ABCA4-associated retinal degenerations spare structure and function of the human parapapillary retina

Abstract

Purpose

To study the parapapillary retinal region in patients with ABCA4-associated retinal degenerations.

Methods

Patients with Stargardt disease or cone-rod dystrophy and disease-causing variants in the ABCA4 gene were included. Fixation location was determined under fundus visualization and central cone-mediated vision was measured. Intensity and texture abnormalities of autofluorescence (AF) images were quantified. Parapapillary retina of an eye donor with ungenotyped Stargardt disease was examined microscopically.

Results

AF images ranged from normal, to spatially homogenous abnormal increase of intensity, to spatially heterogenous speckled pattern, to variably sized patches of low intensity. A parapapillary ring of normal-appearing AF was visible at all disease stages. Quantitative analysis of the intensity and texture properties of AF images showed the preserved region to be an annulus, at least 0.6 mm wide, surrounding the optic nerve head. A similar region of relatively preserved photoreceptor nuclei was apparent in the donor retina. In patients with foveal fixation, there was better cone sensitivity at a parapapillary locus in the nasal retina compared to the same eccentricity in the temporal retina. In patients with eccentric fixation, ~30% had a preferred retinal locus in the parapapillary retina.

Conclusions

Human retinal degenerations caused by ABCA4 mutations spare the structure of retina and RPE in a circular parapapillary region which commonly serves as the preferred fixation locus when central vision is lost. The retina between fovea and optic nerve head could serve as a convenient, accessible and informative region for structural and functional studies to determine natural history or outcome of therapy in ABCA4-associated disease.

ABCA4 gene encodes the ABCR protein (also known as Rim protein) localized to rod and cone photoreceptor outer segments^1–5. Evidence to date supports the involvement of ABCR in the active transport of a complex of all-trans-retinal and phosphatidylethanolamine^6–8. Mutations in ABCA4 cause autosomal recessive forms of retinal degeneration (RD) that are associated with varied fundus appearance, wide-ranging disease severity and a diverse set of clinical diagnoses^9–17. The single common phenotypic feature reported in the vast majority of patients with ABCA4-associated RD (ABCA4-RD) is the involvement of the macula^e.g.2,9–17 with very rare exceptions¹³.

The detailed natural history of spatio-temporal disease progression in ABCA4-RD is not known. Based on the predominant involvement of the macula in ABCA4-RD and effects on the peripheral retina in a subset of those patients, it is parsimonious to hypothesize a generalized central to peripheral (centrifugal) expansion of the retinal degeneration with age in most patients. Studies in Stargardt disease (STGD) of unknown genotype have commonly provided evidence for a centrifugal expansion of the degeneration^18–21 sometimes sparing the fovea or the foveola^21–23. Stages of ABCA4-RD disease observed in cross-sectional studies have been generally consistent with the centrifugal expansion hypothesis^11,15–17. Some reports in STGD, however, have shown a retinal region encircling the optic nerve which may be protected from retinal degeneration and forms an island of exception to the centrifugal expansion hypothesis^20,24–28. In the current work, we use autofluorescence (AF) imaging in the vicinity of the parapapillary retina to examine whether, and under what conditions, this region is spared from retinal degeneration in ABCA4-RD, and whether such sparing has implications in terms of visual function.

Methods

Patients

The study population was a subset (n=41; those who had fixation testing) of patients recently reported¹⁶. They had clinical diagnoses of STGD¹¹ or cone-rod dystrophy (CRD)¹⁵ and one or more changes in the ABCA4 gene considered to be disease-causing variants. Patient numbers in the current work refer exactly to those in the previous publication¹⁶. Research procedures were in accordance with institutional guidelines and the Declaration of Helsinki. All patients gave written informed consent.

Visual function studies

A complete eye examination was performed in all subjects including Goldmann kinetic perimetry with V-4e and I-4e targets. Preferred locus of fixation was determined with optical coherence tomography (OCT, Carl Zeiss Meditec, Dublin, CA). Specifically, after alignment of the eye with the OCT instrument, patients were asked to fixate the “landmark” spot (bright red light) and a 4.5 mm long scan was moved until the cross-section of the anatomical fovea could be identified at the center of the scan. Then, an OCT scan and a corresponding fundus image were obtained for documentation and fixation location was quantified with respect to the location of the anatomical fovea.

