July 2024
Volume 65, Issue 8
Open Access
Retina  |   July 2024
Sequence of Morphological Changes Preceding Atrophy in Intermediate AMD Using Deep Learning
Author Affiliations & Notes
  • Sophie Riedl
    Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medical University of Vienna, Vienna, Austria
  • Ursula Schmidt-Erfurth
    Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medical University of Vienna, Vienna, Austria
  • Antoine Rivail
    Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medical University of Vienna, Vienna, Austria
  • Klaudia Birner
    Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medical University of Vienna, Vienna, Austria
  • Julia Mai
    Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medical University of Vienna, Vienna, Austria
  • Wolf-Dieter Vogl
    Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medical University of Vienna, Vienna, Austria
    RetInSight, Vienna, Austria
  • Zhichao Wu
    Centre for Eye Research Australia, Royal Victorian Eye and Ear Hospital, East Melbourne, Australia
    Department of Surgery (Ophthalmology), The University of Melbourne, Melbourne, Australia
  • Robyn H. Guymer
    Centre for Eye Research Australia, Royal Victorian Eye and Ear Hospital, East Melbourne, Australia
    Department of Surgery (Ophthalmology), The University of Melbourne, Melbourne, Australia
  • Hrvoje Bogunović
    Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medical University of Vienna, Vienna, Austria
  • Gregor S. Reiter
    Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medical University of Vienna, Vienna, Austria
  • Correspondence: Ursula Schmidt-Erfurth, Department of Ophthalmology and Optometry, Medical University of Vienna, Währinger Gürtel 18–20, Vienna 1090, Austria; [email protected]
Investigative Ophthalmology & Visual Science July 2024, Vol.65, 30. doi:https://doi.org/10.1167/iovs.65.8.30
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Sophie Riedl, Ursula Schmidt-Erfurth, Antoine Rivail, Klaudia Birner, Julia Mai, Wolf-Dieter Vogl, Zhichao Wu, Robyn H. Guymer, Hrvoje Bogunović, Gregor S. Reiter; Sequence of Morphological Changes Preceding Atrophy in Intermediate AMD Using Deep Learning. Invest. Ophthalmol. Vis. Sci. 2024;65(8):30. https://doi.org/10.1167/iovs.65.8.30.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose: Investigating the sequence of morphological changes preceding outer plexiform layer (OPL) subsidence, a marker preceding geographic atrophy, in intermediate AMD (iAMD) using high-precision artificial intelligence (AI) quantifications on optical coherence tomography imaging.

Methods: In this longitudinal observational study, individuals with bilateral iAMD participating in a multicenter clinical trial were screened for OPL subsidence and RPE and outer retinal atrophy. OPL subsidence was segmented on an A-scan basis in optical coherence tomography volumes, obtained 6-monthly with 36 months follow-up. AI-based quantification of photoreceptor (PR) and outer nuclear layer (ONL) thickness, drusen height and choroidal hypertransmission (HT) was performed. Changes were compared between topographic areas of OPL subsidence (AS), drusen (AD), and reference (AR).

Results: Of 280 eyes of 140 individuals, OPL subsidence occurred in 53 eyes from 43 individuals. Thirty-six eyes developed RPE and outer retinal atrophy subsequently. In the cohort of 53 eyes showing OPL subsidence, PR and ONL thicknesses were significantly decreased in AS compared with AD and AR 12 and 18 months before OPL subsidence occurred, respectively (PR: 20 µm vs. 23 µm and 27 µm [P < 0.009]; ONL, 84 µm vs. 94 µm and 98 µm [P < 0.008]). Accelerated thinning of PR (0.6 µm/month; P < 0.001) and ONL (0.8 µm/month; P < 0.001) was observed in AS compared with AD and AR. Concomitant drusen regression and hypertransmission increase at the occurrence of OPL subsidence underline the atrophic progress in areas affected by OPL subsidence.

Conclusions: PR and ONL thinning are early subclinical features associated with subsequent OPL subsidence, an indicator of progression toward geographic atrophy. AI algorithms are able to predict and quantify morphological precursors of iAMD conversion and allow personalized risk stratification.

