Autofluorescence Imaging
Basic principles
Fundus autofluorescence (FAF) imaging is a noninvasive imaging method for in vivo mapping of naturally or pathologically occurring fluorophores of the ocular fundus. The dominant sources are fluorophores accumulating in lipofuscin (LF) granules in the retinal pigment epithelium (RPE).1 In the absence of RPE cells, minor fluorophores including collagen and elastin, e.g., in choroidal blood vessel walls, may also become visible. Bleaching phenomena and loss of photopigment may result in increased FAF by reduced absorbance anterior to the RPE level.
Retinal pigment epithelium and lipofuscin
The RPE constitutes a polygonal monolayer between the neurosensory retina and the choroid and is essential for vision. Given multiple essential physiological functions of the RPE, it is not surprising that RPE dysfunction has been implicated in a variety of retinal diseases (reviewed by Schmitz-Valckenberg et al.2).
A hallmark of aging is the gradual accumulation of LF granules in the cytoplasm of RPE cells. It is thought that progressive LF accumulation is mainly a byproduct of the constant phagocytosis of shed photoreceptor outer-segment discs.3–5 Several lines of evidence indicate that adverse effects of excessive LF accumulation represent a common downstream pathogenetic mechanism in various monogenic macular and retinal dystrophies as well as in multifactorial complex retinal disease entities, including age-related macular degeneration (AMD).3,4,6–8
Apparently, once formed, the RPE cell has no means of either degrading or transporting LF material and granules into the extracellular space via exocytosis. Subsequently, these granules are trapped in the cytoplasmic space of the postmitotic RPE cells. Previous studies have shown that various LF components such as A2-E (N-retinylidene-N-retinylethanol-amine), a dominant fluorophore, possess toxic properties which may interfere with normal cell function via various molecular mechanisms, including impairment of lysosomal degradation due to inhibition of the lysosomal adenosine triphosphate-dependent proton pump.9–12 Other components of LF include precursors of A2-E, molecules formed by the mixture of oxygen-containing moieties within photo-oxidized A2-E, reactions between retinoids and other constituents other than ethanolamine, and peroxidation products of proteins and lipids.13,14 The molecular composition of LF may possibly be dependent on specific underlying molecular mechanisms. Zhou and associates demonstrated with an in vitro assay a link between inflammation, activation of the complement system, oxidative damage, drusen, and RPE LF.15 They suggested that products of the photo-oxidation of RPE LF components could serve as a trigger for the complement system which could predispose the macular area to a chronic, low-grade inflammatory process over time.
Detection of LF and its constituents is facilitated by its autofluorescent properties. When stimulated with light in the blue range, LF granules typically emit a green–yellow fluorescence.16,17 The distribution of LF in postmitotic human RPE cells and its accumulation with age have been extensively studied in vitro, applying fluorescence microscopic techniques.5,6,8
Near-infrared autofluorescence
Near-infrared autofluorescence (NIA) images can also be obtained in vivo, most commonly and easily by using the indocyanine green angiography mode of the scanning laser ophthalmoscope, i.e., without dye injection.18,19 Due to the excitation and emission in the red end of the spectrum, the topographic distribution of fluorophores other than LF may be studied by this technique. It has been suggested that the NIA signal is largely melanin-derived.18–20 As such, Keilhauer and Delori18 further speculated that, to varying degrees, choroidal sources contributed to this signal. Gibbs et al.21 investigated NIA in humans and mice and suggested that melanosomes in the RPE and choroid were likely the dominant origin of the signal. Except for measurements in cell cultures at low magnification, their analyses were limited to excitation at 633 nm, in contrast to in vivo NIA, which is generated at 795 nm. Using a customized magnification lens attached to the front of the confocal scanning laser ophthalmoscope (cSLO), Schmitz-Valckenberg and coworkers studied the distribution of the NIA signal in retinal cross-sections of a human donor eye and correlated ex vivo autofluorescence measurements to in vivo findings in a rat animal model.22 They observed that the NIA signal was spatially confined to the RPE monolayer and melanin in the choroid.
Macular pigment imaging
Macular pigment, consisting of lutein and zeaxanthin, extensively accumulates along the axons of the cone photoreceptors in the central retina.23–25 As has been reported, a number of functions have been proposed for macular pigment,24,25 including filtration of blue light which may reduce photo damage and glare, minimization of the effects of chromatic aberration on visual acuity, improvement in fine-detail discrimination, and enhancement of contrast sensitivity. Neutralization of reactive oxygen species by macular pigment may have a protective effect on the neurosensory retina. Although there may be a large variation with regard to the concentration of macular pigment, the pattern of distribution is relatively uniform in the normal population. It generally shows a peak concentration at the foveal center and rapidly decreases with eccentricity, with very little present at about 8° of eccentricity.
