Lipofuscin and Autofluorescence Metrics in Progressive STGD

Abstract

purpose. To evaluate Stargardt disease (STGD) progression and relative lipofuscin levels via autofluorescence image analysis.

methods. The relationship between focally increased autofluorescence (FIAF), geographic atrophy (GA) and focally decreased autofluorescence (FDAF) was analyzed in serial, registered autofluorescence (AF) scans of 10 patients with STGD (20 eyes, 40 scans; mean follow-up, 2.0 years) using automated techniques.

results. GA progressed uniformly in a transition zone with minimal FIAF. Only 4.3% of FIAF progressed to GA or FDAF, despite significant progression of GA (median 30%/year) and FDAF (mean, 29%/year). As a spatial predictor, the mean chance of FIAF for progression to FDAF was 4.3% ± 4.4%, significantly less than that of random areas (6.7% ± 4.0%, P = 0.029, Mann-Whitney test). In the seven eyes with GA, the mean chance of FIAF in the transition zone for transition to GA was 12% ± 8.9%, significantly less than that of random areas (33% ± 3.6%, P = 0.026, Mann-Whitney test).

conclusions. Autofluorescent flecks and FIAF deposits with AF levels elevated above the initial macular background were less likely in the short term (2 years) to transform to GA and FDAF (AF levels below the final background) than random areas, suggesting additional mechanisms beyond direct lipofuscin toxicity. FIAF/FDAF levels were observed to fluctuate, with focal remodeling of FIAF and FDAF, or rarely, even transition of FDAF to FIAF. FDAF tended to develop, not coincident with, but adjacent to initial FIAF. Because AF identifies these characteristic biological markers so specifically, autofluorescence metrics merit consideration in the study of STGD.

There is considerable interest in the effect of lipofuscin on retinal pigment epithelium (RPE) function and its role in retinal diseases. Lipofuscin granules accumulate with age in postmitotic RPE lysosomal compartments as phagocytotic remnants of photoreceptor outer segment discs. ^{1 2 3 4 5} Findings in others studies have suggested that lipofuscin and its constituent A2E exerts toxic effects on normal RPE cellular processes. ^{6 7 8 9 10 11}  

Lipofuscin granules also accumulate more rapidly than with normal aging in monogenic retinal disorders, such as Best disease, Stargardt disease (STGD), and complex degenerative diseases such as age-related macular degeneration (AMD). ^{8 12 13} However, the precise influence of these granules remains uncertain. 

Lipofuscin accumulation has been examined in vivo with fundus autofluorescence (FAF) and confocal scanning laser ophthalmoscopy. Several studies have demonstrated that hyperfluorescent FAF signals are reliable markers of lipofuscin in RPE cells. ^{14 15 16 17}  

Abnormal lipofuscin accumulation may occur in the junctional zones of geographic atrophy (GA) in AMD. ¹⁶ The importance of this phenomenon remains unclear. A previous study based on a small case series of three eyes suggested that areas of increased FAF predict development or enlargement of geographic atrophy. ¹⁸ A subsequent study of eight eyes, in which used image analysis was used for precise localization of FIAF and subsequent development of new GA, found the predictive value of FIAF to be approximately the same as chance. ¹⁹ It has also been asserted that areas without FIAF will not progress to GA (i.e., that lack of FIAF has a negative predictive value (NPV) for development of GA). ¹⁸ However, the same subsequent study did not confirm this. ¹⁹ More generally, the use of multimodal image registration and digital analysis of lesion colocalization can provide rigorous quantitative testing of hypotheses initially based on qualitative observations. ²⁰  

The view that excess lipofuscin is the driving force for RPE cell death in STGD is also widely held, even though it is recognized that some patients with STGD have normal or low levels of lipofuscin compared with normal age-matched control subjects. ²¹ This theory is partly explained in this article by duration of disease, with longer duration associated with lower AF levels and the examples with lower AF in turn mostly associated with central atrophy. However, there are examples without atrophy and fairly similar 2D AF scans, but very different AF levels (normal to high). Thus, although there is a general trend for more atrophy and decreasing AF with time, outside areas of frank atrophy, absolute AF levels can vary widely. The goal of this study was therefore to use digital techniques to determine whether lipofuscin accumulation in relative excess above background levels, as measured by FIAF, is a direct spatial precursor for GA progression or for the development of focally decreased (relative to background levels) autofluorescence (FDAF) in a 2-year time frame. This possibility will be investigated by analyzing serial AF images in patients with STGD for the spatial relationship of FIAF in the initial images to new GA and new FDAF in the final images. 