Atrophic AMD is characterized by progressive loss of the RPE, photoreceptors, and choriocapillaris. It represents the late non-neovascular stage of AMD, which is responsible for severe and irreversible loss of vision.1,2 Its preceding forms, early and intermediate AMD (iAMD), are characterized by medium to large drusen and the presence of pigmentary abnormalities.3,4 At this stage, optical coherence tomography (OCT) imaging may identify early signs of photoreceptor (PR) loss.5 Furthermore, several risk factors of progression toward atrophic AMD have been described using multimodal imaging with a strong focus on OCT,58 as this modality enables the acquisition of in vivo, three-dimensional, high-resolution volumetric scans in a fast and non-invasive manner.9 Nascent geographic atrophy (nGA) is a precursor of GA on OCT, defined by subsidence of the outer plexiform layer (OPL) and inner nuclear layer, and/or the presence of a hyporeflective wedge within Henle's nerve fiber layer.8,10 Among the individual features used to define nGA, OPL subsidence has been reported to be associated with the highest risk for developing GA.11 The Classification of Atrophy Meeting program subsequently published definitions of complete and incomplete RPE and outer retinal atrophy (c/iRORA) based on the layers of the outer retina involved in the atrophic process.5,12 A recent study observed that the association with GA development was significantly stronger with nGA compared with incomplete RORA.13 
In parallel to these various qualitative assessments of outer retinal integrity, fast automated evaluation and accurate quantification of OCT biomarkers have recently become feasible via artificial intelligence (AI).14 Using deep learning, reliable and precise quantification of various OCT biomarkers has been achieved in AMD.1517 These OCT-based atrophic precursors incorporate changes at the PR level18,19; however, reliable in vivo data elucidating temporal changes, leading up to the occurrence of these features are lacking. The purpose of this study was to use AI-based methods to quantify morphological changes in the outer retina in iAMD and thereby enable precise investigation of their spatiotemporal association with respect to OPL subsidence as an early biomarker of atrophy. 
Methods
Study Design and Patient Cohorts
LEAD Study Cohort and Imaging Procedures
This study is a post hoc analysis of 280 eyes of 140 participants20,21 randomized to the sham treatment arm of the Laser Intervention in Early Stages of Age-Related Macular Degeneration (LEAD) study, a prospective, multicenter, randomized controlled clinical trial investigating the effectiveness of subthreshold nanosecond laser treatment in iAMD (clinicaltrials.gov identifier: NCT01790802).22 The specific eligibility criteria for participants included in this current study have been described in detail previously.20 In brief, participants had to be 50 years of age or older with a diagnosis of iAMD, presenting with at least one druse of more than 125 µm diameter within 1500 µm from the fovea, as assessed by color fundus photography. Exclusion criteria included late neovascular and preexisting atrophic AMD, including OCT-defined atrophy (nGA or greater areas of atrophy).8 All participants provided written informed consent, including consent to the post hoc use of the data for extended analysis via an opt-out approach, and institutional review board approval was obtained at each participating center. All study procedures were conducted in accordance with the Declaration of Helsinki and all data were pseudonymized fully. Furthermore, approval for this analysis was obtained from the Ethics Committee at the Medical University of Vienna. 
Spectral domain OCT (SD-OCT) imaging was performed on a 6-monthly basis up to a period of 3 years using Spectralis HRA + OCT (Heidelberg Engineering, Heidelberg, Germany). The imaging protocol comprised OCT volume scans with 1024 A-scans and 49 B-scans in a 20° × 20° field of view centered on the fovea, with 25 frames averaged per B-scan. 
Subcohorts for the Current Analysis
  • OPL subsidence cohort (OPLS) cohort: Of the 280 eyes of 140 participants, a subgroup of eyes that developed OPL subsidence within the 36-month follow-up were identified for this analysis.
  • OPL subsidence + RORA (OPLS+RORA) cohort: Within the OPLS group, a subset of eyes that developed incomplete or complete RORA within the 3-year observation period was identified.
  • Cohort without OPL subsidence and/or RORA (NOPLS + NRORA): This subgroup was identified to compare the investigated changes to eyes showing neither OPL subsidence nor RORA.
Image Analysis
Manual Grading for Study Cohort Identification
For the purpose of identification of the OPLS, OPLS + RORA, and NOPLS + NRORA cohorts, two features were graded manually by two graders (G.S.R. and S.R.), independent of the gradings of nGA, which were performed previously on this data by the authors of the original study group. OPL subsidence, specifically subsidence of the anterior and posterior boundary of the OPL, defined as the outer nuclear layer (ONL)/OPL and OPL/inner nuclear layer junction, was manually annotated on an A-scan basis/in the z-axis for all visits. Moreover, the presence of RORA on OCT was graded at the last study visit to identify the OPLS + RORA cohort. 
Categorization Into Topographic Areas
To examine morphological changes in a precise, topographically differentiated manner, SD-OCT volumes were categorized into the following topographic areas within a 3-mm diameter circle centered on the fovea. 
  • OPL subsidence area (AS): The en face area affected by OPL subsidence at subsidence onset. Inherently, this area is not present in eyes of the NOPLS + NRORA cohort.
  • Drusen area (AD): The OPL subsidence-free en-face area, affected by drusen with a height of more than 40 µm at the time point of subsidence onset.
  • Reference area (AR): En face area remaining after excluding AS and AD.
Figure 1 shows three example cases, including B-scans of the complete follow-up, including color-coded area classification, defined at the time point of OPL subsidence occurrence. The central 3-mm diameter region was chosen because the risk of developing atrophy or nGA was shown to be highest in the central 3 mm.8 Similarly, the topographic distribution of OPL subsidence predominantly matched the predefined 3-mm area (Fig. 2). Additionally, drusen build-up is faster in the foveal and parafoveal areas compared with the perifoveal area, indicating an area of increased risk for atrophy development that was previously investigated in The Blue Mountains Eye Study.23,24 Furthermore, we selected this approach to focus on drusen-related atrophy development rather than atrophy associated with subretinal drusenoid deposits,25 which has been shown to be more frequent in the parafoveal and perifoveal areas.26,27 
Figure 1.
 
Three example cases of progressive morphological changes and OPL subsidence occurring during the 36-month follow-up period. Six-monthly representative B-scans of the OCT volumes (A, C, E) and corresponding colored overlays (B, D, F), indicating different topographic areas, defined at the time point of OPL subsidence occurrence: OPL subsidence area (AS, blue); drusen area (AD, yellow) and reference area (AR, green). The occurrence of OPL subsidence (vertical red arrow) precedes topographically corresponding drusen regression (yellow asterisk) and various degrees of choroidal HT (pink arrowheads), indicative of progressive RORA. Thinned ONL and PR in AS compared with AD is evident in visits leading up to OPL subsidence occurrence.
Figure 1.
 