Peak absorption of luteal pigment is at 460 nm. These absorption properties can be readily recorded in vivo by blue-light autofluorescence imaging.26 Therefore, blue FAF imaging can also be used to determine the topographic distribution of macular pigment. Compared to other methods, including heterochromatic flicker photometry, the advantage of FAF imaging is its objective acquisition technique which is not dependent on psychophysical cooperation by the examined individual.
Techniques of fundus autofluorescence imaging
Recording of autofluorescence images is noninvasive and requires relatively little time.
The intensity of naturally occurring fluorescence of the ocular fundus is about 2 orders of magnitude lower than the background of a fluorescein angiogram at the most intense part of the dye transit.1 Absorption of light with reduction of the fluorescence signal, or excitation and emission of light with an increase in the fluorescence signal by anatomical structures anterior to the retina, may further complicate or interfere with the detection of the FAF signal. In the eye, the principal barrier is the crystalline lens which has highly fluorescent properties in the short-wavelength range (excitation between 400 and 600 nm results in peak emission at c. 520 nm). With increasing age and particularly the development of nuclear lens opacities, the fluorescence of the lens becomes even more prominent.
Pioneering work on the spectral analysis of the origin of the autofluorescence signal was performed by Delori and coworkers1 using a fundus spectrometer. In parallel, von Rückmann et al., in their landmark paper, described the use of cSLO for FAF imaging.27
Fundus spectrophotometer
The fundus spectrophotometer by Delori and coworkers1 was designed to analyse systematically the excitation and emission spectra of the autofluorescence signals originating from small retinal areas (2° diameter) of the fundus. By incorporating an image intensifier diode array as a detector, a beam separation in the pupil, and confocal detection to minimize contribution of autofluorescence from the crystalline lens, this device allowed the absolute measurements of autofluorescence. These authors showed that fundus fluorescence is emitted across a broad band from 500 to 800 nm. Both at the center of the fovea and at 7° temporally, optimal excitation occurred at 510 nm with peak emission at approximately 630 nm, indicating the predominance of a fluorophore at these excitation and emission spectra. There was a significant increase with age and the recording along a horizontal line through the fovea showed a minimum fluorescence at the fovea, a maximum intensity at 7–15° from the fovea, and a decrease toward the periphery, most likely reflecting the concomitant distribution of macular pigment and melanin interfering with the emission of the dominant fluorophore. The optic disc was characterized by a less intense signal. The relationship with age and the topographic distribution of the dominant fundus fluorophore were consistent with those of RPE LF as measured in the RPE of human donor eyes.3,5
Along with autofluorescence recordings in patients with several pathological conditions, the initial work by Delori et al.1 demonstrated that LF is the dominant source of intrinsic fluorescence of the ocular fundus. However, the small area sampled by the fundus spectrometer as well as the customized relatively complex instrumentation and techniques were not practical for recording fundus autofluorescence from patients in a clinical setting.
Scanning laser ophthalmoscopy
Confocal scanning laser ophthalmoscopy (cSLO) optimally addresses the limitations of the low intensity of the autofluorescence signal and the interference of the crystalline lens. It was used initially by von Rückmann and coworkers in a clinical imaging system.27 The confocal scanning laser ophthalmoscope projects a low-power laser beam on the retina which is swept across the fundus in a raster pattern.28 The intensity of the reflected light at each point, after passing through the confocal pinhole, is registered by means of a detector, and a two-dimensional image is subsequently generated. Confocal optics insure that out-of-focus light (i.e., light originating outside the adjusted focal plane, but within the light beam) is suppressed and, thus, the image contrast is enhanced. This suppression increases with the distance from the focal plane and signals from sources anterior to the retina, i.e., the lens or the cornea, are effectively reduced.