Methods

Patient Selection and Image Acquisition

Images of 20 eyes of 10 patients with STGD were selected retrospectively from a database of patients imaged from 2002 to 2008 at Columbia University. All patients had STGD by ABCA4 genotyping. Each eye had an initial and a final AF image (40 total images, mean follow-up, 2.0 years). FIAF and FDAF were present in all eyes. GA was present in seven eyes. The ages of the patients ranged from 6 to 57 years (mean age, 34 years). The age of onset, mean 27 years, was defined as the age at which visual loss was first noted. The dataset included four males and six females: nine Caucasians and one Asian. 

After pupillary dilation, fundus AF images had been recorded using a confocal SLO (HRA or HRA2; Heidelberg Inc, Heidelberg, Germany). This instrument uses blue laser light at 488 nm for illumination and a barrier filter at 500 nm to limit the captured light to autofluorescent structures. The AF images consisted of bitmapped laser scans centered on the macula. Each image was an average of three to six scans composed by the SLO software. We required good quality (second-order vascular branches definable throughout the macula) in both the initial and final images, so that FAF abnormalities would be well characterized. 

The initial and final AF images were registered in commercial software (MatLab R2006; The MathWorks, Inc., Natick, MA). We tested accuracy by superimposing the two images as a layered image (Photoshop CS2; Adobe Systems Inc., San Jose, CA) and flickering one of them on and off. The constant features are the retinal vasculature, and corresponding vessels should remain stationary. If registration was inaccurate, a new registration was performed by picking additional vascular landmarks in the computer program (MatLab; The MathWorks) and continuing until the result was satisfactory. Because they were precisely superimposed, both image scales were identical and exact spatial relations between FAF abnormalities in the initial and final images could be determined. The image pair was then cropped to a square including the fovea. All subsequent analysis was performed on these images. 

The study adhered to the tenets of the Declaration of Helsinki and received approval by the institutional review board of New York Presbyterian Hospital. 

Image Analysis

To make quantitative assessments of FIAF and FDAF relative to the image background in the presence of significant background variability, the AF image was leveled with a 12-zone quadratic polynomial mathematical model of the background in the manner previously described for AF images in AMD. ¹⁹ The mathematical model for leveling the background also includes luteal compensation by a Gaussian distribution in subjects in which the fovea is not completely atrophic. It uses published data on the spatial distribution of macular pigment and is described in more detail in an earlier paper in the context of fundus photography, but the concept is the same. ²² The math model of the background also specifically excludes hyperautofluorescent pixels as input. In each of the 12 zones, areas of FIAF are located by the two-threshold, three-class Otsu method and removed from the background pixels used as input to the model. This ensures that bright, nonbackground pixel values do not artificially elevate the model in their vicinity. (For details see Ref. ¹⁹ , p 2656). 

When GA was present, it was segmented separately before the remainder of the image was leveled with the model. The definition of GA was large (>150 μm in size), homogeneous, and well demarcated areas of profound hypoautofluorescence (gray levels within three gray-scale values of those found in the disc) to distinguish it from small areas of less profound hypoautofluorescence, which then were defined as FDAF if they met the following criterion. Briefly, after masking any GA and leveling the background, the mean and SD ς of the resulting leveled image were used to define the thresholds for increased FAF and decreased FAF. These thresholds were set at 1.5 ς above and below the mean, respectively, to determine the total FIAF and FDAF in the image. 

The choice of 1.5 ς was empiric, because we found that it gave the best selection of visually evident abnormalities in the STGD images. The imposition of a single threshold also lent some objectivity to an otherwise very subjective task. However, we recognize that that no single method or threshold can be the final arbiter of abnormality versus normality. For this reason, a sensitivity analysis was also conducted to measure the impact of redefining FIAF/FDAF thresholds at 1.0 and 2.0 ς above/below the mean in four patients. Indeed, there is no currently agreed on standard for AF abnormalities. Thus, when abnormalities were clearly present to visual inspection, the purpose of the method was to identify them efficiently so that the scientist did not spend countless tedious hours drawing them by hand for further analysis. However, manual intervention was still sometimes necessary. For example, vessel fragments were often incorrectly identified as FDAF by the basic threshold. These were removed partly by morphologic criteria in the automated system, but also manually in some cases. The morphologic criteria enabled in the software (MatLab; The MathWorks) were based on two observations: (1) Single vessel fragments were usually much more eccentric than FDAF lesions. (2) However, vessel fragments with bifurcations were not necessarily eccentric. Therefore, (1) all FDAF connected components with eccentricity greater than 0.975 were removed, and (2) the remaining connected components were skeletonized and bifurcation points removed. The separated components were retested, and the highly eccentric components were removed as before. The remaining segments and bifurcation points were reconstructed and retested to complete the analysis. These highly conservative criteria resulted in most vessel fragments, but rarely any FDAF lesion, being removed. The remaining corrections were performed manually. This portion of the algorithm could have been further optimized, but such a process was beyond the scope required for this article. 

