February 2024
Volume 65, Issue 2
Open Access
Retina  |   February 2024
RPE Curvature Can Screen for Early and Intermediate AMD
Author Affiliations & Notes
  • Rene Cheung
    School of Optometry and Vision Science, University of New South Wales, Sydney, Australia
    Centre for Eye Health, University of New South Wales, Sydney, Australia
  • Matt Trinh
    School of Optometry and Vision Science, University of New South Wales, Sydney, Australia
    Centre for Eye Health, University of New South Wales, Sydney, Australia
  • Yoh Ghen Tee
    School of Optometry and Vision Science, University of New South Wales, Sydney, Australia
    Centre for Eye Health, University of New South Wales, Sydney, Australia
  • Lisa Nivison-Smith
    School of Optometry and Vision Science, University of New South Wales, Sydney, Australia
    Centre for Eye Health, University of New South Wales, Sydney, Australia
  • Correspondence: Lisa Nivison-Smith, School of Optometry and Vision Science, UNSW Australia, Sydney, NSW 2052, Australia; l.nivison-smith@unsw.edu.au
Investigative Ophthalmology & Visual Science February 2024, Vol.65, 2. doi:https://doi.org/10.1167/iovs.65.2.2
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Rene Cheung, Matt Trinh, Yoh Ghen Tee, Lisa Nivison-Smith; RPE Curvature Can Screen for Early and Intermediate AMD. Invest. Ophthalmol. Vis. Sci. 2024;65(2):2. https://doi.org/10.1167/iovs.65.2.2.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose: Diagnosing AMD early optimizes clinical management. However, current diagnostic accuracy is limited by the subjectivity of qualitative diagnostic measures used in clinical practice. This study tests if RPE curvature could be an accurate, quantitative measure for AMD diagnosis.

Methods: Consecutive patients without AMD or normal aging changes (n = 111), with normal aging changes (n = 107), early AMD (n = 102) and intermediate AMD (n = 114) were recruited. RPE curvature was calculated based on the sinuosity method of measuring river curvature in environmental science. RPE and Bruch's membrane were manually segmented from optical coherence tomography B-scans and then their lengths automatically extracted using customized MATLAB code. RPE sinuosity was calculated as a ratio of RPE to Bruch's membrane length. Diagnostic accuracy was determined from area under the receiver operator characteristic curve (aROC).

Results: RPE sinuosity of foveal B-scans could distinguish any eyes with AMD (early or intermediate) from those without AMD (non-AMD or eyes with normal aging changes) with acceptable diagnostic accuracy (aROC = 0.775). Similarly, RPE sinuosity could identify intermediate AMD from all other groups (aROC = 0.871) and distinguish between early and intermediate AMD (aROC = 0.737). RPE sinuosity was significantly associated with known AMD lesions: reticular pseudodrusen (P < 0.0001) and drusen volume (P < 0.0001), but not physiological variables such as age, sex, and ethnicity.

Conclusions: RPE sinuosity is a simple, robust, quantitative biomarker that is amenable to automation and could enhance screening of AMD.