In contrast to the 2° retinal field of the fundus spectrophotometer, the cSLO allows imaging over larger retinal areas. To reduce background noise and to enhance image contrast, a series of several single images is usually recorded (reviewed by Schmitz-Valckenberg et al.2). For the final fundus autofluorescence image, a number of these frames (usually out of 4–32) are averaged and pixel values are normalized. Given the high sensitivity of the cSLO and the high frame rate of up to 16 frames per second, FAF imaging can be performed within seconds and at low excitation energies which are well below the maximum retinal irradiance limits of lasers established by the American National Standards Institute and other international standards.29
With the cSLO, excitation is usually induced in the blue range (λ = 488 nm), and an emission filter between 500 and 700 nm is used to detect emission of the autofluorescence signal. The most widely used cSLO system for FAF imaging is the Heidelberg retina angiograph/Heidelberg Spectralis. One key advantage of the Spectralis system is the simultaneous acquisition of optical coherence tomography (OCT) recordings that allow for both averaging of several OCT B-scans in order to enhance the signal-to-noise ratio and the synchronous topographic alignment of FAF intensities with OCT findings.30 Other previous systems, such as the Rodenstock cSLO and the Zeiss prototype SM 30 4024 for FAF imaging, are no longer commercially available. Nidek has recently introduced the F-10 cSLO platform that also allows for FAF imaging (Fig. 4.1).
Fundus camera
The relatively weak fundus autofluorescence signal, absorption effects of the crystalline lens, nonconfocality, and light-scattering effects are important limitations of fundus camera-based systems for FAF recordings. Delori and coworkers described a modified fundus camera for FAF imaging.31 Their design included the insertion of an aperture in the illumination optics of the camera in order to minimize the loss of contrast caused by light scattering and fluorescence from the crystalline lens. However, the modification also resulted in the restriction of the field of view to a 13° diameter circle; this, together with the complex design, is the likely reason why this configuration has not been further pursued. In 2003, Spaide32 reported the modification of a commercially available fundus camera system by shifting the excitation and emission wavelengths for fundus autofluorescence imaging towards the red end of the spectrum in order to suppress the fluorescence originating from the lens (Fig. 4.2). The relatively inexpensive purchase of an additional filter set, together with the broad availability of the flash fundus camera, may make this an attractive alternative. These operate with excitation in the green spectrum and emission is recorded in the yellow–orange spectrum.33
Fig. 4.2 Range of excitation and emission for different camera systems. cSLO, confocal scanning laser ophthalmoscopy; FC, fundus camera.
In addition to the different excitation light (green versus blue) for FAF recording, other major technical differences between fundus camera systems and the cSLO setup must be considered (Table 4.1). In particular, the absence of confocal optics makes the fundus camera prone to light scattering and generation of secondary reflectance light that interferes with the FAF detection. The visualization of subtle FAF alterations is challenging with the modified fundus camera, as shown in one study of patients with geographic atrophy (GA) secondary to AMD.34
Table 4.1
Summary of technical differences between the confocal scanning laser ophthalmoscope (cSLO) and the modified fundus camera for fundus autofluorescence imaging
| cSLO | Modified fundus camera |
| One excitation wavelength (laser source) Large emission spectrum (cutoff filter) |
Bandwidth filters for excitation and emission |
| Continuous scanning at low light intensities in a raster pattern | One single flash at maximum intensities |
| Confocal system | Entire cone of light |
| Laser power fixed by manufacturer, detector sensitivity adjustable | Flash light intensity, gain and gamma of detector adjustable |
| Imaging processing with averaging of single frames and pixel normalization | Manual contrast and brightness |
Wide-field imaging
Peripheral FAF images can also be recorded with a recently introduced wide-field scanning laser ophthalmoscope (P200Tx, Optos). This system allows for FAF acquisition in less than 2 seconds by using green light excitation (532 nm). FAF recordings beyond the vascular arcades may be particularly helpful for assessment of the peripheral extension of retinal diseases (Fig. 4.3).
Fig. 4.3 Patient with geographic atrophy due to age-related macular degeneration. The image was recorded by a wide-field scanning laser ophthalmoscope (P200Tx, Optos). This system allows for fundus autofluorescence acquisition in less than 2 seconds by using green light excitation (532 nm). Note the peripheral extension of abnormal fundus autofluorescence signal nasal to the optic disc.
Interpretation of fundus autofluorescence images
The FAF image shows the spatial distribution of the intensity of the FAF signal for each pixel in gray values (arbitrary values from 0 to 255). Per definition, low pixel values (dark) illustrate low intensities and high pixel values (bright) illustrate high intensities. The topographical distribution of FAF in normal eyes demonstrates a consistent pattern, as illustrated in Fig. 4.4.27 A diffuse FAF signal over the posterior pole can be seen, while retinal vessels (due to an absorption phenomenon by blood contents, i.e., hemoglobin) and the optic nerve head (absence of autofluorescent material) are characterized by a very low signal and appear dark. Showing a high degree of interindividual variability, decreased FAF intensities at the macular area with a minimum in the fovea are observed; these are caused by absorption of short-wavelength light due to luteal pigment (lutein and zeaxanthin).