There is also the interesting question of what happens when the method is applied to a normal image. It obviously makes no sense to insist that there are abnormal areas in a normal scan! The answer lies in our earlier publication in which the mathematical model was first introduced and validated on normal AF. ¹⁹ The fact that the model is accurate to within the noise levels for normal images was demonstrated in detail in that article on the relation of FIAF and GA in AMD. Put another way, for normal images, as in the figure in that study, the method basically identifies noise in the image (and vessels for FDAF) rather than abnormalities. Precisely, we found that 99.7% of pixels in the leveled image (other than vessels) fell within 2.0 ς of the mean in each. Thus, only tiny fractions of pixels, on the order of 0.3%, would be identified as abnormal by this criterion. No particular change with age was noted in this normal population, with ages ranging from 32 to 58. By contrast, if the gray levels of the image had a normal distribution, then gray levels above 2.0 ς would comprise 2.3% of the image. 

For each pair of registered images, a region of interest (ROI) was defined that included all significant FIAF and new FDAF. Because the areas of FDAF were variably scattered in the different images, an ROI was used in a nonstandardized but intuitively obvious manner (Fig. 1) . Calculations of lesion areas and predictive values of FIAF were all made with reference to this ROI where the disease activity of interest resided. In those cases in which actual GA was also present, the 250 μm border zone around the GA was taken as a separate ROI just as for GA in an AMD image, and separate calculations of the predictive values of FIAF for new GA formation were performed with respect to this region. New dark pixels at the edge of established GA were classified as new GA, but otherwise were classified as new FDAF. Thus, we studied two classes of dark regions on each AF scan: GA and FDAF, and the predictive values of FIAF were calculated separately for each of these two classes. 

Measurements

All areas of FIAF in the original image and new FDAF and new GA in the final image were expressed as percentages of the ROI (e.g., pFIAF = 3.2%). Those pixels of FIAF in the original image that overlay new FDAF in the final image (in the registered image pair) were denoted by the intersection FIAF ∩ newFDAF. As described previously in the case of GA, ¹⁹ the positive predictive value (PPV) that pixels with increased FAF would become new FDAF was given by:  

\[\mathrm{PPV_FIAF}{=}\ \frac{p(\mathrm{FIAF}\ {\cap}\ \mathrm{newFDAF})}{p(\mathrm{FIAF})}{\times}100\%\]

Thus, this was the percentage probability that any pixel with increased FAF in the initial image would become part of new FDAF in the final image. Note that in this setting we have adopted the term PPV as a convenient nomenclature for comparing the observed results in our patients to theoretical results that would hold if all FIAF turned to FDAF in the period of observation. Obviously, the theoretic result was not expected, but it provided a standard for comparing the behavior of other areas of the macula when both were measured by the same standard. 

The probability that any pixel in the original image falls by random chance alone into new FDAF is equal to the area of new FDAF, or p(newFDAF). That is, the PPV of random guessing is exactly as good as the percentage area of new FDAF:  

\[\mathrm{PPV_Chance}{=}p(\mathrm{newFDAF})\]

The NPV of increased FAF—that pixels without increased FAF (denoted ∼FIAF) would not become atrophic (denoted ∼newFDAF)—was given by:  

\[\mathrm{NPV_FIAF}{=}\ \frac{p({\sim}\mathrm{FIAF}\ {\cap}\ {\sim}\mathrm{newFDAF})}{p({\sim}\mathrm{FIAF})}{\times}100\%\]

This is the probability that any pixel without increased FAF in the initial image did not develop FDAF in the final image. The NPV of chance—that is, that a randomly chosen pixel does not fall into new FDAF—is exactly the percentage of the region that is not newFDAF:  

\[\mathrm{NPV_Chance}{=}100\%{-}p(\mathrm{newFDAF})\]

Note in summary that whereas the image-analysis method evaluated each image on a pixel basis, the measurements in this section were always in terms of total areas, or ratios of total areas, of pixels in the various categories. Thus, for each eye, all observations on a pixel basis were aggregated, first into relevant areas of FIAF, and then ratios to determine the PPV, as single numbers. These single numbers were the outcome variables that were compared in the results, and individual pixels were no longer considered. Hence, the approach was to try to describe complex phenomena occurring throughout the macula in simple terms by single numbers. Also, in aggregating observations based on large numbers of pixels, rather than individual pixels or small lesions, the effects of noise were more likely to average out. 

The predictive values of FIAF for the development of new GA were all defined analogously: precisely, in each equation, new FDAF is replaced by new GA. For a more complete discussion of the GA case, see Hwang et al. ¹⁹ In the case where both GA and FDAF were present, we calculated the probabilities for FIAF to transform into new GA or new FDAF separately. 

