacities were included. Ethical approval was obtained from the Aston University, Human Sciences Ethical Committee, and the tenets of the Declaration of Helsinki were followed. Informed consent was obtained from each patient including detailed explanations of all procedures before participation. Thirty-one eyes of 20 patients were eligible for the study and ranged in age from 48 to 79 years (mean 67, SD 6.7), 16 patients were women and 4 were men. A subset of the patient group, 13 eyes of 8 patients, attended 6 months (mean 185.8 days, SD 37.4) later for further images, using the F-10 (Nidek) and RS-3000 OCT (Nidek). All imaging was performed on patients with dilated pupils, to optimize image quality. Fundus photography was carried out with a nonmydriatic camera (EOS 10D camera, 6.3 megapixels; Canon Inc., Tokyo, Japan) to acquire stereoscopic pairs of digital images and 30° field images centred at the macula were stored as high-quality jpeg files (large/fine, approximately 2.4 MB, 3072 × 2048); 20° (horizontal dimension, or 40° field of view) images were captured in retro-mode using the F-10 and stored as high-quality jpeg files. The F-10 (Nidek) is a confocal SLO (Fig. 1), which non-invasively scans the fundus with a Class 1 laser. Different modes of imaging include reflectance modes for various wavelengths (490, 532, 660 and 790 nm), fluorescein angiography, indocyanine green angiography and retro-mode imaging (790 nm; Fig. 3). It captures images with a field of view of 40° or 60°, an optical resolution of 16–20 μm and image size of up to 1024 × 720 pixels. The image size setting for macular images was 800 × 600 pixels. Retro-mode imaging employs an aperture with a modified central stop, which is deviated laterally from the confocal light path, and may be positioned to the left or right side of the fundus (Fig. 2). The scattered light passing through the deviated aperture gives a shadow to features such as subretinal deposits, thus enhancing the contrast and delineation of the features (Fig. 3). Fundus photograph (top left), F-10 retro-mode image (bottom left) and F-10 reflectance images for infrared 790 nm (top right) and green 532 nm (bottom right), acquired in a patient with macular drusen. Optical coherence tomography images were acquired using the RS-3000, a spectral domain OCT and SLO infrared imaging system. The axial and transverse resolutions are 7 and 20 μm, respectively, and an automated tracking function is employed during scanning. The fundus was imaged with the 'macular map' scan, in which the appropriate macular field was selected, which was approximately the central 3000 μm in each patient, to acquire an image cube of 64–128 B-scans. The images were manipulated using commercially available software (Photoshop 8.0; Adobe Systems Inc., San Jose, CA, USA) such that the retinal positions of the OCT scans as shown by the infrared images were matched to the retro-mode images using the flicker on and off and transparency features. Fluorescein angiography was not undertaken in this study, as it was considered unethical to expose the volunteer patients who were all asymptomatic, to an invasive procedure with potential risk. Fundus photography images were graded in a random order for the retinal features according to the International Classification and Grading System for ARM and AMD (Bird et al. 1995). Stage of disease was determined according to the stages of severity defined by an epidemiologic study, based on progression rates of features over a 6.5-year period (van Leeuwen et al. 2003). Grading and stage determination was carried out by two independent, masked graders (JMG & JHA). The colour fundus photographs were graded using custom software, written in Liberty BASIC (Shoptalk Systems, Framingham, MA, USA) for number of drusen and predominant drusen diameter within the central 3000 μm circle, by two masked graders (HK & PG), experienced in grading AMD and diabetic retinopathy. The software allows for mapping of the circular grading grid onto the fundus image, manual marking of drusen and a measurement tool. The subretinal deposits visible on the retro-images were assessed in the same way. For the purpose of this study, the term drusen refers to the deposits visible on colour fundus photography. The term subretinal deposits refers to the elevations visible in retro-mode images, even though some may coincide with drusen on colour fundus photography. Pairs of retro-mode images from the same imaging session were graded for drusen size and number and according to parameters of image quality (Strauss et al. 2007), for one eye of each patient. Where data variables had non-Gaussian distributions, they were transformed to achieve normal distributions. Paired t tests were performed to assess the differences between imaging methods and within-session retro-mode image reproducibility. Drusen size and drusen number data were converted to ordinal data according to the definitions specified in the International Classification and Grading System (Bird et al. 1995) to