Published: May 26th, 2023
This research describes a workflow to determine and compare autofluorescence levels from individual regions of interest (e.g., drusen and subretinal drusenoid deposits in age-related macular degeneration [AMD]) while accounting for varying autofluorescence levels throughout the fundus.
Fundus autofluorescence (FAF) imaging allows the noninvasive mapping of intrinsic fluorophores of the ocular fundus, particularly the retinal pigment epithelium (RPE), now quantifiable with the advent of confocal scanning laser ophthalmoscopy-based quantitative autofluorescence (QAF). QAF has been shown to be generally decreased at the posterior pole in age-related macular degeneration (AMD). The relationship between QAF and various AMD lesions (drusen, subretinal drusenoid deposits) is still unclear.
This paper describes a workflow to determine lesion-specific QAF in AMD. A multimodal in vivo imaging approach is used, including but not limited to spectral domain optical coherence tomography (SD-OCT) macular volume scanning and QAF. Using customized FIJI plug-ins, the corresponding QAF image is aligned with the near-infrared image from the SD-OCT scan (characteristic landmarks; i.e., vessel bifurcations). The foveola and the edge of the optic nerve head are marked in the OCT images (and transferred to the registered QAF image) for accurate positioning of the analysis grids.
AMD-specific lesions can then be marked on individual OCT BScans or the QAF image itself. Normative QAF maps are created to account for the varying mean and standard deviation of QAF values throughout the fundus (QAF images from a representative AMD group were averaged to build normative standard retinal QAF AMD maps). The plug-ins record the X and Y coordinates, z-score (a numerical measurement that describes the QAF value in relation to the mean of AF maps in terms of standard deviation from the mean), mean intensity value, standard deviation, and number of pixels marked. The tools also determine z-scores from the border zone of marked lesions. This workflow and the analysis tools will improve the understanding of the pathophysiology and clinical AF image interpretation in AMD.
Fundus autofluorescence (FAF) imaging provides a noninvasive mapping of naturally and pathologically occurring fluorophores of the ocular fundus1. The most common blue (488 nm excitation) autofluorescence (AF) excites lipofuscin and melanolipofuscin granules of the retinal pigment epithelium (RPE)2,3,4. The distribution and increase/decrease of granules play a central role in normal aging and various retinal diseases, including age-related macular degeneration (AMD)5.
A further development of FAF, quantitative fundus autofluorescence (QAF), now allows the accurate determination of topographically resolved retinal AF intensities4,6. By incorporating a reference into the optical pathway of the FAF imaging device, AF intensities can be compared between devices, time points, and subjects. This technique has resulted in a paradigm shift with regard to a presumed pathogenetic factor in AMD, which for a long time was speculated to be due to excessive lipofuscin accumulation in RPE cells7. Histologic and clinical quantification of AF, however, has revealed a decrease in AF in AMD (due to the redistribution and loss of autofluorescent lipofuscin and melanolipofuscin granules), instead of the proposed increase in AF8,9,10.
Monitoring AF has clinical implications. Von der Emde et al. and others have showed that AF is not only decreased but also further decreases in the course of AMD in high-risk, intermediate AMD eyes8,9. Additionally, histological studies suggest that most AMD-affected RPE cells show a characteristic behavior with granule aggregation and extrusion prior to RPE cell loss via subduction, sloughing, migration, or atrophy13,14,15,16. This further indicates that AF loss might be a trigger or a surrogate signal of impending disease progression.
QAF studies so far have only evaluated AF globally at the posterior pole-using prefabricated grid polar coordinate systems (e.g., QAF8/Delori Grid)17. Using prefabricated grids to measure AF results in multiple AF values on predetermined areas per eye of a subject. Investigating AF values in this way might miss local changes in areas with pathologically altered AF, for example, in AMD atop or close to drusen or subretinal drusenoid deposits (SDDs). Drusen, and to a higher degree SDDs, are associated with a high risk of developing late AMD and vision loss. Drusen in particular have a typical cycle of increasing in size over many years and may deteriorate rapidly prior to atrophy. It is conceivable that, for example, global AF decreases in AMD, but increases or is even further reduced in and around these specific disease-related focal lesions.