Three example cases of progressive morphological changes and OPL subsidence occurring during the 36-month follow-up period. Six-monthly representative B-scans of the OCT volumes (A, C, E) and corresponding colored overlays (B, D, F), indicating different topographic areas, defined at the time point of OPL subsidence occurrence: OPL subsidence area (AS, blue); drusen area (AD, yellow) and reference area (AR, green). The occurrence of OPL subsidence (vertical red arrow) precedes topographically corresponding drusen regression (yellow asterisk) and various degrees of choroidal HT (pink arrowheads), indicative of progressive RORA. Thinned ONL and PR in AS compared with AD is evident in visits leading up to OPL subsidence occurrence.
Figure 2.
 
En face topographic distribution of OPL subsidence in the RORA subgroup. En face map showing the cumulative topographic distribution of OPL subsidence lesions over all patients. Pseudo heat map colors correspond to the percentage of eyes affected by a lesion at a specific location. Concentric rings indicate the fovea-centered 3 and 6 mm diameters.
Figure 2.
 
En face topographic distribution of OPL subsidence in the RORA subgroup. En face map showing the cumulative topographic distribution of OPL subsidence lesions over all patients. Pseudo heat map colors correspond to the percentage of eyes affected by a lesion at a specific location. Concentric rings indicate the fovea-centered 3 and 6 mm diameters.
Automated Image Analysis
To extract subclinical imaging features, all SD-OCT volumes were processed and analyzed with a comprehensive set of automated AI-based image segmentation and biomarker quantification tools. Specifically, several deep learning-based convolutional neural networks (CNNs) were used to segment the retinal layer and relevant imaging biomarkers. The PR layer was segmented with a U-shaped CNN architecture, which was extensively validated previously17 and is part of a commercially available OCT-based monitoring tool for GA.28 The network delineates the PR layer defined as the region between the inner border of the ellipsoid zone and the inner border of the RPE. From the segmentation output, a PR thickness topographic map was then calculated. Similarly, to detect drusen, another U-net was used, which detects drusen as the region between the outer boundary of the RPE and the Bruch's membrane.29 For ONL segmentation, a custom CNN, trained on a different iAMD patient cohort, was developed and specialized for layer boundary regression, which delineates upper and lower ONL boundaries.30 To ensure that the resulting ONL thickness map was quantified correctly, the layer segmentations were manually corrected where necessary. Finally, a custom method for quantifying choroidal hypertransmission (HT) was developed. It uses a signal processing approach, where locally the amount of A-scan transmission under Bruch's membrane is evaluated as a deviation from the OCT-wide A-scan transmission statistics because, in an iAMD cohort, HT can be considered an outlier. Thus, it expresses the HT extent in normalized intensity units. 
Statistical Analysis
Two main analyses were performed to characterize the morphological changes leading up to the development of OPL subsidence and investigate how these differ between the abovementioned topographic areas. 
First, to perform a crude analysis of the differences in PR thickness, ONL thickness, drusen height and HT between the topographic areas AS, AD and AR leading up to the occurrence of OPL subsidence, linear mixed models (LMMs) were fitted at each time point. Results are reported as means and confidence intervals (CIs). To investigate whether early differences in PR and/or ONL thickness between topographic areas (AS vs. AR and AS vs. AD) are significant Wald-tests were applied, starting at the earliest time point (−18 months). The P values were subjected to Bonferroni–Holm correction to account for multiple testing. 
In a second, more in-depth analysis, LMMs were fitted to both PR and ONL thicknesses to investigate how the change of both these features over time differs between the three topographic areas, and whether PR and ONL thickness are in addition impacted by a compressing effect of drusen. To this end, topographic area, time, interaction of area and time, and drusen height were included as covariates into the LMMs. Owing to the model design, only eyes, in which all three topographic areas were present (OPLS and OPLS + RORA subgroups) were included in this analysis. Measurements within the same eye and/or within the same patient were adjusted for with random effects. A two-sided significance level of 0.05 was applied. Statistical analyses were performed in Python statsmodels 0.10.1. 
Results
Cohort Characterization and OPL Subsidence Distribution
Of a total of 280 eyes of 140 participants with iAMD, 29 eyes of 29 participants were excluded for this analysis for various reasons, which are listed in detail in Supplementary Figure S1. Fifty-three eyes from 43 participants developed OPL subsidence during follow-up and were included in the OPLS subgroup. The mean age in this cohort was years 72 (95% CI, 70–74 years) and 76% were female. Of these, 36 eyes from 30 individuals were identified for the OPLS + RORA subgroup. The mean age was 71 years (95% CI, 69–74 years) in this subgroup and 69% were female. In these eyes, an en face distribution map of OPL subsidence shows the lesions to be concentrated in a 3-mm diameter ring centered on the fovea (Fig. 2). In terms of cumulative OPL subsidence area, 83% and 74% of OPL subsidence occurrences presented within the central 3-mm diameter in both subgroups, respectively. The NOPLS + NRORA subgroup, showing neither of atrophic features, consisted of 198 eyes of 111 participants. The mean age was 69 years (95% CI, 68–70 years), and 77% were female. 
Morphological Changes Leading up to OPL Subsidence in the Subsidence, Drusen, and Reference Areas
The time course of quantitative morphological features in the different topographic areas AS, AD, and AR is shown for all subgroups in Figure 3. Time point zero marks the first occurrence of OPL subsidence (OPLS and OPLS + RORA subgroups) or the last available visit (NOPLS + NORA subgroups). Presentation and course of all morphological features in AD and AR of the NOPLS + NRORA subgroups was comparable to the corresponding areas of the other subgroups with largely overlapping CIs. Inherently, no trajectory for AS is shown for the NOPLS + NRORA, because OPL subsidence did not occur in these eyes during the observation period. Supplementary Tables S1, S2, S3, and S4 include all numerical results of the respective LMMs. 
Figure 3.
 