In all patients with documented stable fixation at the anatomical fovea, horizontal profiles of cone sensitivity were obtained with orange (600 nm; 200 ms; 1.7° diameter) stimuli presented on a white (2.7 log phot-td) background on a modified²⁹ computerized static perimeter (HFA, Zeiss Humphrey Systems, Dublin, CA).

AF imaging and analyses

Imaging was done with a confocal scanning laser ophthalmoscope (HRA, Heidelberg Engineering, Dossenheim, Germany) as previously described¹⁶. AF images were obtained with 488 nm excitation and >500 nm emission. All images were acquired with a lateral magnification wherein a 30°×30° square field was sampled onto 512×512 pixels. Custom-written software (MATLAB 6.5, Natick, MA) was used to first transform each file containing series of consecutive images of AF intensity into a stack of 8-bit raw images. Frames with blinks or mid-frame eye movements were discarded, and remaining frames were spatially transformed to correct for imaging-system derived distortions. This corrected stack was loaded into an image processing program (ImageJ, ver 1.34n, available in the public domain at http://rsb.info.nih.gov/ij). The images in the stack were automatically registered using a pair of programs (turboreg and stackreg³⁰, available in the public domain at http://bigwww.epfl.ch/algorithms.html). Mean value of the AF intensity was calculated at each pixel after registration. A wide field image montage was assembled from 6–8 images by manually specifying retinal landmark pairs corresponding to each other in overlapping segments using custom-written software (MATLAB 6.5). Images from left eyes were transformed into equivalent right eyes and further transformed to register the anatomical fovea and the center of the optic nerve head (ONH) to predetermined locations based on mean normal results. The intensity of each image was normalized by the mean intensity at the center of the ONH. The resulting intensities were mapped to a custom pseudocolor scale which was slightly modified from a previously published version¹⁶.

In order to quantify the local heterogeneity of the AF intensity, run-length^31–33 was calculated for a fixed criterion in eight principal directions for each pixel. The mean run-length of a pixel represents the size of a local region showing homogeneity of intensity. Further details of this method are in Appendix 1. All intensity and run-length images were transformed into pseudo-profiles using semi-polar integral analysis in the neighborhood of the ONH. Further details of this method are in Appendix 2.

Histopathology

A sample of the right retina of a previously published³⁴ eye donor (Foundation Fighting Blindness, FFB#219) was available for study. In brief, the patient was a 62-year-old woman who died of lung cancer. At the time of death, the donor had advanced macular degeneration from STGD but without a known molecular cause. The eye was enucleated at 4 hours and 20 minutes postmortem and was slit at the pars plana. Ten minutes later, it was placed into fixative³⁴. The tissue sample containing the nasal half of the optic nerve head was embedded without osmication in glycol methacrylate (JB-4; Polysciences, Wilmington, DE), sectioned at 2 μm, and stained with Richardson’s methylene blue/azure II mixture. Sections were examined with a microscope (Leica DMR, Deerfield, IL) and photographed with Kodak EliteCHROME ASA 400 film with a calibration slide.

Results

Accumulated lipofuscin in RPE cells is the dominant fluorophore that contributes to the topographic variation of AF intensity across the human retina upon excitation with short wavelength light^16,35–40. Features of AF intensity distribution across the central retina in normal eyes have been described^39,41. A representative normal subject shows a deep trough of AF signal at the fovea (Fig. 1) corresponding to the absorption of the excitation light by macular pigment; and, there is loss of signal at the retinal blood vessels corresponding to the absorption by blood. The AF signal originating from the normal ONH is below the level of detection. The peak of AF intensity forms an approximate circle at an eccentricity of ~3 mm corresponding to the highest density of rod photoreceptors in the human retina^35,41,42.

Figure 1.