Accurate detection of AMD in its early stages is critical for appropriate clinical management, including patient education of dietary and lifestyle modifications to decrease the risk of progression,1 Age-related Eye Diseases Study (AREDS) supplement recommendations2 and timely ophthalmology referral. However, detecting nonexudative AMD remains challenging, with nearly one-quarter of patients in primary and tertiary care misdiagnosed.3 Assistance from automated models does not necessarily improve on this with 34.2% to 55.3% of eyes with early AMD misclassified as healthy by various artificial intelligence applications.4,5 
Poor detection and diagnosis of the early stages of AMD by clinicians relates to the difficulty in quantifying drusen size by eye from color fundus photographs.6,7 Classification is further complicated by the fact that not all drusen (i.e., drusen sized <63 µm) are considered pathological.8 Automated and semiautomated methods of quantifying drusen area,9,10 volume1012 and number12,13 from RPE thickness can aid clinical interpretation,14 but may be confounded by physiological variations in RPE thickness from age, ethnicity, and axial length/refraction.15 RPE thickness-based measures also lack sensitivity to small RPE elevations, meaning that they typically underestimate the contributions of small drusen.16 Alternative optical coherence tomography (OCT)-based AMD biomarkers such as outer retinal disruptions, shallow irregular RPE elevations,17 and hyper-reflective foci17,18 have been proposed, but mostly rely on subjective, qualitative interpretation making them time-consuming and less amenable to automation and future integration into artificial intelligence–based assistive technologies. 
RPE alterations induced by drusen not only affect thickness, but also the curvature of the hyper-reflective RPE band on OCT. Thus, measuring the curvature of the RPE band could potentially indicate the presence of drusen and other AMD lesions in OCT B-scans. Measures of curvature already exist in other fields of science, most notably environmental science, where the curvature or meandering of a river is quantified as the ratio of river length:straight line distance and termed sinuosity.19 The ratio component makes sinuosity a robust measure, as it allows direct comparison of bands regardless of variations in the length of B-scans or their width. 
By adapting sinuosity to quantify RPE curvature, that is, the ratio of RPE band length to Bruch's membrane band length on OCT, curvature changes induced by AMD lesions can be assessed. In theory, as these two structures are apposed in the healthy retina,20 sinuosity should be close to one. In contrast, alterations in the curvature of the RPE bands from drusen should drive this ratio above 1. Importantly, the impact of physiological variabilities is theoretically minimized in this ratio-based measure because it is not thickness based. Indeed, other OCT biomarkers that use ratios21 rather than thickness measures22,23 have been shown to be effective for assessment of ocular diseases and are relatively unaffected by physiological variability. 
Thus, the aim of this retrospective, cross-sectional study was to optimize methods for quantifying RPE sinuosity from OCT B-scans and subsequently, measure its diagnostic accuracy for early and intermediate stages of AMD. 
Methods
Study Population
Consecutive, unique patients seen at the Centre for Eye Health between January 1, 2016, and May 31, 2022, were screened for inclusion into the study (n = 28,568) until the target sample size of 212 patients with AMD was reached. The Centre for Eye Health is an optometry-led referral center that provides intermediate-tier eye care services, including advanced diagnostic imaging and/or disease management in conjunction with ophthalmologists in the local health district.24 All patients consented to use of their deidentified data for research and the study was approved by a University of New South Wales Human Research Ethics Advisory Committee. 
Four groups of individuals were recruited by identifying consecutive patients meeting inclusion criteria including individuals with a diagnosis of (1) early AMD, (2) intermediate AMD (iAMD), (3) normal aging changes based on small drusen or drupelets, or (4) no AMD or aging changes (termed the non-AMD group), which are the same categories developed by the Beckman Initiative for Macular Research Classification Committee to classify nonexudative or atrophic AMD.8 Inclusion criteria for all groups were age over 50 years, refractive error of less than 6 diopters, and availability of a macular Spectralis OCT (Heidelberg Engineering, Inc., Heidelberg, Germany) volume scan with a B-scan quality score of 15 or higher. For all groups, patients with other concomitant macular pathology or systemic disease were included to mitigate selection bias and determine the robustness of the biomarker for AMD screening in typical populations. Concomitant macular pathology was defined as ocular diagnoses potentially affecting the macula, such as epiretinal membranes, optic nerve head disease, or glaucoma suspects (a complete list of pathologies is presented in Supplementary Table S1). Systemic conditions were defined as the presence of cardiovascular disease or diabetes. The presence of these factors was accounted for in multivariate analysis. Eyes with late AMD were excluded. 
Sample Size and Study Eye Selection
A pilot sample of 40 eyes—20 consecutive individuals in the iAMD group, and 20 individuals in the non-AMD group—was assessed. For the final study, the required sample size was estimated from the outcomes of the pilot sample.25 For both studies, the more diseased, eligible eye was selected for inclusion. The study eye was selected randomly for all other individuals where both eyes met the inclusion criteria. 
Data Extraction
Available patient characteristics including date of birth, sex, ethnicity, systemic disease status, visual acuity, refractive error, date of consultation, study and fellow eye diagnosis, and self-reported ethnicity were extracted from clinical records and patient reports using practice management software, Bp VIP.net (Version 2.1.530.017; Best Practice Software, Hamilton, New Zealand) and recorded in Microsoft Excel. 
Eyes were also classified for the presence of key AMD structural markers: pigmentary abnormalities, reticular pseudodrusen and drusen volume. Color fundus and the highest quality en face macular OCT scan available (Cirrus HD-OCT, Carl Zeiss Meditec, Dublin, CA) were graded manually for the presence of pigmentary abnormalities. Macular OCT B-scans were graded for presence of reticular pseudodrusen using previously established definitions for OCT grading, that is, reticular lesions defined as five or more hyper-reflective mounds or triangular lesions above the RPE on one or more B-scans (Spectralis OCT).26 Drusen volume within the central 5mm was recorded using advanced RPE analysis software (Cirrus HD-OCT, Carl Zeiss Meditec). 
OCT B-Scan Selection and Segmentation
OCT B-scan selection and segmentation of the RPE and Bruch's membrane was performed independently by two experienced clinician graders (R.C., M.T.), who were randomly allocated 50% of study eyes. Macular volumes of the highest axial resolution (5 µm) from the Spectralis OCT device (Heyex, Version 2.5.1; Heidelberg Engineering GmbH, Heidelberg, Germany) available for each study eye were reviewed. For the pilot study, the foveal B-scan and four horizontal B-scans superior and four inferior to the fovea (interscan distance of 120–240 µm) were extracted. For the final study, only the foveal B-scan was extracted for the study population. For the pilot and final study, B-scans were segmented using proprietary software (Heyex, Version 2.5.1; Heidelberg Engineering GmbH). All segmentations of the RPE and Bruch's membrane were generated automatically by proprietary software and then independently reviewed by two experienced graders who manually corrected any segmentation errors as previously described.27,28 Each grader reviewed the other's segmentation and any disagreements in segmentations were resolved by discussion and consensus between graders before further data extraction. OCT images and associated B-scan characteristics of selected volumes including the OCT macular volume scan protocol, B-scan length, and x- axis scaling ratio (describing image compression [µm/pixel]) were extracted from OCT records using Heyex. 
RPE Sinuosity Quantification
The RPE and Bruch's membrane segmentation images were uploaded to MATLAB (MathWorks, Natwick, MA). RPE and Bruch's membrane lengths were subsequently measured and extracted using customized code (Fig. 1). The ratio of RPE to Bruch's membrane length was then calculated to obtain RPE sinuosity for each B-scan, whereby the ratio is expected to be close to 1 in healthy retina where the RPE and Bruch's membrane are apposed.20 Ratios were expressed as a percentage difference from 1 (RPE–Bruch's membrane sinuosity difference), that is, an RPE sinuosity of 1.05 represents a 5% difference in the curvature of the RPE to Bruch's membrane. For the pilot study, RPE sinuosity was expressed as a single value from the foveal scan and the mean, median, or maximum value calculated across multiple scans. For the final study, only RPE sinuosity of the foveal B-scan was calculated. 
Figure 1.
 