The course of OCT feature development during progression of iAMD. Changes over time in PR thickness, ONL thickness, drusen height and intensity of choroidal HT in the OPLS (dashed line), OPLS + RORA (solid line) and NOPLS + NRORA (alternating dashed/solid line) subgroups. Number of eyes included in the respective groups is indicated by n. Differentiated topographic zones are compared, as each time course is plotted with respect to mean values within OPL subsidence area (AS, red), drusen area (AD, yellow) and reference area (AR, blue). The time axes of the plots are relative to the time point of OPL subsidence occurrence (in the OPLS and OPLS + RORA subgroup) or the last available visit (in the NOPLS + NRORA subgroup), denoted by the dotted vertical line at month 0, respectively. Bars indicate 95% CIs.
Figure 3.
 
The course of OCT feature development during progression of iAMD. Changes over time in PR thickness, ONL thickness, drusen height and intensity of choroidal HT in the OPLS (dashed line), OPLS + RORA (solid line) and NOPLS + NRORA (alternating dashed/solid line) subgroups. Number of eyes included in the respective groups is indicated by n. Differentiated topographic zones are compared, as each time course is plotted with respect to mean values within OPL subsidence area (AS, red), drusen area (AD, yellow) and reference area (AR, blue). The time axes of the plots are relative to the time point of OPL subsidence occurrence (in the OPLS and OPLS + RORA subgroup) or the last available visit (in the NOPLS + NRORA subgroup), denoted by the dotted vertical line at month 0, respectively. Bars indicate 95% CIs.
PR thickness was significantly reduced in AS with a mean 20 µm and 18 µm (95% CI, 17–22 and 16–21) compared with both AD (23 µm and 24 µm [95% CI, 21–26 and 20–27]) and AR (both 27 µm [95% CI, 24–30 and 23–30]) 12 months before subsidence occurred in the OPLS and OPLS + RORA subgroups, respectively (P < 0.009 and P < 0.006). 
ONL thickness was likewise decreased significantly in the OPLS and OPLS + RORA cohorts (both P < 0.008) in AS compared with AD and AR starting as early as 18 months before subsidence occurred (both 84 µm [95% CI and 79–89, 78–91] vs. 94 µm and 98 µm [95% CI, 89–99 and 92–104] and 98 µm and 102 µm [95% CI, 93–103 and 96–108], respectively). 
Drusen height was comparable between AS and AD up to the onset of OPL subsidence (Fig. 3 bottom left), at which time it showed a marked decrease to a mean of 33 µm and 35 µm [95% CI, 28–39 and 27–42] in AS, compared with AD in which an increase to a mean of 70 µm and 69 µm [95% CI, 63–78 and 60–78] was observed in the OPLS and OPLS + RORA subgroups, respectively. 
Longitudinal analysis of choroidal HT revealed a trend of increasing HT in AS, pronouncedly at the time point of subsidence. HT in AD and AR remained stable throughout the observation period. 
PR Thickness as Influenced by Time, Topographic Area, and Drusen Height
We highlight the results of the LMM, reporting PR thickness in the OPLS + RORA subgroup as a function of topographic area, time and drusen height. Similar results were obtained in the OPLS subgroup; complete data are shown in Table. In addition to PR thickness being significantly lower in areas developing OPL subsidence (AS) at the time point of its occurrence (−13 µm ± 1; P < 0.001 compared with AD and −14 µm ± 1; P < 0.001 compared with AR), it also showed significant thinning over time of 0.8 ± 0.1 µm per month (P < 0.001). The observed rate of thinning in this area was significantly faster compared with that in both AD and AS (both by 0.6 ± 0.1 µm per month; P < 0.001) These results are corrected for a significant compressive effect on PR of 0.1± 0.01 µm per µm drusen height, which was revealed by the model (P < 0.001). 
Table.
 
Output of the LMM for PR and ONL Thickness for the OPLS and OPLS + RORA Subgroups
Table.
 