Standardized images of autofluorescence in a representative normal subject (A) and three patients (B-D) with ABCA4-associated retinal degeneration (ABCA4-RD). Excitation wavelength was 488 nm. Intensities are mapped to a pseudocolor scale shown; results of P42 were uniformly scaled (X 0.7) to fit the 256-level dynamic range. Black represents no image data and purple corresponds to the dark-level of the detector. All eyes are shown as equivalent right eyes.

An early sign of ABCA4-disease detectable non-invasively is the abnormality of AF intensity^{14,16,28,43,44}. Illustrating an early ABCA4-disease stage is Patient #42 (P42, see ref. 16), a 23 year-old man clinically diagnosed¹¹ with STGD phenotype I and molecularly diagnosed with a G1961E mutation in the ABCA4 gene. He had best corrected visual acuities of 20/40 and 20/100 in the right and left eyes, respectively. Both eyes showed a full extent of Goldmann kinetic visual field with the V-4e and I-4e test targets but a relative central scotoma with the I4e target. AF imaging of his left eye shows a dramatic increase of intensity across the central retina (Fig. 1B, displayed as equivalent right eye). The abnormal intensity is spatially homogeneous throughout most of the central retina except near the fovea where there are small regions of AF intensity loss with resulting local spatial heterogeneity. Notable is a region of normal-appearing AF intensity encircling the ONH. A more advanced disease stage of ABCA4-RD is illustrated by P46, a 48 year-old woman clinically diagnosed with STGD phenotype II and molecularly diagnosed with a R2030Q mutation in the ABCA4 gene. She had best corrected visual acuities of 20/200 and 20/20 in the right and left eyes, respectively, and full peripheral extent of kinetic fields. Centrally, there was a small island of retained function (with the V-4e target) surrounded by an absolute scotoma and a larger relative scotoma. AF imaging of her right eye shows intensities distributed mostly within the normal range but there is abnormal spatial heterogeneity (Fig. 1C) with a speckled pattern. This pattern has been hypothesized to correspond to microscopic variation in the rates of abnormal lipofuscin accumulation or to the patchy loss of a subset of RPE cells or to the reduction of OS shedding as the overlying retina degenerates¹⁶. Two regions of presumed RPE atrophy are also apparent as patches of low intensity in the superior and inferior parafovea. Of note, there is a spatially homogenous circular region of normal-appearing intensity at the parapapillary retina. A severe disease stage of ABCA4-RD is represented by P37, a 41 year-old man clinically diagnosed with STGD phenotype II and molecularly diagnosed with a IVS40+5G>A mutation in the ABCA4 gene. He had best corrected visual acuities of 20/400 and 20/200 in right and left eyes, respectively. Kinetic perimetry with the V-4e target showed superior field limitation but was otherwise full in peripheral extent; an absolute central scotoma was present in each eye. There was no detection of the I-4e target. AF imaging of his right eye shows extensive regions of presumed RPE atrophy across the posterior pole with small intervening regions of detectable autofluorescence probably originating from retained RPE. A normal-appearing parapapillary ring is visible (Fig. 1D).

To confirm and extend the visual impressions of parapapillary preservation, we quantified AF results in ABCA4-RD patients and compared them to a group of normal subjects. Distribution of lipofuscin accumulation was estimated from AF intensities. Spatial heterogeneity of lipofuscin accumulation was derived using run-length analysis where mean run-length of a pixel represents the size of a local region showing homogeneity of intensity. Intensity and run-length images centered on the ONH were vertically bisected to form nasal and temporal halves, transformed from rectangular to polar coordinates, and integrated along the angular dimension to produce pseudo-profiles. Data from four patients overlaid on normal limits represent the spectrum of results observed (Fig. 2). P20 represents a normal parapapillary region and the surrounding area: both the intensity as well as local homogeneity of the AF imaging within 2.5 mm of the center of the ONH fall within or near normal limits (Fig. 2A). P42 shows abnormally increased AF intensity in combination with nearly-normal mean run-length distributed across the parapapillary region and the surrounding area (Fig. 2B). P47 and P34 on the other hand, show normal or nearly normal intensity distributions associated with biphasic run-length plots showing normal results in a parapapillary region surrounded by an abnormally reduced run-length at greater eccentricities both in temporal and nasal retinas (Fig. 2C,D). It is important to note that regions of atrophy (such as in Fig. 2D) are masked (see Appendices 1, 2) and do not contribute to the pseudo-profile results.