(A, B) The original foveal OCT B-scan was extracted and then binarized to obtain the (C, D) RPE and (E, F) Bruch's membrane (BM) segmentation line. Note differences in the RPE segmentation line for an eye with iAMD (left) and no AMD (right).
Figure 1.
 
(A, B) The original foveal OCT B-scan was extracted and then binarized to obtain the (C, D) RPE and (E, F) Bruch's membrane (BM) segmentation line. Note differences in the RPE segmentation line for an eye with iAMD (left) and no AMD (right).
Statistical Analysis
Paired and independent sample t tests and Pearson's χ2 test were used to assess continuous and categorical variables against outcomes, respectively, with P values of <0.05 considered significant. The diagnostic accuracy of RPE sinuosity was calculated by evaluating the area under the receiver operator characteristic (aROC) curve.25 The accuracy of using RPE sinuosity to (1) screen for early or iAMD, (2) identify iAMD, and (3) discriminate stages of AMD (i.e., early from iAMD) was interpreted using aROC according to predefined criteria: 0.7 to 0.8 as acceptable, 0.8 to 0.9 as excellent, and greater than 0.9 as outstanding.29 A multivariable linear mixed-effects model was used to assess associations between baseline characteristics to RPE sinuosity in addition to concomitant macular pathology, systemic disease (presence of cardiovascular or diabetic disease), the presence of reticular pseudodrusen, pigmentary abnormalities, drusen volume in the central 5 mm, fellow eye AMD status, and RPE sinuosity for all participants. All statistical analyses were performed using SPSS (version 25; IBM, Armonk, NY). 
Results
Pilot Study
A pilot study was first conducted to optimize the protocol for calculating RPE sinuosity. Optimization aimed to achieve maximum diagnostic accuracy of RPE sinuosity alongside minimal image processing to increase the potential of the measure for clinical translation. 
First, the role of B-scan length was explored by assessing RPE sinuosity across the entire B scan length vs. B-scans cropped horizontally to the central 1 mm. Larger extents of 3 mm have also been examined30; however, we chose to review the central 1 mm of the foveal region, because this is where outer retinal disruptions in retinal disease are typically examined.31,32 RPE sinuosity was significantly greater for the central 1 mm vs. the entire B-scan length in AMD groups (P = 0.04) (Fig. 2A), but diagnostic accuracy was unaffected (1 mm aROC, 0.917; entire scan aROC, 0.893) (Fig. 2B). 
Figure 2.
 