Output of the LMM for PR and ONL Thickness for the OPLS and OPLS + RORA Subgroups
ONL Thickness as Influenced by Time, Topographic Area, and Drusen Height
The results of the LMM model for ONL thickness are reported in detail for the OPLS + RORA subgroup in this section. Complete results are shown in Table, confirming similar findings for the OPLS cohort. Likewise, as for PR, ONL thickness was statistically significantly lower in AS at the time point of subsidence occurrence in comparison to AD and AR (88 ± 3 µm vs. 111 ± 2 µm and 105 ± 2 µm, respectively, both P < 0.001) and also showed statistically significant thinning over time of 0.8 ± 0.1 µm per month (P < 0.001), a rate significantly accelerated compared with that of AD and AR by 0.8 ± 0.2 µm per month and 0.6 ± 0.2 µm per month, respectively (both P < 0.001). A statistically significant compressive effect on ONL thickness of 0.3 ± 0.02 µm per µm drusen height was detected and corrected for in the model (P < 0.001). 
Discussion
AMD is a progressive disease, whereby the non-neovascular component advances continuously, eventually resulting in extensive and irreversible loss of retinal tissue and visual function. The progression from iAMD to atrophic AMD has been a focus of research for the last decade with the aim to detect and characterize the morphological features associated with disease progression and identify imaging biomarkers, highly sought for with respect to participant selection for therapeutic interventions, which have recently been introduced for late atrophic AMD.31,32 OCT enables visualization of the atrophic process in a three-dimensional and high-resolution manner. The build-up of drusen volume, followed by rapid regression, as well as the presence of hyper-reflective foci and progressive degeneration of the RPE, have been associated with a higher risk of progression to late AMD.3335 
Our aim was to identify the earliest subclinical signs of outer retinal degeneration on OCT imaging, topographically preceding OPL subsidence, a feature associated with a high risk of atrophy development.8,10 To this end, we applied objective AI-based topographic quantification of signs of outer retinal degeneration, which enable insights into longitudinal changes prior to the development of OPL subsidence, specifically by comparison to regions with drusen without subsidence or otherwise unaffected regions. We expect this study to contribute to the search for potential biomarkers for future treatment evaluation. 
In this study, gradings by the investigators of this study in Vienna identified 53 eyes that developed OPL subsidence during the 36-month follow-up period, of which 36 eyes developed RORA during the follow-up period. The incidence of OPL subsidence was higher than the development of nGA, which requires additional subsidence of the inner nuclear layer and/or presence of a hyporeflective wedge-shaped band, in 40 eyes in this same cohort by the investigators in Melbourne, Australia (R.H.G. and Z.W.; who originally defined nGA) as reported in recent studies.20,21 Ferrara et al.11 have reported OPL subsidence to be the most prognostic single OCT feature of nGA for progression toward atrophy, which was the rationale for selecting the occurrence of this single feature as early atrophic endpoint up until which morphological changes were investigated. 
Interestingly, the distribution of OPL subsidence graded as part of this study was mainly found to localize within the central 3 mm, mostly sparing the fovea. This finding is consistent with the typical initial appearance of clinical GA, which develops in the parafovea and expands in a ring-like configuration with foveal sparing.1 Our findings are also consistent with previous reports on nGA, which develop mainly in the central 3-mm area.8 The drusen load in the central macular area and its faster drusen build-up make this area prone to disease progression, often associated with drusen regression and progression to late AMD.23,24,36 Although drusen volume has been reported as a significant risk factor for atrophic progression on the eye level, drusen height itself does not relate to the risk of drusen collapse and topographically corresponding resulting atrophy. This is underlined by the finding of our longitudinal observation, which clearly show similar drusen load in the drusen and subsidence topographic areas up until the time of subsidence occurrence (Fig. 3, bottom left). This suggests that OPL subsidence-associated atrophy occurs topographically colocalized with drusen. However, the identification of areas destined to become atrophic is not feasible based on drusen height alone. Therefore, the subclinical changes, detected by our analyses, discussed in the following, may help to distinguish areas with and without the risk of developing outer retinal atrophy, despite a similar drusen load. 
Our analyses, using high precision AI-supported measurements were able to reveal distinct, robustly quantified, subclinical changes, topographically corresponding with areas of future outer retinal atrophy and atrophy-associated changes. Progressive PR and ONL thinning, both signs of PR degeneration, were detectable as early as 12 and 18 months before OPL subsidence, respectively, and showed progressive thinning up to the time point of subsidence occurrence. The temporal association of drusen regression and increased HT with the time point of OPL occurrence, highlight the progression toward RPE atrophy in this topographic region. Morphological trajectories in the drusen and reference zones, with which the subsidence zone was compared, were comparable with the same zones of a comparison subgroup of the LEAD study, showing neither OPL subsidence nor RORA during the observation period. This finding highlights the confined nature of the observed subclinical changes to precise topographic locations of future atrophic progression. 
Recent findings, correlating the attenuation of ellipsoid zone reflectivity on OCT to areas with choriocapillaris flow deficits in iAMD, underline the relevance of changes of the PR layers, as visualized by OCT, with respect to disease pathomorphology.19 Compared with ellipsoid zone reflectivity, our models measure PR outer segment thickness, also correcting for the compression effect in areas with drusen. In the AREDS2 ancillary OCT study, disruption of PR layers and thinning were observed before the onset of new GA, which is compatible with the results of our study.37 Our AI-based results, showing PR thinning to precede the occurrence of OPL subsidence, which topographically developed mainly in the parafoveal area, are in line with investigations by Curcio et al.38 in human autopsy eyes, describing an initial loss of parafoveal rods compared with foveal cones. Loss of cell nuclei in the ONL, together with the degeneration of photoreceptors, were suggested as features of the aging retina, which might not affect other retinal layers necessarily at the time of early degeneration.39 Migration of PR cells exceeding Henle's fiber layer and the depopulation of ONL have also been discussed in the progression of GA secondary to non-neovascular AMD.40 In summary, all of these findings support the progressive alterations on the PR level, which were demonstrated by longitudinal in vivo OCT imaging and concomitant AI-based image analyses in our study, confirming previous reports on histology in the aging retina and non-neovascular AMD. 
A limitation of this investigation is its post hoc design. However, the large prospective longitudinal observational cohort evaluated in this study, along with its specific inclusion criteria concerning atrophy, are inherent strengths of this study. Another limitation of this study is that the analyses were only performed on a specific subcohort of eyes showing OPL subsidence, and our findings thus cannot be generalized to eyes where OPL subsidence does not develop during AMD progression. Moreover, this work does not include predictive models of OPL subsidence, but rather comprehensively reports on the quantification of early atrophic changes in regions precisely correlating to OPL subsidence. However, RORA developed in more than one-half of the eyes showing OPL subsidence by the end of the 3-year observation period and the lacking difference in early morphological changes between eyes with and without RORA may let us speculate that atrophy will develop in an even higher percentage of eyes given a longer follow-up duration. 
In conclusion, we found that OPL subsidence, an early imaging marker of atrophy, is even preceded by PR and ONL thinning compared with regions with drusen only or otherwise unaffected regions. These changes are early and subtle, yet robustly identifiable by means of AI analysis. Such identification and quantification of these subclinical biomarkers is a task that is not feasible for human graders or retinal experts in their clinical routine. Furthermore, our analysis is an important step toward personalized prediction of disease progression, improves patient management in AMD, and might prove relevant to evaluate subclinical effects of early treatment and identify suitable patients for future interventional trials targeting atrophic AMD. 