calculate a weighted kappa statistic to measure agreement between graders. Bland–Altman plots were constructed for comparisons between imaging methods and graders (Bland & Altman 1986). A one-way anova was performed to examine the effect of stage of AMD on the normalized dependent variables. Classification of stage of disease resulted in 9 eyes at stage 0, 10 eyes at stage 1, 4 eyes at stage 2, 2 eyes at stage 3 and 6 eyes at stage 4 of disease. Drusen were present in all eyes, 25 eyes had drusen and pigmentary changes only, 4 eyes exhibited geographic atrophy and 2 eyes had signs of choroidal neovascularization. The data variable, number of drusen, was square root transformed, and a logarithmic transformation was applied for size of drusen, to achieve normal distributions. This was confirmed by the Kolmogorov–Smirnov test, p = 0.555 and p = 0.663, for drusen number and drusen size, respectively. Between stages of AMD, the variance of the transformed data was homogeneous for number of drusen (Levene's statistic = 0.663, p = 0.619) but not for size of drusen (Levene's statistic = 4.437, p = 0.002). The effect of stage on drusen number and size in colour fundus images was assessed. A significant (one-way anova: F = 5.913, p < 0.001) variation of number of drusen (square root) was noted when comparing by stage of severity of AMD. A significant variation of size of drusen (log base 10) was found when comparing by stage of severity of AMD (F = 11.211, p < 0.001). Table 1 shows the mean number and predominant drusen diameter graded for colour fundus photography and retro-mode imaging. There were significant differences between imaging methods for number (paired t test: t = −9.314, df = 61, p < 0.001) and size of drusen (t = 3.009, df = 53, p = 0.004). A consistently greater number of deposits were detected by the retro-mode images, and the mean difference was 48 deposits (Fig. 4A). As a percentage of the number of drusen on colour fundus images, retro-mode deposits were increased by a mean of 413% compared to colour fundus images. Drusen were graded slightly larger in fundus photography, by a mean difference of 5 μm (Fig. 4B). As a percentage of drusen size on colour fundus images, retro-mode deposits were on average 8% smaller than colour fundus drusen. Greater levels of agreement were seen when fewer drusen were present, whereas there was no clear relationship between the magnitude of drusen diameter and agreement between imaging modes (Fig. 4A,B). Overall, deposits visible on retro-mode images but not on colour fundus images were small in diameter, <63 μm. Bland–Altman plots. The black line indicates the mean difference, and the grey lines represent the 95% limits of agreement. Plots represent the difference in drusen number counted on fundus photography and retro-mode images (A), the difference in drusen size between imaging types (B), the difference between drusen number counted by grader 1 and grader 2 (C) and the difference between drusen size measured by grader 1 and grader 2 (D). Although there was a significant difference between graders for number (p < 0.001) and size (p < 0.001) of drusen, the agreement between graders when applying the categories used in the International Classification and Grading System was substantial for number of deposits (weighted kappa value 0.69, standard error 0.09) and moderate for size of deposits (weighted kappa value 0.42, standard error 0.08), for both imaging types. Grader 2 (PG) tended to grade fewer and larger drusen than grader 1 (HK). For drusen number, the difference was greater when there were more than 100 drusen (Fig. 4C), and there was no clear bias in the difference between graders for size of drusen (Fig. 4D). Overall small deposits less than approximately 40 μm were not evident in OCT images, which may be explained by the presence of mild lenticular opacities or unstable fixation. Larger drusen, identified as being present in both fundus photography and retro-mode images, manifested on OCT scans as displacement of the RPE (Fig. 5). The large subretinal deposits in retro-mode imaging were consistent with the appearance of drusen on OCT imaging (Fig. 5; solid arrow) in all eight patients. Large deposits that were present in retro-mode images but not present in colour fundus images manifested in OCT images as a displacement of the RPE (Fig. 5; outline arrow). However, not all large retro-mode deposits were consistently detected in OCT images. Fundus photograph, retro-mode image and optical coherence tomography (OCT) images. Solid arrow indicates drusen visible in all three imaging types. Outline arrow indicates retro-mode deposit not visible on fundus photograph, but visible on OCT scan, whose displacement of the RPE is consistent with the appearance of drusen. Comparison between the second and first set of retro-mode images revealed appreciable change in four of 13 eyes where two of the images showed enlargement of deposits and two images showed increased confluence of deposits (Fig. 6), where no change over