Figure 2.

Detailed analysis of autofluorescence abnormalities in a region around the optic nerve head (ONH) in four ABCA4-RD patients (A–D). The contrast of each grayscale image is uniformly stretched for better visibility of features. Standardized image data from temporal (T) and nasal (N) semi-annular retinal regions from the edge of the ONH to an eccentricity of 2.5 mm (white circles overlaid on images) have been analyzed and shown as pseudo-profiles of intensity and run-length. Black lines represent data from patients and gray regions show the normal range.

ABCA4-RD patients were divided into two groups: those with normal AF intensity pseudo-profiles within or surrounding the parapapillary region (Fig. 3A, left panel) and those with abnormal results (Fig. 3A, right panel). The abnormalities were either temporal to the ONH or nasal to the ONH or both. Mean run-length pseudo-profiles ranged from normal to abnormal in both groups (Fig. 3B). Run-length abnormalities were dominated by numbers smaller than the mean normal values implying an increase in local heterogeneity of AF intensities consistent with that visually appreciated from the images. The extent of run-length abnormality increased as a function of distance from the center of the ONH (Fig. 3B). Similarly, the percent of patients showing normal run-length decreased as a function of distance from the ONH (Fig. 3C). Using 90% as a criterion, the region of preserved parapapillary retina extended to 1.5–1.6 mm from the center of the ONH; this corresponded to a ~0.6 mm (~2°) wide annular region (Fig. 3C, dashed lines) spared from the structural alterations of the RPE typically visible on AF imaging of patients with ABCA4-RD.

Figure 3.

Quantification of the extent of parapapillary preservation in ABCA4-RD patients. (A) Pseudo-profiles in the group of patients (left panel) showing normal autofluorescence intensity in the vicinity of the parapapillary region versus the remaining patients (right panel) showing abnormal results. (B) Pseudo-profiles of mean run-length in the two groups of patients as in panel A. (C) Percent of patients in each group showing normal run-length results as a function of eccentricity from the center of the optic nerve. Vertical dashed lines delineate the extents of parapapillary preservation defined as the eccentricity at which 90% of the patients show normal run-length. (D) Histopathology in a donor eye with Stargardt disease showing the parapapillary region nasal to the optic nerve head. Two regions (black rectangles) are shown at higher magnification as insets. Arrowheads point to photoreceptor nuclei, which are more numerous in the parapapillary region.

Histopathology results from a donor eye with STGD³⁴ (but without a known molecular diagnosis) allowed examination of retinal consequences of parapapillary sparing of the RPE, as visible on AF images. In the STGD eye, a zone immediately nasal to the optic nerve head showed 3–4 rows of retained photoreceptor nuclei with some inner segments and some very short outer segments (Fig. 3D). At further eccentricities, there was greater pathology in the photoreceptor layer which overlay disorganized or absent RPE. The zone of relatively spared retina extended ~0.7 mm from the nasal edge of the ONH and thus appeared to correspond to the RPE preservation observed with AF imaging methods.

In order to evaluate the visual consequences of a structurally-spared parapapillary region, we first considered psychophysical thresholds obtained in the central retinas of a subset of patients (8/41=19.5%) with documented foveal fixation. The majority of these patients showed parafoveal losses of cone sensitivity with foveal thresholds either within the normal range or mildly reduced (Fig. 4A). Differences in the depth and extent of the parafoveal defects represented the severity of the central retinal disease in these patients. Unexpected was the apparent asymmetry of the visual function defect between the nasal and temporal paramacular retinas. For example, at 3 mm (10°) eccentricity, mean L/M cone sensitivity was significantly better (mean difference=0.69 log, Student’s t-Test, P=0.05) in the nasal retina near the ONH than in the temporal retina (Fig. 4A).

Figure 4.