Different local AF patterns could also have prognostic relevance for disease progression. For example, autofluorescence levels might be used to assess whether drusen are increasing in size or are already in regression to atrophy. It has already been shown that altered AF perilesional patterns in geographic atrophy largely impact atrophy progression over time18. Additionally, local autofluorescence patterns could reveal further details into the health of the RPE. Oftentimes, the optical coherence tomography (OCT) shows hyperreflectance into the choriocapillaris, although the RPE layer appears intact. A multimodal approach combining local QAF values and OCT might help differentiate lesions with a high risk of RPE disruption and impending atrophy.
One reason spatially resolved analyses in studies have not been performed is because the most commonly used manufacturer's software does not provide a tool for these types of analysis. AF properties of different lesions dependent on the AMD disease stage could further explain the pathogenesis of AMD. Therefore, a tool to measure regional, lesion-specific AF would be desirable. To accurately compare lesions that are located throughout the retina, the workflow needs a way to account for varying degrees of AF in the human fundus19. Most centrally, AF is characteristically lower due to the shadowing effects of macular pigment and differing granule counts20,21.
AF reaches its peak at ~9° (distance to the fovea in all directions) and decreases to a greater extent peripherally4. Therefore, if one were to compare absolute values of AF levels from soft drusen (located at the fovea and parafovea in low AF areas) and SDDs (located paracentrally in high AF areas), the results would not be comparable22. Inspired by the work by Pfau et al. and the concept of sensitivity loss (correcting sensitivity measured in AMD for the hill of vision [declining retinal sensitivity with distance to the fovea] of healthy controls) for fundus controlled perimetry, AF is compared to standardized AF values throughout the macula23,24. The results are reported as z-scores (numerical measurement of a region of interest value's relationship to the mean).
The purpose of this study is to evaluate the use of a new tool for measuring local QAF levels in different types of lesions in patients with AMD. This tool is designed to measure autofluorescence levels of lesions identified on OCT scans. This enables the assessment of local autofluorescence levels in lesions, such as soft drusen or SDDs, and allows for tracking of AF changes from lesions over time. The potential utility of this tool is to enable a new structural biomarker that estimates the health of the RPE and may have prognostic value for the investigated lesions.
The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the University of Bonn (protocol code 305/21). Written informed consent was obtained from all subjects involved in the study. We required all participants in the video to sign release forms granting us permission to use their likeness and personal information in the creation of an online video.
1. Quantitative autofluorescence (QAF) image acquisition
2. Image export
3. Open-source plug-ins for QAF analysis-installing the pipeline
NOTE: The presented QAF software is an open-source plug-in named "Spectralis pipeline" created for the open-source software ImageJ (FIJI expansion)26.
4. Setup - data storage
NOTE: To allow a seamless workflow, it is recommended to set up the folder structure as follows. First, set up a folder for each study subject. Oculus dexter (OD) and oculus sinister (OS) refers to the right and left eye, respectively, and these abbreviations are used throughout this workflow.
5. Conversion of the QAF XML file into a QAF image (plug-in used: QAF_xml_reader)
6. Registering QAF images with the OCT image (plug-in used: Register_OCT_2)
NOTE: This step is needed to accurately align the OCT image with the QAF image, so that lesions in the QAF images and OCT BScans are aligned.
7. Creating an averaged QAF image for comparison (plug-in used: StandardRetina/BatchStandardRetina)
NOTE: QAF values are strongly dependent on the retinal location (e.g., central shadowing caused by macular pigment). Therefore, the QAF values of drusen should be compared to standard QAF values of the same region. As a prerequisite for analysis, the StandardRetina creates an enface map of averaged QAF images (for example, from an aged-matched control cohort). The resulting enface map shows a pixel-by-pixel map of an average QAF value for the central retina.
8. Annotating regions of interest for analysis (plug-in used: Mark_BScans_OCT)
Viewing the output
To adequately analyze and draw conclusions from the results, it is important to understand the output file of Mark_Bscans_OCT. The first three columns are labeled after the case ID, the laterality of the file, and the imaging modality that was chosen. The fourth column is referred to by mode and is labeled z-score. Note that as of writing this text, Mark BScans can only calculate all lesions in one go; the rows refer to iso-hulls, whose distances from the outer edge of the lesion are specified in the lower and upper columns of the spreadsheet. Iso-hulls measure AF in z-scores (in case of QAF) in a specified circumference surrounding the lesion. Note that the minimum value of a pixel in an iso-hull can be found in the columns labeled min, the columns labeled median, max, mean, and stdev, respectively contain the median, maximum, mean, and standard deviation of the mean of the pixel values in aniso-hull. The column n contains the total number of pixels in an iso-hull. Figure 1 shows a singular marked soft drusen of an 84-year-old male patient with intermediate age-related macular degeneration (iAMD).