Total quantities of FIAF and FDAF were also measured in each image, and the yearly rate of change in serial image pairs was calculated. For brevity in discussion, we refer to decreasing FIAF or increasing FDAF or GA, either locally or globally, as a decreasing AF transition, with the understanding that all quantities are relative to the AF background. Likewise, increasing FIAF or decreasing FDAF is termed an increasing AF transition. Note that when total measured FIAF is decreasing, either the background (absolute) AF is increasing and/or the (absolute) AF in the flecks is decreasing, which in turn may depend on the stage of the disease. Similarly, when total FDAF is increasing, either the background (absolute) AF is increasing and/or the (absolute) AF in FDAF areas is decreasing. 

Results

Progression and Remodeling of FDAF/FIAF

Although the long-term trend was for decreasing AF overall, focally increased and decreased AF were observed to be in a constant process of remodeling, with these lesions interchanging position and intensity (Figs. 1 2) . Bright flecks did not necessarily turn dark, but often simply faded. Unexpectedly, even FDAF lesions were observed to turn bright in five patients (Fig. 2) . Indeed, the total amount of FIAF and FDAF could change, in either direction. Thus, although the mean rates of change were mostly in the decreasing AF direction (for FDAF 30% ± 45%/year; range, −51% to +147%/year; for FIAF −16% ± 35%/year, range −91% to +65%/year), the range in each case included several examples of increasing AF (Table 1) . Individual cases still should be interpreted with care for the relative nature of the measurements. For example, the large drop in detected FIAF (−43% OD) in 7-year-old patient 6 (Table 1)could have been caused by a generalized rise in background AF in this young patient with rapid disease progression, resulting in fewer areas of distinctly elevated AF. GA lesions, of course, remained atrophic, and the area of GA could only increase. 

Predictive Power of FIAF for New FDAF and New GA

As mentioned in the Methods section, the population being studied for purposes of the final statistical comparisons was not the population of pixels in a given eye, but rather the study population of 20 eyes. Thus, one set of observations was the set of 20 measurements of PPV for FIAF; another was the set of measurements of PPV for chance. In these eyes over a median period of 2.0 years, the mean positive predictive power of focally increased autofluorescence for transition to focally decreased autofluorescence was 4.3% ± 4.4%, significantly less than that of random pixel selection (6.7% ± 4.0%, P = 0.029, Mann-Whitney test). The mean negative predictive power of FIAF for transition to FDAF was 93% ± 4.0%, the same as that of chance (93% ± 4.0%, NS; Table 1 ). 

For the seven eyes with GA, the mean chance of FIAF in the transition zone for transition to GA was 12% ± 8.9%, significantly less than that of random areas (33% ± 3.6%, P = 0.026, Mann-Whitney test). The mean negative predictive power of FIAF for transition to GA was 67% ± 3.5%, the same as that of chance (67% ± 3.6%, NS; Table 2 ). Only one eye (patient 10, OD) had a PPV of FIAF for new GA greater than chance. For each of the other six eyes with GA, this value was substantially less than chance (Table 2) . 

Age of onset and duration of disease are presented in Table 1 . Although there was, as expected, fading of flecks and increasing FDAF or even atrophy later in the disease course, the findings remained unchanged when age and duration were considered. Specifically, the PPV of chance remained greater than that of FIAF for age above or below the mean, for duration of disease above or below the mean, and in the presence or absence of atrophy. That is, the PPV of chance remained greater than that of FIAF in these six subgroups: age > 34 years (6.7% vs. 3.9%); age < 34 years (6.5% vs. 4.9%); duration of disease > 7 years (8.5% vs. 5.1%); duration of disease < 7 years (5.2% vs. 3.0%); eyes with atrophy (7.7% vs. 4.9%); and eyes without atrophy (6.2% vs. 3.4%). 

There was no correlation between total initial FIAF load and subsequent rate of total GA or FDAF progression (Tables 1 2) . 

Sensitivity of Predictive Power to Definition of FIAF

Increased and decreased FAF were empirically defined to be 1.5 SD above/below the mean image intensity in the leveled image. To consider the possibility that another definition could have changed the calculated predictive values, a sample set of calculations in four sets of serial images in which increased FAF was defined as 1.0 or 2.0 SD above the mean was performed. It gave no systematic improvement in PPVs for FDAF. Two of the eight new values were better than chance, and the rest were near or below chance (Table 3) . A similar calculation for the only eye with GA and a PPV for FIAF and GA greater than chance (patient 10 OD, Table 3 ) showed slight improvement for ς = 2.0 and no change for ς = 1.0. 

The GA Transition Zone

In the seven eyes with GA, as a fraction of the 250-μm border zone, the mean new GA was 33% ± 4%, and GA progressed at a median yearly rate of 30%/years with respect to baseline. However, these transition zones generally had minimal FIAF (mean, 3% ± 1%) and displayed one of three qualitative AF types: a neutral AF level, with patches of minimally elevated and decreased AF as in Figure 3(four eyes); predominantly FDAF (two eyes of two patients); and mixed patches of FIAF/FDAF (one eye). Only the last eye had significant FIAF in the transition zone. 