time on fundus photography was evident. Change in retro-mode images over time. The image on the right was captured 6 months after the left. Solid arrows indicate a deposit which enlarged. Outline arrows indicate formation of new deposits in the image on the right. A comparison between pairs of retro-mode images from the same imaging session was made by one observer (JHA). No significant differences were found between retro-mode image pairs for drusen number (paired t test: t = −0.509, df = 19, p = 0.617) and drusen size (t = 1.327, df = 19, p = 0.200). There were no significant differences between image pairs from the same session for the following graded parameters of image quality: overall image quality (t = 0.213, df = 19, p= 0.834), contrast (t = 0.175, df = 19, p = 0.863), sharpness (t = 0.438, df = 19, p = 0.666), resolution (t = 1.453, df = 19, p = 0.163), noise (t = 0.547, df = 19, p = 0.591) and artefacts (t = −0.195, df = 19, p = 0.0.847), In this study, the comparison between drusen quantification on retro-mode images and digital fundus photographs was investigated. The images taken using the F-10 in retro-mode show a pseudo-three-dimensional appearance to drusen, which is consistent with the enhanced imaging of drusen using infrared SLO in indirect mode (Manivannan et al. 1994; Hartnett & Elsner 1996). Large subretinal deposits visible in the retro-mode images appear to be consistent with the appearance of retinal drusen on OCT imaging; however, the smaller subretinal deposits less than approximately 40 μm in diameter, which were present in large numbers, were not apparent on OCT imaging. The retro-mode deposits appeared in significantly greater numbers than the drusen in the same area in fundus photography. This is in agreement with observation of a single-patient case study of macular drusen (Manivannan et al. 1994), and a study of ten patients with exudative AMD in which drusen were manually counted using acetate overlays (Hartnett & Elsner 1996). Conversely, no significant difference between drusen area in colour fundus slides and indirect SLO slides was noted when manually marking drusen onto an acetate sheet over slides of five eyes of six patients (Kirkpatrick et al. 1995). Drusen are deposits of extracellular material which lie between the RPE and Bruch's membrane. Histologically, drusen smaller than 25–30 μm, the diameter of two RPE cells, are not clinically detectable (Sarks et al. 1999). Various types of preclinical drusen, <25 μm in diameter, have been identified histologically, which occur in the formation of clinically detectable drusen. In eyes that have small numbers of drusen, these may be the entrapment of coated membrane bodies between the RPE and Bruch's membrane. In eyes with many hard drusen, preclinical drusen are small focal plaques of thickened hyalinized Bruch's membrane or microdrusen, 1–2 μm in diameter, composed of dense amorphous material (Sarks et al. 1999). Microdrusen occur discretely or in rows and were noted to be a frequent occurrence in eyes with many drusen and may have a role in soft drusen formation (Sarks et al. 1999). It has been suggested that the subretinal deposits seen in indirect SLO mode, which do not correspond to drusen in images derived by colour fundus photography, are either identical to drusen or material under the RPE or within Bruch's membrane (Hartnett & Elsner 1996). Because the resolution of the F-10 is 16–20 μm, drusen smaller than 25 μm were not graded in this study. For all image pairs, of retro-mode and colour fundus images, subretinal deposit number was greater than drusen number, with the exception of two image pairs graded by grader 1(HK), where there was a difference of <7. In all images where subretinal deposits in retro-mode were more numerous, the subretinal deposits which did not correspond to drusen in the colour fundus photographs did not have a different appearance to the deposits which did correspond to drusen. The most commonly used grading scales for AMD (Klein et al. 1991; Bird et al. 1995; The Age-Related Eye Disease Study Research Group 2001) use the circle C0 that has a diameter of 63 μm, as the smallest increment for grading drusen size. Comparison between grading of drusen size for the imaging methods shows good agreement because 87% of diameter differences in grading were <63 μm and the 95% limits of agreement were 95.9 μm (Fig. 4). Inter-grader agreement was clinically acceptable and 74% of diameter differences in grading between graders were <63 μm, despite only moderate level of agreement calculated between graders for predominant drusen size (weighted κ value 0.42). Graders showed substantial agreement for number of drusen (weighted κ value 0.69). The shadow effect to subretinal deposits imaged in retro-mode was observed to create light or dark deposit edges. This differed between images captured by the apertures in opposite positions. Where a high-contrast edge was seen on a particular drusen edge, a low-contrast border would be seen at the same edge when imaged