Comparison of RPE–Bruch's membrane (BM) sinuosity difference values between iAMD (orange circles) and non-AMD (blue circles) subgroups for whole vs. the central 1 mm of foveal B-scans (A) and foveal B-scan vs. multiple (five) B-scans within the vertical 1 mm spaced 120 µm or 240 µm apart (C, E, and G; left). The diagnostic accuracy of using the foveal B-scan, mean, median and maximum sinuosity to predict iAMD (B, D, F, and H; right).
Figure 2.
 
Comparison of RPE–Bruch's membrane (BM) sinuosity difference values between iAMD (orange circles) and non-AMD (blue circles) subgroups for whole vs. the central 1 mm of foveal B-scans (A) and foveal B-scan vs. multiple (five) B-scans within the vertical 1 mm spaced 120 µm or 240 µm apart (C, E, and G; left). The diagnostic accuracy of using the foveal B-scan, mean, median and maximum sinuosity to predict iAMD (B, D, F, and H; right).
Second, the role of B-scan density was examined by comparing RPE sinuosity of the foveal scan alone to summary RPE sinuosity measures derived from vertically adjacent B-scans with an anatomic coverage of 1 mm in the vertical direction including mean, median and maximum sinuosity. Summary measures were derived from either nine B-scans spaced 120 µm or five spaced 240 µm apart based on previous reports that large drusen length is 140 µm or larger on SD-OCT.33 There was no significant difference in RPE sinuosity between a single foveal scan vs. the mean (P = 0.08) or median (P = 0.10) RPE sinuosity of multiple B-scans (Supplementary Table S2). Although there was a significant difference between the RPE sinuosity of a single foveal scan vs. the maximum RPE sinuosity across multiple B-scans for iAMD (P < 0.0001), there was no difference in the diagnostic accuracy for identifying iAMD between these measures (Figs. 2C–H). Based on our examination of these two conditions in the pilot study (modifying horizontal length and density of B-scans included in sinuosity calculations), RPE sinuosity for the final study was calculated from the entire length of foveal scans, because this minimized image processing, but still afforded high diagnostic accuracy. Furthermore, the potential impact of image artefacts and noise on segmentation accuracy is also minimized by this approach. 
Study Population
For the final study, 434 eyes were consecutively recruited for each group and included: 102 with early AMD, 114 with iAMD, 107 with normal aging changes, and 111 in the non-AMD group. Pairwise comparisons indicated that intermediate and early AMD groups were significantly different from the non-AMD group for age (P < 0.0001), sex (P = 0.002–0.01), ethnicity (P = 0.004–0.006), the presence of systemic disease (P = 0.004–0.01), concomitant macular disease (P < 0.0001), reticular pseudodrusen (P < 0.001), and fellow eye status (P < 0.0001). The iAMD and early AMD groups were significantly different from the normal aging changes group for age (P < 0.0001), systemic disease (P = 0.04–0.05), reticular pseudodrusen (P < 0.0001), and fellow eye status (P < 0.0001) only. Between the two AMD groups, age, ethnicity, fellow eye status, reticular pseudodrusen, pigmentary abnormalities, and drusen volume were all significantly different (P < 0.0001–0.01). As such, all of these factors were considered in subsequent multivariable analyses (Table 1). Note that Bruch's membrane length differed between OCT scan protocols and may be influenced by axial length, which is associated with ocular magnification effects34 and retinal curvature variations.35 
Table 1.
 
Differences in Patient and Scan Characteristics Across Subgroups
Table 1.
 