Acknowledgments
Presented at the Annual Macula Society Meeting 2021 and the 2021 ARVO Annual Meeting. 
Trial Registration: Laser Intervention in Early Stages of Age-Related Macular Degeneration (LEAD) study (clinicaltrials.gov identifier: NCT01790802). 
Disclosure: S. Riedl, None; U. Schmidt-Erfurth, Genentech (C), Heidelberg Engineering (C), Kodiak (C), RetInSight (C), Novartis (C), Roche; A. Rivail, None; K. Birner, None; J. Mai, None; W.-D. Vogl, None; Z. Wu, None; R.H. Guymer, Roche (C), Genentech (C), Apellis (C), Novartis (C), Bayer (C); H. Bogunović, Heidelberg Engineering (F), Apellis (F); G.S. Reiter, RetInSight (F) 
References
Sayegh RG, Sacu S, Dunavolgyi R, et al. Geographic atrophy and foveal-sparing changes related to visual acuity in patients with dry age-related macular degeneration over time. Am J Ophthalmol. 2017; 179: 118–128. [CrossRef] [PubMed]
Sunness JS. The natural history of geographic atrophy, the advanced atrophic form of age-related macular degeneration. Mol Vis. 1999; 5: 25. [PubMed]
Mitchell P, Liew G, Gopinath B, Wong TY. Age-related macular degeneration. Lancet. 2018; 392: 1147–1159. [CrossRef] [PubMed]
Ferris FL, 3rd, Wilkinson CP, Bird A, et al. Clinical classification of age-related macular degeneration. Ophthalmology. 2013; 120: 844–851. [CrossRef] [PubMed]
Guymer RH, Rosenfeld PJ, Curcio CA, et al. Incomplete retinal pigment epithelial and outer retinal atrophy in age-related macular degeneration: Classification of Atrophy Meeting Report 4. Ophthalmology. 2020; 127: 394–409. [CrossRef] [PubMed]
Schmidt-Erfurth U, Waldstein SM, Klimscha S, et al. Prediction of individual disease conversion in early AMD using artificial intelligence. Invest Ophthalmol Vis Sci. 2018; 59: 3199–3208. [CrossRef] [PubMed]
Bogunovic H, Montuoro A, Baratsits M, et al. Machine learning of the progression of intermediate age-related macular degeneration based on OCT imaging. Invest Ophthalmol Vis Sci. 2017; 58: BIO141–BIO150. [CrossRef] [PubMed]
Wu Z, Luu CD, Ayton LN, et al. Optical coherence tomography-defined changes preceding the development of drusen-associated atrophy in age-related macular degeneration. Ophthalmology. 2014; 121: 2415–2422. [CrossRef] [PubMed]
Schmidt-Erfurth U, Klimscha S, Waldstein SM, Bogunovic H. A view of the current and future role of optical coherence tomography in the management of age-related macular degeneration. Eye (Lond). 2017; 31: 26–44. [CrossRef] [PubMed]
Wu Z, Luu CD, Hodgson LAB, et al. Prospective longitudinal evaluation of nascent geographic atrophy in age-related macular degeneration. Ophthalmol Retina. 2020; 4: 568–575. [CrossRef] [PubMed]
Ferrara D, Silver RE, Louzada RN, Novais EA, Collins GK, Seddon JM. Optical coherence tomography features preceding the onset of advanced age-related macular degeneration. Invest Ophthalmol Vis Sci. 2017; 58: 3519–3529. [CrossRef] [PubMed]
Sadda SR, Guymer R, Holz FG, et al. Consensus definition for atrophy associated with age-related macular degeneration on OCT: Classification of Atrophy Report 3. Ophthalmology. 2018; 125: 537–548. [CrossRef] [PubMed]
Wu Z, Goh KL, Hodgson LAB, Guymer RH. Incomplete retinal pigment epithelial and outer retinal atrophy: longitudinal evaluation in age-related macular degeneration. Ophthalmology. 2023; 130: 205–212. [CrossRef] [PubMed]
Schmidt-Erfurth U, Sadeghipour A, Gerendas BS, Waldstein SM, Bogunovic H. Artificial intelligence in retina. Prog Retin Eye Res. 2018; 67: 1–29. [CrossRef] [PubMed]
Schlegl T, Waldstein SM, Bogunovic H, et al. Fully automated detection and quantification of macular fluid in OCT using deep learning. Ophthalmology. 2018; 125: 549–558. [CrossRef] [PubMed]
Schlegl T, Bogunovic H, Klimscha S, et al. Fully automated segmentation of hyperreflective foci in optical coherence tomography images. arxiv Prepr 2018.
Orlando JI, Breger A, Bogunović H, et al. An Amplified-target loss approach for photoreceptor layer segmentation in pathological OCT scans. In: Fu H, Garvin MK, MacGillivray T, Xu Y, Zheng Y, eds. Ophthalmic Medical Image Analysis. Cham: Springer International Publishing; 2019: 26–34.
Reumueller A, Schmidt-Erfurth U, Salas M, et al. Three-dimensional adaptive optics-assisted visualization of photoreceptors in healthy and pathologically aged eyes. Invest Ophthalmol Vis Sci. 2019; 60: 1144–1155. [CrossRef] [PubMed]
Borrelli E, Sacconi R, Zuccaro B, et al. Photoreceptor alteration in intermediate age-related macular degeneration. Sci Rep. 2020; 10: 21036. [CrossRef] [PubMed]
Wu Z, Bogunovic H, Asgari R, Schmidt-Erfurth U, Guymer RH. Predicting progression of age-related macular degeneration using OCT and fundus photography. Ophthalmol Retina. 2021; 5: 118–125. [CrossRef] [PubMed]
Wu Z, Kumar H, Hodgson LAB, Guymer RH. Reticular pseudodrusen on the risk of progression in intermediate age-related macular degeneration. Am J Ophthalmol. 2022; 239: 202–211. [CrossRef] [PubMed]
Guymer RH, Wu Z, Hodgson LAB, et al. Subthreshold nanosecond laser intervention in age-related macular degeneration: the LEAD randomized controlled clinical trial. Ophthalmology. 2019; 126: 829–838. [CrossRef] [PubMed]
Pollreisz A, Reiter GS, Bogunovic H, et al. Topographic distribution and progression of soft drusen volume in age-related macular degeneration implicate neurobiology of fovea. Invest Ophthalmol Vis Sci. 2021; 62: 26. [CrossRef] [PubMed]
Joachim N, Mitchell P, Burlutsky G, Kifley A, Wang JJ. The incidence and progression of age-related macular degeneration over 15 years: The Blue Mountains Eye Study. Ophthalmology. 2015; 122: 2482–2489. [CrossRef] [PubMed]
Spaide RF. Outer retinal atrophy after regression of subretinal drusenoid deposits as a newly recognized form of late age-related macular degeneration. Retina. 2013; 33: 1800–1808. [CrossRef] [PubMed]
Curcio CA, Messinger JD, Sloan KR, McGwin G, Medeiros NE, Spaide RF. Subretinal drusenoid deposits in non-neovascular age-related macular degeneration: morphology, prevalence, topography, and biogenesis model. Retina. 2013; 33: 265–276. [CrossRef] [PubMed]
Reiter GS, Told R, Schranz M, et al. Subretinal drusenoid deposits and photoreceptor loss detecting global and local progression of geographic atrophy by SD-OCT imaging. Invest Ophthalmol Vis Sci. 2020; 61: 11. [CrossRef] [PubMed]
RetInSight-GmbH. RetInSight secures the first MDR CE-certification for its AI-based GA monitor for the management of geographic atrophy. Online Press Release 2023.
Asgari R, Orlando JI, Waldstein S, et al. Multiclass segmentation as multitask learning for drusen segmentation in retinal optical coherence tomography. arXiv e-prints 2019:arXiv:1906.07679.
Orlando JI, Gerendas BS, Riedl S, et al. Automated quantification of photoreceptor alteration in macular disease using optical coherence tomography and deep learning. Sci Rep. 2020; 10: 5619. [CrossRef] [PubMed]
Liao DS, Grossi FV, El Mehdi D, et al. Complement C3 inhibitor pegcetacoplan for geographic atrophy secondary to age-related macular degeneration: a randomized phase 2 trial. Ophthalmology. 2020; 127: 186–195. [CrossRef] [PubMed]
Khanani AM, Patel SS, Staurenghi G, et al. Efficacy and safety of avacincaptad pegol in patients with geographic atrophy (GATHER2): 12-month results from a randomised, double-masked, phase 3 trial. Lancet. 2023; 402: 1449–1458. [CrossRef] [PubMed]
Schlanitz FG, Baumann B, Kundi M, et al. Drusen volume development over time and its relevance to the course of age-related macular degeneration. Br J Ophthalmol. 2016; 101: 198–203. [CrossRef] [PubMed]
Reiter GS, Hacker V, Told R, et al. Longitudinal changes in quantitative autofluorescence during progression from intermediate to late age-related macular degeneration. Retina. 2021; 41: 1236–1241. [CrossRef] [PubMed]
Schmidt-Erfurth U, Bogunovic H, Grechenig C, et al. Role of deep learning-quantified hyperreflective foci for the prediction of geographic atrophy progression. Am J Ophthalmol. 2020; 216: 257–270. [CrossRef] [PubMed]
Curcio CA, Zanzottera EC, Ach T, Balaratnasingam C, Freund KB. Activated retinal pigment epithelium, an optical coherence tomography biomarker for progression in age-related macular degeneration. Invest Ophthalmol Vis Sci. 2017; 58: BIO211–BIO226. [PubMed]
Pasricha MV, Tai V, Sleiman K, et al. Local anatomic precursors to new-onset geographic atrophy in age-related macular degeneration as defined on OCT. Ophthalmol Retina. 2021; 5: 396–408. [CrossRef] [PubMed]
Curcio CA, Medeiros NE, Millican CL. Photoreceptor loss in age-related macular degeneration. Invest Ophthalmol Vis Sci. 1996; 37: 1236–1249. [PubMed]
Gartner S, Henkind P. Aging and degeneration of the human macula. 1. Outer nuclear layer and photoreceptors. Br J Ophthalmol. 1981; 65: 23–28. [CrossRef] [PubMed]
Li M, Huisingh C, Messinger J, et al. Histology of geographic atrophy secondary to age-related macular degeneration: a multilayer approach. Retina. 2018; 38: 1937–1953. [CrossRef] [PubMed]
Figure 1.
 