Fig. 4.4 Color fundus photograph (A) and fundus autofluorescence image (B) of the right eye of a normal subject imaged with the confocal scanning laser ophthalmoscope (Heidelberg retina angiograph, HRA 2, Heidelberg Engineering, Heidelberg, Germany). Topographical distribution of fundus autofluorescence intensity shows typical background signal with a dark optic disc (absence of autofluorescent material) and retinal vessels (absorption). Further, intensity is markedly decreased over the fovea due to the absorption of the blue light by yellow macular pigment. (Reproduced with permission from Schmitz-Valckenberg S, Fleckenstein M, Scholl HP, et al. Fundus autofluorescence and progression of age-related macular degeneration. Surv Ophthalmol 2009;54:96–117.)
Using pixel gray values, typical ratios between the intensity of the fovea and perifoveal macula have been established in normal subjects (reviewed by Schmitz-Valckenberg et al.2). Based on these findings, qualitative descriptions of localized FAF changes are widely used. Usually, the FAF signal over a certain retinal location is categorized in decreased, normal, or increased intensities in comparison to the background signal of the same image.
When analyzing absolute intensities on averaged but nonnormalized FAF images (after ensuring that the normalization of the pixel histogram is turned off), a great variability of the mean gray value for a certain retinal location is usually noted when FAF images are subsequently acquired from the same subject directly one after the other using the same imaging device. A systematic analysis by Lois and coworkers35 reported good intraobserver and moderate interobserver reproducibility when comparing the absolute mean pixel value of a 16 × 16 pixel square on the retina. In this report, the image resolution is not provided. When assuming an image resolution of 256 × 256 pixels and a 40° × 30° field (as these settings were published in previous studies using the same cSLO by the same group), the 16 × 16 pixel box would encompass a retinal area of c. 2° × 1.9°. Hence, moderate interobserver reproducibility would just have been achieved over a rather large retinal area, but was not shown for the anatomical resolution of the imaging system.
Several confounding factors have to be taken into account when comparing absolute FAF intensities between different examinations and different individuals. This not only includes standardization of settings (laser power, detector sensitivity, correction of refractive errors, and image-processing steps, including the number of averaged images), but also eye movements, position of the patient in the chin rest, orientation of the camera, distance between the camera and the cornea, fluctuations of laser power, and short-term dynamic changes in FAF intensities caused by prolonged exposure to the excitation light or previous dark adaption (reviewed by Schmitz-Valckenberg et al.2).
Recently, Delori and coworkers introduced a method for quantitative autofluorescence measurements by insertion of an internal FAF reference to account for variable laser power and detector sensitivity.36 Quantified autofluorescence is calculated accounting for the calibrated reference, the zero gray level, and the magnification (refractive error). For retinal degenerations and related diseases, this approach may enhance the understanding of disease processes, and may serve as a diagnostic aid, as a more sensitive marker of natural disease progression, and as a tool to monitor the effects of therapeutic interventions targeting LF accumulations.
Clinical applications
Age-related macular degeneration
Early AMD
Early manifestation of AMD include focal hypo- and hyperpigmentation at the level of the RPE as well as drusen with extracellular material accumulating in the inner aspects of Bruch’s membrane.37 Drusen may be distinguished based on size (small versus large) and morphology (hard versus soft). Postmortem analyses demonstrated that some molecular species in drusen material possess autofluorescent properties.
In vivo FAF changes in early AMD have been described by several authors using the cSLO and the fundus camera, respectively (reviewed by Schmitz-Valckenberg et al.2). Interestingly, drusen visible on fundus photography are not necessarily correlated with notable FAF changes and areas of increased FAF may or may not correspond with areas of hyperpigmentation or soft or hard drusen (Fig. 4.5). Overall, larger drusen are more frequently associated with notable FAF abnormalities than smaller ones, with the exception of basal laminar drusen. Crystalline drusen typically demonstrate a corresponding decreased FAF signal.