Differences in Patient and Scan Characteristics Across Subgroups
RPE Sinuosity Between AMD Severities
The distribution of RPE sinuosity differences for each subgroup are shown in Figure 3. RPE sinuosity was significantly greater in early and iAMD eyes vs. those without AMD pathology (non-AMD and normal aging subgroups) (Fig. 3; blue vs orange data points; P < 0.0001). RPE sinuosity was also significantly greater for iAMD vs. early AMD (P < 0.0001). The median RPE–Bruch's membrane sinuosity difference was less than 1% for all groups, except for iAMD, and higher for early and iAMD groups vs. eyes without AMD. 
Figure 3.
 
Scatter plots of the foveal RPE–Bruch's membrane (BM) sinuosity difference for non-AMD and normal aging subgroups (blue circles) and AMD subgroups (orange circles). Mean values (red dashed lines), CI (black brackets), and median RPE-BM sinuosity differences values are shown beneath subgroup labels.
Figure 3.
 
Scatter plots of the foveal RPE–Bruch's membrane (BM) sinuosity difference for non-AMD and normal aging subgroups (blue circles) and AMD subgroups (orange circles). Mean values (red dashed lines), CI (black brackets), and median RPE-BM sinuosity differences values are shown beneath subgroup labels.
Diagnostic Performance
Based on the separable distributions of RPE sinuosity between subgroups, we then assessed the diagnostic accuracy of using RPE sinuosity to screen for AMD and discriminate between AMD severities (Fig. 4). For AMD screening (i.e., the identification of eyes with any stage of AMD), foveal RPE sinuosity could identify eyes with early or iAMD from non-AMD eyes or those exhibiting normal aging changes with acceptable performance (aROC = 0.775). RPE sinuosity could also screen for iAMD from non-AMD subgroups pooled together (aROC = 0.871), as well as from the normal aging changes or the non-AMD groups alone with excellent performance (aROC = 0.820–0.881). RPE sinuosity, however, was unable to distinguish between eyes with early AMD from eyes without AMD or normal aging changes with acceptable diagnostic accuracy (aROC = 0.652–0.686). For differentiating AMD severity, foveal RPE sinuosity could discriminate iAMD eyes from early AMD eyes with acceptable performance (aROC = 0.737). Further information on the diagnostic performance of discriminating between subgroups is included in Supplementary Table S3
Figure 4.
 
Diagnostic performance of RPE sinuosity for discriminating between subgroups including (A) any AMD, (B) early AMD, and (C) iAMD vs normal aging changes and non-AMD and (D) early vs iAMD.
Figure 4.
 
Diagnostic performance of RPE sinuosity for discriminating between subgroups including (A) any AMD, (B) early AMD, and (C) iAMD vs normal aging changes and non-AMD and (D) early vs iAMD.
Factors Affecting RPE Sinuosity
Finally, to identify person-level, eye-level, and image-level characteristics impacting RPE sinuosity and its ability to screen for AMD, iAMD, or differentiate AMD severity, univariate and multivariate analyses were performed. Univariable analysis revealed that RPE sinuosity was affected by age (P < 0.0001), ethnicity (Asian; P < 0.0001), systemic disease (P = 0.01), fellow eye AMD status (P < 0.0001–0.002), pigmentary abnormalities (P < 0.0001), reticular pseudodrusen (P < 0.0001), and drusen volume (P < 0.0001) (Table 2). OCT B-scan protocol was not included as a variable in univariate or multivariate analysis as 93% of images were captured using the 30° × 25° scan type (Supplementary Table S4) and B-scan length is already accounted for. 
Table 2.
 
Univariate and Multivariate Analysis of Factors Affecting RPE Sinuosity
Table 2.
 