An example of each of the three types is presented in Figure 4 . 

Discussion

The relationships between FIAF, FDAF, and GA in active STGD appear to be complex and in this study did not follow a consistent or monotonic progression, either from FIAF to FDAF or from FIAF to GA, as might be expected from direct lipofuscin toxicity. Total FIAF levels did not always decrease, nor did FDAF quantities always increase, as generally supposed, but could fluctuate for a period, with focal remodeling of FIAF and FDAF, or rarely, even focal transition of FDAF to FIAF. There was a striking tendency for FDAF to develop, not coincident with, but adjacent to initial FIAF. 

The appearance of increasing FIAF could simply be explained by the stage of disease activity that was captured by the serial images. In active disease, more lipofuscin could be being produced and imaged as flecks with time. It is more difficult to explain decreasing total FDAF on a cellular basis. It is unlikely that diseased cells headed to apoptosis, which were already low on lipofuscin, somehow managed to regain lipofuscin. Instead, perhaps there is a mechanism of local remodeling of the RPE layer that results in adjacent confluent areas of thickening and thinning. If the number of lipofuscin granules stays constant during changes in cell shape, they could be more tightly packed in the axis of autofluorescence visualization, resulting in FIAF, and more widely dispersed in that same axis, resulting in FDAF. Such a mechanism is consistent with the observations in this study. 

The transition zones for areas of GA in our seven cases fell into three recognizable types, only one case of which contained significant FIAF. This is contrary to AMD, in which there is often significant FIAF in the border zone. ^{18 19} Qualitatively, these types differ mostly in overall fluorescence levels and may just represent different stages in the same process (Fig. 4) . 

Quantitative analysis of these phenomena also provided unexpected findings. Although qualitative study certainly showed the variability of lesion expression and relationships, one might still have expected that most lesion transformations would display decreasing AF transitions. To test this hypothesis, precise registration of serial images, coupled with automated identification of lesion types in the original and final images, allowed computation of correlations not feasible by manual methods or visual inspection. These data, however, did more than suggest a mere lack of association between focal lipofuscin deposits, as imaged by AF, and subsequent RPE damage (loss of lipofuscin), as evidenced by focally decreased AF and GA. The mean predictive values for FIAF, significantly less than that of chance, suggest that initial FIAF and subsequent FDAF tend to avoid coinciding. The qualitative spatial relationships, as illustrated in Figures 1 and 2and as we commonly observed, were consistent with these calculations and further demonstrated a striking tendency for these lesions, while avoiding coincidence, to occur adjacent to one another. The biological implications are unclear. For example, if lipofuscin and/or melanolipofuscin were accumulating next to the RPE that was undergoing damage and losing lipofuscin, perhaps the later appearance of hypoautofluorescent RPE adjacent to such lipofuscin deposits could be explained. This sequence would parallel that noted in our previous study of AF and drusen in AMD, in which FIAF at first colocalized with drusen in early AMD and then was found adjacent to drusen and GA as atrophy supervened. ²⁰  

The study has several limitations: First, it was a retrospective study of 20 eyes in 10 patients, which may not be representative of the phenotypically heterogeneous STGD population and could reflect some selection bias. Also, imaging on successive days or weeks was not possible for assessing reproducibility of the measurements. Noise and image acquisition variability produced image variation in successive AF images. However, an advantage of the method is that by leveling the background, acquisition illumination variability was largely eliminated (see the figures). In addition, with the good-quality scans used in this study, the key abnormalities were visually evident and detected by the method. 

Second, images were derived from the HRA/HRA2 software, which registers and averages multiple individual AF scans. In each case, the number of scans averaged may differ due to scan quality and availability, resulting in differences in image dynamic range and signal-to-noise ratio. The resulting images also therefore report relative AF levels, not absolute levels. To acquire absolute levels would require spectrophotometric measurements on a point-by-point basis, as in the seminal studies by Delori et al. ^{14 15} Thus, this study describes the relationship between relative levels; precisely, between (relatively) elevated AF levels in the original image to (relatively) decreased levels in the final image. We cannot know if an area of normal background AF on a given STGD patient image actually has normal lipofuscin content, or perhaps higher, or even lower lipofuscin content than a normal control. ²¹ Bright flecks are thus simply focal areas of higher lipofuscin concentration than background levels. However, if lipofuscin were toxic, one logical prediction would still be for a dose–response in which bright flecks were more likely than random areas to turn dark in time. The finding herein, albeit for an average interval of only 2 years in this chronic disease, was that bright flecks were significantly less likely than random spots to turn dark. This pattern was maintained even in our patient with a 5-year follow-up (patient 8, Fig. 2 ). 