Fig. 4.5 Classification of abnormal autofluorescence patterns in early age-related macular disease. Corresponding color fundus photographs and fundus autofluorescence (FAF) images are shown. Eight phenotypic patterns are differentiated: (1) Normal (A, B): homogeneous-background FAF and a gradual decrease in the inner macula toward the fovea due to the masking effect of macular pigment. Only small hard drusen are visible in the corresponding fundus photograph. (2) Minimal change (C, D): only minimal variations from normal background FAF. There is limited irregular increase or decrease in FAF intensity due to multiple small hard drusen. (3) Focal (E, F): several well-defined spots with markedly increased FAF. Fundus photograph of the same eye with multiple hard and soft drusen. (4) Patchy (G, H): multiple large areas (over 200 µm in diameter) of increased FAF corresponding to large, soft drusen and/or hyperpigmentation on the fundus photograph. (5) Linear (I, J): characterized by the presence of at least one linear area of markedly increased FAF. A corresponding hyperpigmented line is visible on the fundus photograph. (6) Lace-like (K, L): multiple branching linear structures of increased FAF. This pattern may correspond to hyperpigmentation on the fundus photograph or to no visible abnormalities. (7) Reticular (M, N): multiple, specific small areas of decreased FAF with brighter lines in between. The reticular pattern not only occurs in the macular area but is found more typically in a superotemporal location. There may be visible reticular drusen in the corresponding fundus photograph. (8) Speckled (O, P): a variety of FAF abnormalities are noted to occupy a larger area of the FAF image. There seem to be fewer pathologic areas in the corresponding fundus. (Reproduced with permission from Bindewald A, Bird AC, Dandekar SS, et al. Classification of fundus autofluorescence patterns in early age-related macular disease. Invest Ophthalmol Vis Sci 2005;46:3309–14.)
Delori and coworkers described a pattern of FAF distribution associated with drusen which consists of decreased FAF in the center of the druse with a surrounding annulus of increased FAF.31 It has been speculated that this appearance is caused by attenuated RPE at the center and tangential orientation of RPE cells at the edges of the druse. A reduced turnover and a net increase in the amount of LF of the RPE cells at the edges would lead to the increased signal. Interestingly, this ring-like appearance of drusen with FAF imaging is much more pronounced when imaged with a flash fundus camera. Several authors have consistently reported that confluent drusen and large foveal soft drusen (drusenoid RPE detachments) topographically correspond well with mildly increased FAF using cSLO.38–40 With a fundus camera-based system, large soft drusen have a slightly decreased FAF signal at their centers and are surrounded by a faint ring of increased signal. Multiple foci and/or irregular areas of FAF are observed when several small, hard or soft drusen coalesce. Focal areas of increased FAF are typically found in the vicinity of drusen with overlying areas of pigment-clumping or adjacent to long-standing and crystalline drusen.
So-called reticular pseudodrusen have been identified as a risk factor for the development of late-stage AMD. In patients with GA this specific phenotypic pattern, which is best recognized by infrared reflectance and FAF imaging, can be detected in over 60% of eyes with GA.41 The precise morphological correlate of this distinct pattern is controversial. Speculations range from abnormalities in the inner choroid42 to subretinal deposits43; the latter speculation is based on the spectral domain (SD)-OCT changes recorded in the presence of reticular pseudodrusen.30,43,44
The spectrum of FAF findings in patients with early AMD was classified by an international expert group.40 Pooling data from several retinal centers, a system with eight different FAF patterns was developed, including normal, minimal change, focal increased, patchy, linear, lace-like, reticular, and speckled pattern (Fig. 4.5). This classification demonstrates the relatively poor correlation between visible alterations on fundus photography and notable FAF changes. Based on these results, it was speculated that FAF findings in early AMD may indicate more widespread abnormalities and a greater extent of disease than is ophthalmoscopically visible. The changes seen in FAF imaging at the RPE cell level may precede the occurrence of visible lesions as the disease progresses. This classification system may help to identify specific high-risk characteristics for disease progression and may be of value in future interventional trials. Furthermore, it may be of use in molecular genetic analysis to identify one or several genes conferring risk for the development of certain AMD manifestations.
Recent approaches to investigate FAF findings in AMD patients have included the use of image analysis software to compare pixel values and topographically map and register alterations visible on FAF images with fundus photographs or reflectance images.32,38 Differences in the percentage of areas with focally increased FAF intensity between eyes with various AMD manifestations have been reported. One study reported that the fellow eyes of patients with unilateral exudative AMD in the other eye tended not to exhibit FAF abnormalities. Another analysis showed that patients with exudative AMD in one eye had larger amounts of areas with abnormal autofluorescence in the fellow eye than did the eyes of patients with early disease and without a history of exudative AMD. Unfortunately, because of differences in imaging devices and the use of different image analysis protocols, comparisons between these studies are difficult and further investigation is required (reviewed by Schmitz-Valckenberg et al.2).