Univariate and Multivariate Analysis of Factors Affecting RPE Sinuosity
After adjusting for these covariables, foveal RPE sinuosity was no longer significantly affected by age, ethnicity, systemic disease, or pigmentary abnormalities (P = 0.13), but remained strongly associated with other AMD disease features, including fellow eye AMD status (P < 0.0001–0.04), reticular pseudodrusen (P < 0.0001), and drusen volume (P < 0.0001) (Table 2). 
Discussion
This study evaluated RPE sinuosity, a novel method of quantifying RPE curvature, as a method for screening and differentiating between disease stages in AMD before conversion to late disease. Pilot results revealed that B-scan length or density did not significantly alter the diagnostic accuracy of RPE sinuosity for identifying these cases. Subsequent analysis of the full study population confirmed that the diagnostic performance of foveal RPE sinuosity alone was acceptable to excellent for screening and could distinguish AMD severity classification with minimal potential confounding effects from nondisease covariables. These results suggest that RPE sinuosity is a promising, quantitative AMD biomarker. 
Foveal RPE Sinuosity Can Screen for iAMD But Not Early AMD
Accurately detecting intermediate or early AMD optimizes patient management by ensuring appropriate lifestyle and dietary recommendations are provided1 and review schedules are adjusted relative to the risk of vision loss.36,37 Distinguishing early AMD from healthy eyes is important to minimize psychological distress, as even a diagnosis of early AMD is associated with a 17% decrease in quality of life.38 
Our study found RPE sinuosity of the foveal B-scan could identify any AMD (early and iAMD combined) or iAMD alone with acceptable to excellent diagnostic accuracy. Other OCT-based screening approaches have reported greater levels of accuracy for identifying iAMD (area under the curves of 0.93–0.98)39,40; however, these investigations have typically combined iAMD and late AMD cases into one category, oversampled late AMD relative to the community prevalence of less than 1%41,42 and excluded eyes with concomitant macular pathology.40,43 In contrast, our population only included individuals with earlier AMD stages as well as individuals with aging changes or other macular pathology, better reflecting the challenges associated with screening in a real-world context. Additional work assessing RPE sinuosity on larger real-world datasets with images of varying quality will further determine the robustness of this biomarker for screening purposes. 
RPE sinuosity was unable to achieve acceptable performance for early AMD screening. Other studies have also reported low diagnostic accuracy for early AMD with 39% and 47% to 79% of cases misclassified by clinicians and deep learning algorithms, respectively.4,5 This finding reflects the challenge of identifying subtle outer retinal disruptions in early AMD for which aggregate RPE sinuosity metrics may be worthwhile exploring, particularly if sinuosity measurements can be obtained efficiently using automated methods,30 promoting accessibility. Assessing the diagnostic accuracy of RPE sinuosity for different horizontal extents, multimodal imaging and functional measures, such as rod-mediated dark adaptation44 and retinal sensitivity tested under mesopic lighting,45,46 could also be used to improve early AMD identification, although the practical aspects of implementing specialized functional testing in screening settings needs to be considered.47,48 
RPE Sinuosity Can Distinguish Between Early and iAMD
RPE sinuosity could identify intermediate from early AMD with acceptable performance. This point is significant, considering that iAMD is the diagnostic threshold at which AREDS supplements are recommended for decreasing the risk of progression to late disease49,50 and supplements decrease the cost of intravitreal injections over the lifetime of a patient.51 Clinical signs of AMD are misdiagnosed frequently,52 particularly in early AMD,53 and nearly one-quarter of early AMD cases are misdiagnosed as iAMD by deep-learning algorithms exploring AMD classification.5,39 Previous work using OCT imaging for AMD diagnosis has shown high specificity (86.9%) and low sensitivity (21.1%) in identifying early AMD. In comparison, RPE sinuosity could distinguish early AMD from all other cohorts with only 77.8% specificity at the same sensitivity level.39 These errors have been attributed to difficulties in differentiating medium from large drusen, which distinguishes early from intermediate disease, and the use of severity grading systems based on AMD features visible on color fundus images rather than OCT or other imaging modalities being explored. Using RPE sinuosity for AMD screening and stratification represents a shift in approach from qualitative, lesion-based assessment to quantitative methods that can be used reliably to improve overall diagnostic accuracy. 
RPE Sinuosity was Unaffected by Physiological Variables That Impact Other AMD Biomarkers