Three example cases of progressive morphological changes and OPL subsidence occurring during the 36-month follow-up period. Six-monthly representative B-scans of the OCT volumes (A, C, E) and corresponding colored overlays (B, D, F), indicating different topographic areas, defined at the time point of OPL subsidence occurrence: OPL subsidence area (AS, blue); drusen area (AD, yellow) and reference area (AR, green). The occurrence of OPL subsidence (vertical red arrow) precedes topographically corresponding drusen regression (yellow asterisk) and various degrees of choroidal HT (pink arrowheads), indicative of progressive RORA. Thinned ONL and PR in AS compared with AD is evident in visits leading up to OPL subsidence occurrence.
Figure 1.
 
Three example cases of progressive morphological changes and OPL subsidence occurring during the 36-month follow-up period. Six-monthly representative B-scans of the OCT volumes (A, C, E) and corresponding colored overlays (B, D, F), indicating different topographic areas, defined at the time point of OPL subsidence occurrence: OPL subsidence area (AS, blue); drusen area (AD, yellow) and reference area (AR, green). The occurrence of OPL subsidence (vertical red arrow) precedes topographically corresponding drusen regression (yellow asterisk) and various degrees of choroidal HT (pink arrowheads), indicative of progressive RORA. Thinned ONL and PR in AS compared with AD is evident in visits leading up to OPL subsidence occurrence.
Figure 2.
 