with the opposite aperture. In this way, the retro-mode imaging provides directional information and would be expected to better enhance subtle features such as low-lying drusen than infrared SLO indirect imaging. Furthermore, for drusen which are circular and uniformly elevated, the directional effect does not influence the detection capability, but may provide useful spatial information about the nature of deposits that are not uniformly shaped. Retro-mode image quality was observed to improve with accurate focusing and contrast adjustment of the instrument, as well as stability of patient fixation. The reproducibility of retro-mode images captured within the same imaging system gave consistent results in terms of drusen grading and image quality parameter grading. However, this assessment of reproducibility is limited by the variability of subjective grading and would be improved by automated image grading. Supplementary observations regarding the nature of the subretinal deposits were the variations in edge sharpness, which combined with the variations in edge contrast, and may affect automated drusen segmentation of retro-mode images. A previous study attempted computerized automated quantification of drusen area, of indirect SLO images, but this technique was found to have a low sensitivity of 35% (Kirkpatrick et al. 1995). Automated drusen detection was confounded in the study by Kirkpatrick et al. (1995), because it was deemed inappropriate to apply a simple thresholding image processing technique to both light and dark edges, and the edge-detection method lacked specificity because of the large variability of the sharpness of the drusen edges (Kirkpatrick et al. 1995). For images where there were many small subretinal deposits at the image edge, it was noted that the image quality in retro-mode was not uniform across the entire image, even after careful focusing of the instrument. The subretinal deposits appeared to fade or blur at the periphery of the retro-mode images, which was outside the grading area used in this study. This was further investigated by the capture of images at different fundus areas, using different points of fixation. It was noted that the corresponding drusen that appeared faded in the periphery of one image were clearly imaged when central in another image. Increased subretinal deposits in the area temporal to the optic disc were also observed in two patients. The peripheral image degradation could be explained by the retinal curvature and increased distance from the light source causing the brightness profile of the image to be reduced. Peripheral image degradation to subretinal deposits in the 20° images in this study occurred on average in the peripheral 3° outside an approximately horizontal elliptical area of good image quality. Therefore, the peripheral features of a retro-mode image, in this area, should be ignored because useful clinical information can only be obtained from the central 17° of a 20° image. It is recommended that further or multiple images be captured where the features of interest lie central to the image. Patients with small pupils and lenticular opacities were excluded from the study, to optimize the retro-mode image quality. Small pupil size owing to a lack of retinal illumination causes poor contrast of the image. Mild lenticular opacities, where a degradation to colour fundus images was visible, had no observable detrimental effect on retro-mode images. A disadvantage of retro-mode imaging is that retinal features such as small vessels are not easily observable. The results demonstrate that significantly more subretinal deposits, consistent with the OCT appearance of drusen, are detected in retro-mode imaging than in standard fundus photography. However, a limitation of the study is that only a subset of eyes were imaged with OCT as well as retro-mode imaging. Comparison of retro-mode images over time revealed the enlargement and confluence of the deposits in a small number of patients (Fig. 6). Alteration to the appearance of drusen has not previously been observed over such a short period. A recent study noted that apparent regression of drusen may lead to misclassification in clinical trials evaluating the efficacy of pharmacological intervention (Sallo et al. 2009). A larger scale longitudinal follow-up of patients is required to confirm whether these subretinal deposits are sub-clinical drusen which will eventually manifest as clinically visible drusen or other features of disease. Retro-mode imaging provides a rapid non-invasive technique which is useful in the monitoring of subtle drusen changes in the progression of AMD and may have the ability to detect changes in drusen over much shorter periods of time than previously found. This may have implications for the conduct of future clinical trials that intend to use drusen detection as an end-point. The authors acknowledge Birmingham Optical Group for use of the Nidek F-10 and RS-3000 OCT. We thank the reviewers for their constructive comments.