Visual function consequences of the preservation of parapapillary retina in ABCA4 disease. (A) Psychophysical cone sensitivity profiles in ABCA4-RD patients with documented foveal fixation. Two horizontal axes represent the standard perimetric coordinate system centered on the fovea (F) and an alternate coordinate system centered on the optic nerve head (ONH). Hatched region represents the expected location of the ONH; gray region delimits the normal range of cone sensitivity. (B) Graphical representation of retinal fixation loci in one eye of all patients (numbers designate individual patients) with central scotomas and extra-foveal fixation. Data are shown as right-eye equivalent on a schematic showing ONH (circle) and major retinal blood vessels (curved lines). Fixation locus of P30 was in the far supero-nasal retina in the direction pointed to by the arrow. (C) Autofluorescence images of both eyes of P32 with bilateral central atrophy of the RPE. Fixation loci determined individually in each eye are shown (white stars with black outline). The contrast of each grayscale image is uniformly stretched for better visibility of features.

In ABCA4-RD patients with central scotomas the preferred retinal locus of fixation was eccentric, as expected (Fig. 4B). Nineteen patients (46%) had a fixation locus in the superior retina. Ten patients (24% of all patients in the study, 30% of the subset of patients with central scotomas) demonstrated reproducible fixation loci in the parapapillary region (Fig. 4B). Six of these patients (P9, P18, P22, P33, P35, and P45) used their temporal parapapillary retina as the locus of fixation placing their vision between the central scotoma caused by the ABCA4 disease and the physiological blind spot caused by the ONH. Some patients fixated using parapapillary loci in both eyes when tested unilaterally (Fig. 4C).

We evaluated the relationship between a rank based measure of retina-wide disease severity (0=least severe, 100=most severe) ¹⁶ and the location of fixation. Eyes with foveal fixation showed the least disease severity (mean±std=16.6±14.5). Eyes with superior retinal fixation showed intermediate severity (38.3±18.3) and eyes with parapapillary fixation were more severely affected (82.3±10.1). Patients P30 and P38 were among the most severely affected (94.5±4.9) and they showed fixations in the far supero-nasal and temporal retinal loci, respectively (Fig. 4B).

Discussion

ABCA4-RD is generally accepted to show a centrifugal gradient of disease severity with macular photoreceptors and RPE having greater vulnerability to dysfunction and cell loss compared with peripheral retina. The present study documented and explored a common and consistent exception to this presumed centrifugal gradient: the parapapillary retina and RPE are spared from degeneration even at advanced disease stages. This spared region is also shown to be important for visual function in severely affected patients with central scotomas. We offer several hypotheses which, alone or in concert, may help explain these findings.

A disk membrane load hypothesis could be invoked. This hypothesis would suggest that a change in the photoreceptor to RPE ratio near the ONH explains parapapillary sparing. It is reasonable to assume that the local rate of lipofuscin accumulation is related to local ratio of photoreceptors per RPE cell, based on the distribution of lipofuscin and AF intensity in normal subjects corresponding to the spatial density of rod photoreceptors^35,36,39,41, and lipofuscin being derived from ingestion of photoreceptor outer segments⁴⁰. Interestingly, the parapapillary region in normal subjects shows a ring of decreased AF intensity⁴¹ (Figs. 1,2,3) and the extent and shape of this ring is similar to the spared region in ABCA4-RD. A constitutively lower rate of lipofuscin accumulation in the normal parapapillary zone would be consistent with the dramatic decrease in photoreceptor density (from ~140,000 to ~65,000 cells.mm⁻²) that has been observed in this region of human eyes⁴². Reduced number of photoreceptors together with an assumed invariant RPE density⁴⁵, would relieve the disk membrane load of the parapapillary RPE cells and make them less vulnerable to the increase in lipofuscin accumulation¹⁶ seen in ABCA4-RD. Inconsistent with this hypothesis is the local reduction in the photoreceptor to RPE cell ratio observed in the parafoveal region in primates⁴⁶; this region is extremely vulnerable to ABCA4-disease. Consistent with the disk membrane load hypothesis is the dramatic reduction in photoreceptor density seen normally in the far peripheral retina⁴² where retinal histology was found to be unperturbed in an eye donor with STGD³⁴.