Third, as stated, the brief follow-up time is a limitation. What might happen in 5 years of observation is of course speculation. However, one might argue, for example, that additional areas of FIAF would transition to GA, thus giving an increased PPV for FIAF at this point and bolstering a conclusion that FIAF predicts GA after all. However, note that all measurements would have to be redone at the 5-year point. As GA progresses, the PPV for chance (which is simply the fractional area of new GA) increases also. We must also posit that that the PPV of FIAF starts out less than that of chance at the 2-year point, because this notion is supported by the present study. A moment’s reflection now makes it clear that if the PPV of FIAF starts out behind that of chance, it is not sufficient that more FIAF simply turns to GA with time to return to a conclusion that FIAF predicts GA. The trend would have to significantly reverse with time for FIAF to catch up or pass the predictive value of chance, which is always increasing. The point here is to emphasize that even “obvious” conclusions based on reasonable speculation about past or future events may be incorrect. Further follow-up with real data should resolve some of these questions. 

Fourth, another concern may be whether a median 2-year time interval is sufficient to detect significant changes in any case. But as just noted in the Results section, the mean annual rates of change of FIAF and FDAF area were both significant (−16% and +30%, respectively), and GA area progressed at a substantial annual median of 30% also. For example, Figure 3demonstrates significant enlargement in 2 years of central GA into a border zone without FIAF. Figure 2shows marked fading of flecks over 2 years, with appearance of many more areas of FDAF, which, however, did not correspond to the original flecks. 

Fifth, increased and decreased FAF were empirically defined to be 1.5 SD below the mean image intensity in the leveled image. However, we also considered more or less bright populations of flecks, as determined by standard deviations above the mean, to see whether greater or less lipofuscin content affects predictive values significantly (Table 3) . It did not, bolstering the basic conclusions of the study and also providing further evidence against a dose response for lipofuscin load. 

A main strength of the study is the quantitative rather than qualitative nature of the analysis and conclusions. Another is the uniformity of the findings across 20 eyes of 10 patients: Not only were the conclusions statistically significant, but each eye also showed little sign of FIAF turning to FDAF or GA. 

In conclusion, autofluorescent flecks and FIAF deposits with AF levels elevated above the macular background level were less likely in the short term (2 years) to transform to STGD manifestations of GA and focally decreased AF than randomly chosen areas, suggesting a multifactorial and/or time-delayed pathogenesis with RPE damage mediated in part through nonfluorescent intermediates or other mechanisms. Further study over a longer period of observation of the complex time course and relationships of FIAF, FDAF, and GA is warranted, which in turn may offer further insight into the biology of this disease. Advances in AF imaging in the future may also offer data on absolute rather than relative lipofuscin levels, thus allowing AF image analysis to be more definitive about the role of lipofuscin. 

Because AF identifies lipofuscin accumulation and GA so specifically and because these are the most characteristic biological markers of STGD, autofluorescence metrics in STGD merit consideration in the study and eventual therapy for this disease. 

 Supported by National Eye Institute Grant R01-EY015520, the New York Community Trust, and Research to Prevent Blindness.

 Submitted for publication June 18, 2008; revised August 4 and September 25, 2008, and January 4, February 13, and March 5, 2009; accepted June 10, 2009.

 Disclosure: R.T. Smith, None; N.L. Gomes, None; G. Barile, None; M. Busuioc, None; N. Lee, None; A. Laine, None

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Figure 1.

 

Patient 8 (right eye): 5-year progression and remodeling of focally decreased autofluorescence. (A) AF image from 2002 (original). (B) FIAF and FDAF detected by the model in (A). The ghost vessels in the 2002 image (not present in 2007) are an artifact caused by misalignment of one of the raw images used to generate the 2002 scan. The vessels in that image are thus faintly perceived. These ghosts did not affect the segmentation. (C) AF image from 2007 (original), registered to the image from 2002. (D) FIAF and FDAF detected by the model in (C). White ring: the ROI that included all significant FIAF and new FDAF. The FDAF is larger by a factor of 7 in area since 2002. Both FIAF and FDAF are more scattered and peripheral in the later image. However, there is also remodeling of the FDAF centrally. Note in particular the yellow outlined area in (D), which is an island of mottled but not decreased autofluorescence within the larger new area of FDAF. If this island is located, just below the foveal center, in the initial images in (B), it can be seen that it formerly contained some decreased autofluorescence that is now no longer decreased (yellow arrowhead). (E) FDAF is divided into new (not present in 2002) and old (present in 2002). Old FDAF (purple) is clearly visible within the yellow-outlined island that has now returned to normal AF levels. (F) To see the relation of AF levels to new FDAF formation, the FIAF from 2002 was overlaid on the new FDAF in 2007. New FDAF arises both inside and outside the initial FIAF ring, but coincides little. Despite a significant increase in area of FDAF, only 4.8% of FIAF pixels turned to FDAF in 5 years, relative to 7.1% of randomly chosen pixels. New FDAF appears adjacent to initial FIAF but is not coincident with it. An FIAF lesion is less likely to turn to FDAF than another randomly chosen point.