Figure 2 shows the left eye of a representative patient with SDDs marked with the QAF-Workflow tool (Figure 3). SDDs in this patient were associated with reduced AF (z-score = -0.4 ± 0.2). Similarly, the iso-hulls around the SDD demonstrated reduced AF (e.g., closest iso-hull = -0.3 ± 0.3) compared to the StandardRetina. A plausible explanation for this phenomenon might be shadowing effects (reduced translucency) of SDD lesions on the RPE. The use of SDDs was exemplary. The tool enables the assessment of local AF levels in other lesions, such as drusen, as well. Furthermore, the tool allows for tracking AF changes from lesions over time.
Figure 1: A singular marked soft drusen of an 84-year-old male patient with intermediate age-related macular degeneration (iAMD). (A) The QAF image of a left eye with the marked drusen. (B) Close-up of the drusen: brown center representing the marked drusen and colored bands representing the surrounding iso-hulls. The table below depicts the output file. QAF drusen values are compared to corresponding QAF values of the corresponding eccentricity from the StandardRetina. This results in z-scores that represent deviation from the mean of unaffected areas. The blue box shows from left to right: the case ID, laterality of the eye, modality used, and the desired output (in this case, z-scores). Columns within the orange box show the boundaries of the measured area in millimeters (lower = lower bound, upper = upper bound). The green box labels the columns showing the QAF measurements. From left to right, these contain the minimum, median, maximum, number of pixels, mean, and standard deviation of the mean. Each row represents an iso-hull, rows within the blue box represent values within the lesion, and rows within the purple box show the iso-hulls surrounding each lesion (from top to bottom with increasing distance to the lesion). Scale bar = 1 mm. Please click here to view a larger version of this figure.
Figure 2: Marked SDDs in a QAF image of an 80-year-old female patient with early AMD. (A) SDDs can be seen in the QAF image. The same QAF image is shown with imprinted annotations of SDDs. (B) Around each marked lesion, the iso-hulls are depicted with color-coding (light green, dark green, and red). (C) A magnified version of the blue rectangle. The outer edge of each SDD is marked in blue. Abbreviations: QAF = quantitative autofluorescence; AMD = age-related macular degeneration; SDD = subretinal drusenoid deposit. Scale bar = 1 mm. Please click here to view a larger version of this figure.
Figure 3: Workflow to determine AF of lesions. This figure visualizes the software plug-ins needed to determine lesion-specific AF. (A) Image shows a color-coded QAF image that can be used to visualize the distribution of QAF values, but should not be used for further analysis. (B) A QAF image in the foreground, with the infrared image from the SD-OCT scan in the background is shown. This is supposed to visualize the alignment using vessel bifurcations. This can be done using the Register_OCT_2 plug-in. (C) A StandardRetina which is used to measure the z-score values of lesions. StandardRetinas can be created using StandardRetina/BatchStandardRetina.(D) A BScan with blue arrows pointing to SDDs, which are highlighted by yellow lines (note: lesions are always marked below the RPE independent of the location in the z direction) is depicted. (E) All marked lesions are seen imprinted on a QAF image (see Figure 1). The last two steps are done using the Mark_BScans_OCT plug-in. Abbreviations: AF = autofluorescence; QAF = quantitative autofluorescence; SDD = subretinal drusenoid deposit; IR = infrared; RPE = retinal pigment epithelium; SD-OCT = spectral domain optical coherence tomography. Please click here to view a larger version of this figure.
This workflow provides a step-by-step guide to use open-source FIJI plug-in tools to determine and compare AF of AMD-specific lesions. The plug-ins provide easy-to-use templates that do not require any coding knowledge and can be applied by physicians without technical support27. To our knowledge, these tools are one of a kind for lesion-specific AF quantification.