En face topographic distribution of OPL subsidence in the RORA subgroup. En face map showing the cumulative topographic distribution of OPL subsidence lesions over all patients. Pseudo heat map colors correspond to the percentage of eyes affected by a lesion at a specific location. Concentric rings indicate the fovea-centered 3 and 6 mm diameters.
Figure 2.
 
En face topographic distribution of OPL subsidence in the RORA subgroup. En face map showing the cumulative topographic distribution of OPL subsidence lesions over all patients. Pseudo heat map colors correspond to the percentage of eyes affected by a lesion at a specific location. Concentric rings indicate the fovea-centered 3 and 6 mm diameters.
Figure 3.
 
The course of OCT feature development during progression of iAMD. Changes over time in PR thickness, ONL thickness, drusen height and intensity of choroidal HT in the OPLS (dashed line), OPLS + RORA (solid line) and NOPLS + NRORA (alternating dashed/solid line) subgroups. Number of eyes included in the respective groups is indicated by n. Differentiated topographic zones are compared, as each time course is plotted with respect to mean values within OPL subsidence area (AS, red), drusen area (AD, yellow) and reference area (AR, blue). The time axes of the plots are relative to the time point of OPL subsidence occurrence (in the OPLS and OPLS + RORA subgroup) or the last available visit (in the NOPLS + NRORA subgroup), denoted by the dotted vertical line at month 0, respectively. Bars indicate 95% CIs.
Figure 3.
 
The course of OCT feature development during progression of iAMD. Changes over time in PR thickness, ONL thickness, drusen height and intensity of choroidal HT in the OPLS (dashed line), OPLS + RORA (solid line) and NOPLS + NRORA (alternating dashed/solid line) subgroups. Number of eyes included in the respective groups is indicated by n. Differentiated topographic zones are compared, as each time course is plotted with respect to mean values within OPL subsidence area (AS, red), drusen area (AD, yellow) and reference area (AR, blue). The time axes of the plots are relative to the time point of OPL subsidence occurrence (in the OPLS and OPLS + RORA subgroup) or the last available visit (in the NOPLS + NRORA subgroup), denoted by the dotted vertical line at month 0, respectively. Bars indicate 95% CIs.
Table.
 
Output of the LMM for PR and ONL Thickness for the OPLS and OPLS + RORA Subgroups
Table.
 
Output of the LMM for PR and ONL Thickness for the OPLS and OPLS + RORA Subgroups
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×