A light load hypothesis would propose that a local gradient in the penetration of light to the retina or the RPE underlies the finding. According to this hypothesis, parapapillary sparing would occur because of reduction of either lipofuscin accumulation⁴⁷ or photooxidative damage^48,49. The retinal nerve fiber layer (RNFL) forms a prominent feature of the parapapillary retina and light loss due to scattering⁵⁰ within the increasingly thicker RNFL may reduce the amount of light penetrating to deeper layers. However, RFNL thickness topography is asymmetrically distributed around the ONH⁵¹ and therefore unlikely to be the principal cause for the circularly symmetric effect observed in this work. An increase in retinal capillary vascularity has been observed at the parapapillary region of the monkey⁵² but detailed topography of this effect is unknown and the correspondence to the spared parapapillary ring cannot be determined at this time. There could also be a parapapillary increase in melanin concentration protecting RPE cells from light-induced apoptosis⁵³; however, use of infrared excitation to obtain AF images dominated by melanin have not shown telltale signs of a high intensity ring surrounding the ONH to support such a speculation⁵⁴ (Keilhauer CN, et al. IOVS 2005;46:ARVO E-Abstract 1394; Weinberger AWA et al. IOVS 2005;46:ARVO E-Abstract 2585).

A lipofuscin clearance hypothesis would favor increased removal of lipofuscin in the parapapillary region as a cause of the spared annulus in ABCA4-RD. The existence of a clearance mechanism for lipofuscin from RPE cells has been controversial^47,55. If there was an effective mechanism of lipofuscin clearance in humans, local changes in choriocapillaris properties observed in the parapapillary region could affect the accumulation of this substance^27,56.

A neurotrophic factors hypothesis would suggest there may be increases in factors that enhance neuronal survival around the ONH. Consistent with this hypothesis, increased immunoreactivity for basic fibroblast growth factor (bFGF) has been observed around the edge of the ONH in normal mouse retinas⁵⁷. Such a local gradient could provide protection to photoreceptors from degeneration, and this effect has been observed in some animal studies^57–62. Interestingly, human retinas show a centrifugal gradient of bFGF immunoreactivity in rod nuclei⁶³, however, the spatial topography of bFGF or other factors around the human ONH is currently unknown, and a causal relationship between increased neurotrophic factors and preserved parapapillary retina or RPE remains speculative.

Not all human retinopathies show sparing of the parapapillary region and some even show enhanced vulnerability of this region. Molecularly-characterized retinal degenerations in the latter category include Sveinsson’s chorioretinal atrophy caused by dominant mutations in the TEAD1 gene⁶⁴; some phenotypes of X-linked RP caused by RP2 mutations⁶⁵; and Malattia Leventinese/Doyne honeycomb retinal dystrophy caused by a mutation in EFEMP1⁶⁶. Interestingly, vulnerability of the parapapillary region is also seen in normal aging and in age-related macular disease^67–69 (Keilhauer CN, et al. IOVS 2004;45:ARVO E-Abstract 3078). Better understanding of the special properties of the parapapillary region may contribute to understanding of the molecular/cellular disease mechanisms involved in different retinal degenerative diseases with dramatically contrasting effects on this region.

The current work also has implications for further ABCA4-RD clinical studies. The region between the fovea and optic nerve, convenient and accessible for imaging and visual function measurements, may be informative of the entire gamut of ABCA4 disease expression - from barely-detectable pathology at the parafovea at the earliest disease stages to still-detectable function at the parapapilla at the most severe stages. Tests of structure and function densely sampling this region may help determine natural history or outcome of therapy at all severity stages of ABCA4-RD.

Appendix 1: Run-length analysis

AF imaging studies in patients with retinal degenerative diseases have shown that not only the absolute intensity but also the local heterogeneity of the intensity correlates with clinical impression of local disease severity²⁸,⁴⁴. We have recently quantified this heterogeneity in ABCA4-RD by calculating the local standard deviation of AF intensity at a midperipheral locus and shown how an abnormal increase in this measure precedes localized changes in rod and cone mediated visual function¹⁶. In the current work, we used an alternative measure of “local” texture which is more appropriate for regions with local discontinuities (such as the parapapillary region) and also produces quantitative results in terms of retinal distances that are easier to interpret.