QAF values naturally vary across the retina, with values being higher at the periphery and lower in the macula due to uneven lipofuscin and melanolipofuscin distribution within the retina, the low AF of vessels, and uneven macular pigment distribution. Due to the high variation of naturally occurring QAF levels in the retina, analyzing absolute QAF values of lesions directly is not a promising approach. For example, a hypoautofluorescent lesion in the periphery might still have higher absolute QAF values than physiologic fluorescence levels of the macula. The use of a StandardRetina and the use of z-scores to measure the fluorescence levels of drusen correct for this naturally occurring variance of QAF values.
A z-score is a numerical measurement of a region of interest value's relationship to the mean in the StandardRetina. It is calculated by subtracting the mean from an individual from the mean from the StandardRetina at the same location, and then dividing the result by the standard deviation. This standardization allows for the comparison of different QAF images, as the z-score indicates how many standard deviations a value differs from the mean. A positive z-score indicates that the value is above the mean, while a negative z-score indicates that it is below the mean.
It is important to note that there may be potential pitfalls that should be considered. While this method accounts for the varying amount of AF levels throughout the fundus, it may still not be the most accurate way to measure and compare an RPE's AF. Individuals have different levels and topography of macular luteal pigment, and lesions may affect the translucency of the overlying retina as well28,29. It is therefore plausible that the measured reduced AF in areas of SDDs (see representative results) is a consequence of shadowing effects rather than decreased fluorophores in the RPE30,31,32.
We are currently working on a workflow to account for retinal reflectivity, thickness, and quantified macular pigment (using green and blue AF) with linear mixed models. Additionally, so far, QAF uses an age-dependent correction factor to account for lenticular opacification that disregards interindividual differences in the lenticular opacification of participants of a similar age33. We are therefore currently working on a workflow for a personalized correction factor of lenticular autofluorescence and opacification. To reliably extract information of AF from small lesions, adequate test-retest reliability of QAF images is needed. To further differentiate those QAF images where more detailed analysis is viable, we are investigating "QAF image reliability indices" that can predict the test-retest reliability of QAF images. At the current stage, the prudent approach is to acquire duplicate images and investigate the retest reliability of lesion-specific AF.
The presented method of additionally analyzing the iso-hulls of lesions was technically difficult to implement, as iso-hulls of neighboring lesions merge. Areas of merged iso-hulls could be characterized distinctively depending on which lesion is considered. Our solution was to consider all lesions of one type as one lesion and to analyze their periphery as a joint iso-hull. This method, however, drastically reduces the ability to measure the iso-hulls of individual drusen and might be considered a further pitfall of this technique. More technically sophisticated methods to account for merged iso-hulls or suspended reporting of AF in areas of merged iso-hulls could facilitate the analysis of AF in the circumference of lesions in the future.
We used AMD as a model disease for this study. The workflow can be adapted to study lesions in other diseases as well. So far, QAF has been used in many chorioretinal diseases, including recessive Stargardt disease, Bestrophin-1 associated diseases, various forms of retinitis pigmentosa, acute zonal occult outer retinopathy, pseudoxanthoma elasticum, and others17,33,34,35,36,37. As this workflow uses open-source software, we encourage others to duplicate this work in determining lesion-specific AF and expand our knowledge of retinal disorders. In summary, we present a workflow to determine and compare AF levels of different retinal lesions throughout the macula. This workflow paves the way for more in-depth analysis of AF and could facilitate the development of new biomarkers in AMD and beyond.
Leon von der Emde reports receiving payments from Heidelberg Engineering. Merten Mallwitz reports no financial disclosures. Kenneth R. Sloan also reports no financial disclosures. Frank G. Holz reports consulting/personal payments for Acucela, Alcon (C), Gyroscope Allergan Apellis, Bayer Bioeq/Formycon, CenterVue, Roche/Genentech, Geuder, Ivericbio, NightStarX, Novartis, Optos, Oxurion, Pixium Vision, Stealth BioTherapeutics, Zeiss, and GRADE Reading center. Thomas Ach reports consulting/personal payments for Bayer, Apellis, Roche, and Novartis.
This work was funded by the German Ophthalmologic Society (DOG) grant for doctoral students (MW) and the NIH/NEI 1R01EY027948 (TA).
|FIJI (Image J)
Copyright © 2024 MyJoVE Corporation. All rights reserved