Figure 2.

 

Patient 10, left eye: local relationships of FIAF and new FDAF over 2 years. This series shows a typical progression in the totals of FIAF and FDAF over 2 years. (A) Initial image, original. (B) Final image, registered to initial image. On qualitative inspection of the original images, there appear to be fewer flecks of FIAF and more areas of FDAF in the final than in the initial image. Precisely, after automated detection, this impression was confirmed: The area of FDAF was found to have increased 14% and the area of FIAF decreased 52% overall. GA increased dramatically—by a factor of 10. The same is true of the regions enclosed in the white square: (C) initial image, (D) final image, where the area of FDAF had increased 26% and the area of FIAF had decreased 50%. GA was present in the final image, but not in the initial. (E) The FIAF in the initial image is segmented in pink. (F) The FDAF in the final image is segmented in violet. Despite the general trend to decreasing FIAF, a single FDAF lesion in the initial image, (E, white arrow), has also been selected to illustrate increasing AF transformation to a single FIAF lesion in the final image (F, white arrow). The detailed relationship between the initial FIAF and final FDAF is illustrated in (G) with the FIAF from (E) superimposed on the FDAF in (F). It is remarkable how very closely adjacent many of these lesions are, without actually coinciding. This relationship still holds in (H), when only new FDAF, red, is juxtaposed with the initial FIAF.

Table 1.

 

Predictive Value of FIAF for Progression of FDAF

Patient	Eye	Age (y)	Age of Onset (y)	Duration between AF images (y)	New FDAF (PPV_Chance)	FIAF	PPV_FIAF	NPV_Chance	NPV_FIAF	Change FIAF per Year	Change in FDAF per Year
1	OD	52	50	1	3.4	2.0	3.1	97	97	−31.2	97.2
	OS			1	4.6	2.3	0.4	95	95	32.3	−11.8
2	OD	52	45	2	5.3	1.8	1.7	95	95	7.2	14.3
	OS			2	7.7	1.6	4.8	92	92	−13.3	15.2
3	OD	57	42	2	16.8	3.1	13.3	83	83	4.5	13.4
	OS			2	5.9	3.7	5.5	94	94	4.8	−5.3
4	OD	19	14	3	1.4	1.6	1.4	99	99	6.2	19.8
	OS			3	1.0	1.3	0.9	99	99	−8.8	−8.4
5	OD	45	29	1	4.9	4.1	2.7	95	95	−53.2	−51.2
	OS			1	5.3	3.2	0.8	95	95	−91.1	31.5
6	OD	7.6	6.0	1.6	10.1	11.2	1.8	90	89	−42.8	88.9
	OS			1.6	4.4	10.4	0.4	96	95	65.4	−4.7
7	OD	19	18	1	4.8	2.5	4.9	95	95	−41.2	46.4
	OS			1	9.1	4.2	10.6	91	91	−45.1	72.3
8	OD	30	21	5	7.1	11.5	4.8	93	86	−12.9	146.8
	OS			5	14.6	14.7	14.5	85	85	−5.8	47.2
9	OD	12	11	1	*	0.2	, †	*	, †	, †	10.2
	OS			1	*	0.7	, †	*	, †	, †	8.2
10	OD	44	30	2	7.8	4.9	5.7	92	92	−26.9	25.9
	OS			2	5.6	3.0	0.6	94	94	−31.8	33.4
Mean		34	27	2.0	6.7	4.4	4.3	93	93	−15.8	29.5
SD		18	15	1.3	4.0	4.1	4.4	4.0	4.1	35.0	44.7

 Predictive values are percentages. Serial AF images were analyzed. Areas of FIAF were identified on the initial image and areas of FDAF were identified on the final image. Each was measured as a percentage of the region that included both on the registered final image. PPV_FIAF is the probability that pixels with FIAF in the initial image became part of new FDAF in the final image. NPV_FIAF is the probability that pixels without increased FAF did not become FDAF pixels.

*  Omitted because FIAF was too small for comparison of PPV.

†  FIAF amounts were too small for meaningful calculation, but did give a PPV of 0.

Table 2.