Many texture analysis methods have been proposed based on the two major characteristics of images: coarseness and directionality³¹,³². The run-length analysis method combines statistical and structural approaches to image texture³². A primitive, called a gray level run, is defined as the set of consecutive, collinear pixels having the same gray level: image regions with heterogeneous intensity have shorter runs than regions with more homogeneous intensity³³.

To quantify the local AF heterogeneity across the retina we calculated the mean run-length at each pixel (representative example shown in Fig. 5) by using two images: AF intensity image and a mask image showing the location of major blood vessels, optic nerve head and regions of atrophy. AF intensity image was spatially filtered with a Gaussian filter (radius 5 pixels) to minimize the contribution of the noise produced by the avalanche photodiode detector. Run-lengths were calculated along the eight principal directions and averaged. Run-length at each pixel was defined as the pixel-to-pixel distance multiplied by the number of consecutive, collinear neighbors having an intensity within 10% of the pixel. A principal direction was not included in averaging if a pixel corresponding to a mask value or the edge of the frame was reached. Pixel-to-pixel distance was defined as 17.5 μm along the axes and 24.9 μm along the diagonals. The result of this texture analysis method is an image where the value of each pixel represents the mean radius of a circular region over which the AF intensity would be expected to be nearly homogeneous.

Figure 5.

Estimation of local intensity heterogeneity by calculating mean run-length image. (A) A representative autofluorescence (AF) intensity image showing regions of high local heterogeneity and regions of relative homogeneity. The contrast of the grayscale images are uniformly stretched for better visibility of features. (B) Magnification of a 10×10 pixel region of the image in panel A. (C) Intensity values corresponding to each pixel of the magnified grayscale image shown in panel B. The extents of run-lengths in eight principal directions are demarcated (lines) around a chosen pixel of interest (circle); the mean run-length is 1.99 pixels, which corresponds to 0.035 mm on the retina. (D) Gray scale representation of the mean run-length values calculated at each pixel for the image shown in panel A. Regions of local intensity heterogeneity have short run-lengths and darker pixels, regions of local intensity homogeneity have longer run-lengths and lighter pixels.

Appendix 2: Pseudo-profiles using semi-polar integral analysis

Data reduction strategies applied to AF intensity and texture images could facilitate statistical comparison of results between patients. It would be preferable to attempt to retain an intuitive link in the reduced data to the exquisite spatial information contained in the underlying images. In the current work, this link was achieved by taking advantage of the approximate circular symmetry of the parapapillary region.

To perform semi-polar integral analysis (representative example shown in Fig. 6) we used registered pairs of AF intensity (or texture) and mask images. First, a 300×300 pixel region of interest centered on the ONH was cropped and split vertically into temporal and nasal halves (Fig. 6A,B). Then, values of each pixel were sampled onto temporary images representing polar coordinates (Fig. 6C,D). Integration was performed along the angle coordinate excluding masked pixels, i.e. those retinal regions corresponding to major blood vessels, optic nerve head and regions of atrophy. The result of the semi-polar integral (Fig. 6E) is a pseudo-profile that has the intuitiveness of a profile across the ONH but includes quantitative data from all of the parapapillary annulus.

Figure 6.

Demonstration of the semi-polar integral transformation used to perform data reduction on images. (A) A representative autofluorescence intensity image of the parapapillary region is divided into temporal and nasal halves. Radius (r) and angle (θ) axes of polar coordinates are shown with respect to the center of the optic nerve head (ONH). (B) A two-level mask image showing image pixels (black) corresponding to ONH and retinal blood vessels, which are to be excluded from further calculations. (C, D) Application of a rectangular to polar transformation to the intensity and mask images shown in panels A and B. Representative regions (black lines) and locations (white squares) demonstrating the correspondence between rectangular and polar coordinate representations are shown in panels A and C. (E) The result of integrating the semi-polar transformed images along the angle coordinate. Locations corresponding to black pixels of the mask image are not included in the integral.