 

Predictive Value of FIAF for Junctional GA Progression

Patient	Eye	Duration between Images (y)	New GA (PPV_Chance)	FIAF	PPV_FIAF	NPV_Chance	NPV_FIAF	Increase in GA per Year
2	OD	2	24.8	0.8	0.0	75.2	75.0	29.7
2	OS	2	22.5	3.2	1.4	77.5	76.9	13.8
3	OD	2	32.0	3.3	20.7	68.0	67.6	8.2
3	OS	2	25.2	2.1	2.9	74.8	74.4	8.7
5	OD	1	27.0	2.9	0.0	73.0	72.2	29.5
10	OD	2	47.7	9.6	61.8	52.3	53.8	43.7
10	OS	2	49.5	0.1	0.0	50.5	50.5	530.6
Mean		1.960	32.6	3.1	12.4	67.4	67.2	29.5*
SD		1.290	3.6	1.0	8.9	3.6	3.5	N/A

 Predictive data are percentages. Serial AF images were analyzed for eyes with GA. Areas of GA were identified on the initial and final images. FIAF was identified and measured as a percentage of the 250-μm border zone surrounding the initial GA lesion. PPV_FIAF is the probability that pixels with FIAF in the initial image became part of new GA in the final image. NPV_FIAF is the probability that pixels without increased FAF did not become atrophic.

*  Median value.

Table 3.

 

Sensitivity of Predictive Value of FIAF for New FDAF to Standard Deviation Thresholds Used to Define FIAF and FDAF in Four Eyes and Sensitivity of Predictive Value of FIAF for New GA in One Eye

Patient	Eye	SD	New FDAF (PPV_Chance)	FIAF	PPV_FIAF	NPV_Chance	NPV_FIAF
1	OD	2	1.7	0.8	1.3	98.3	98.3
		1.5	3.4	1.8	3.1	96.6	96.6
		1	5.1	1.9	7.7	94.9	95.0
4	OS	2	1.8	0.8	0.0	98.2	98.2
		1.5	1.0	1.3	0.9	99.0	99.0
		1	2.9	2.7	1.0	97.1	97.1
7	OS	2	3.9	3.4	6.0	96.1	96.2
		1.5	9.1	4.2	10.6	90.9	90.9
		1	7.1	14.5	6.0	92.9	92.8
10	OD_FDAF	2	7.7	4.2	3.2	92.3	92.1
		1.5	7.8	4.9	5.7	92.2	92.1
		1	5.6	7.6	4.4	94.4	94.3

Patient	Eye	SD	New GA (PPV_Chance)	FIAF	PPV-FIAF	NPV_Chance	NPV_FIAF
10	OD_GA	2	49.8	5.0	74.9	50.2	51.6
		1.5	47.7	9.6	61.8	52.3	53.8
		1	49.7	9.6	63.2	50.3	51.7

 Predictive data are percentages. Pixels were originally defined as having FIAF (FDAF) if their gray-scale levels were greater than (or less than) 1.5 SD above (or below) the leveled image mean. The threshold used to define FIAF/FDAF was varied from 1.0 to 2.0 SD, to measure the impact on predictive values. PPV_FIAF is the probability that pixels with FIAF in the initial image became part of new FDAF (or new GA) in the final image. NPV_FIAF (negative predictive value) is the probability that pixels without increased FAF did not become FDAF pixels (or become atrophic).

Figure 3.

 

Patient 2, right eye: 2-year progression of geographic atrophy in a 53-year-old man. (A) Initial AF scan from 2006, with a 250-μm border zone in outlined in white. (B) GA segmented in black and FIAF in pink, mostly outside the border zone. The border zone itself had neither significant FIAF nor FDAF, but did have patches of minimally elevated and decreased AF. (C) Final AF scan from 2008, registered exactly with the 2006 scan, with same border zone superimposed. The GA had progressed. (D) Superimposition of the final GA from (C) on the initial GA in (B) allows identification of the new GA (dark violet). It had grown significantly, uniformly in all directions, and was also essentially unrelated to the initial FIAF. In fact, there were only two tiny FIAF pixels in the border zone, and neither turned to GA (predictive value of 0).

Figure 4.

 

The three types of transition zone for geographic atrophy. (A) Patient 3 left eye, initial scan. The 250-μm transition zone has an essentially neutral AF level, similar to that of patient 2, right eye (Fig. 3) , with patches of minimally elevated and decreased AF. (B) Patient 5 right eye, initial scan. The transition zone is predominantly decreased autofluorescence. (C) Patient 10 right eye, initial scan. The transition zone has mixed increased and decreased AF. Note that these three transition zones could simply represent progressive stages in the same basic process, with autofluorescence greatest in type C, neutral in type A, and least in type B. (D) Patient 3 (left eye), final scan, 2-year interval. New geographic atrophy (GA) is indicated in dark violet, and the FIAF from the initial scan in pink is superimposed, showing essentially no correspondence with the new GA. This result is also similar to that of patient 2 (Fig. 3) . (E) Patient 5 (right eye), final scan, 1-year interval. The relationship of initial FIAF and new GA is much like (D). (F) Patient 10 (right eye) final scan, 2-year interval. The initial FIAF in the transition zone largely coincided with new GA, but there was also significant new GA nasally where no FIAF was present. In fact, in each of these cases the GA had grown uniformly circumferentially with no quadrantic preference.