We hypothesized that RPE sinuosity would be resistant to physiological and ocular factors, including age, sex, ethnicity, and refraction that impact other global biomarkers commonly assessed in AMD—central drusen volume, RPE–Bruch's membrane complex thickness, and retinal thickness.15,54,55 Our multivariate analysis confirmed this finding and also showed that the measure is robust to systemic disease, visual acuity, and the presence of concomitant macular pathology. Expectedly, RPE sinuosity was strongly associated with core AMD biomarkers used for classification, which are known to affect the RPE, notably reticular pseudodrusen and drusen volume, as well as other biomarkers for late AMD prognostication, such as fellow eye AMD status.5658 There may be a lack of association to pigmentary abnormalities; pigment is known to migrate toward the inner retina with disease progression, which is difficult to capture using a metric derived from the outer retina.59 This unexpected finding also suggests that pigmentary abnormalities are less specific to AMD within an unselected population because the whole study cohort was evaluated for AMD biomarkers, including the non-AMD and aging changes group. Nevertheless, the associations with other prominent AMD biomarkers supports the hypothesis that RPE sinuosity can serve as an indicator of outer retinal changes induced by AMD and may translate to different populations more effectively than existing quantitative biomarkers that are impacted by physiological variation. 
Limitations
Some limitations that may impact the translatability of results to real-world settings include the use of a single OCT device, OCT images with relatively good quality segmentation, and a predominantly White population. Using a range of OCT devices and output images with varying quality, as well as examining a more diverse population, would provide a more representative evaluation of the screening and diagnostic potential of RPE sinuosity. This approach could also help to determine whether the biomarker could be used with specific screening technologies, such as portable OCT devices, because they typically yield lower resolution images.60 Alterations to RPE curvature also occur in other retinal disease and, thus, may impact sinuosity and, consequently, the diagnostic accuracy of AMD screening.61 However, a breakdown of concomitant diagnoses among patients shows that the majority of conditions in our community-based population primarily affects the inner retina and are not expected to impact RPE integrity (glaucoma suspect, glaucoma or other optic nerve disorders [69.1%] and vitreoretinal interface disorders including epiretinal membranes [13.1%]). Furthermore, the index test is targeted to patients aged 50 years and older to screen for drusenoid changes associated with age-related disease. Misclassifying eyes without AMD that exhibit RPE pathology is also associated with few negative implications for patients as screening tests serve as initial evaluations and these cases are, thus, still appropriately flagged for further assessment.30,62 Collecting axial length data in future work may also better adjust for factors potentially impacting RPE sinuosity.63 The discriminability of aggregate RPE sinuosity metrics for identifying early AMD could also be explored further to determine its potential for early AMD screening and determine the minimum number of B-scans needed to optimize diagnostic accuracy given that its diagnosis is challenging currently. 
Conclusions
RPE sinuosity identified iAMD with excellent diagnostic accuracy and distinguished early from iAMD with acceptable accuracy from a single foveal OCT B-scan. Multivariate analysis also confirmed the metric was associated with lesions for AMD classification and resistant to physiological factors, such as age. These findings suggest RPE sinuosity is a simple, novel, robust biomarker for screening and classifying the early stages of AMD and could be amenable to future automation. 
Acknowledgments
Supported by the National Health and Medical Research Council (NH&MRC) Project Grant (#1174385) and the Australian Government Research Training Program (RTP) Scholarship. Guide Dogs NSW/ACT provides support for the Centre for Eye Health, the location of recruitment. 
This work was presented as a conference abstract at the Imaging in the Eye Conference, Association for Research in Vision and Ophthalmology, April 21–22, 2023, New Orleans, Louisiana. 
Disclosure: R. Cheung, None; M. Trinh, None; Y.G. Tee, None; L. Nivison-Smith, None 
References
Chakravarthy U, Wong TY, Fletcher A, et al. Clinical risk factors for age-related macular degeneration: a systematic review and meta-analysis. BMC Ophthalmol. 2010; 10: 1–13. [CrossRef] [PubMed]
Chew EY, Clemons TE, Agrón E, et al. Long-term effects of vitamins C and E, β-carotene, and zinc on age-related macular degeneration: AREDS report no. 35. Ophthalmology. 2013; 120: 1604–1611. [CrossRef] [PubMed]
Neely DC, Bray KJ, Huisingh CE, et al. Prevalence of undiagnosed age-related macular degeneration in primary eye care. JAMA Ophthalmol. 2017; 135: 570–575. [CrossRef] [PubMed]
Venhuizen FG, van Ginneken B, van Asten F, et al. Automated staging of age-related macular degeneration using optical coherence tomography. Invest Ophthalmol Vis Sci. 2017; 58: 2318–2328. [CrossRef] [PubMed]
Burlina PM, Joshi N, Pacheco KD, et al. Use of deep learning for detailed severity characterization and estimation of 5-year risk among patients with age-related macular degeneration. JAMA Ophthalmol. 2018; 136: 1359–1366. [CrossRef] [PubMed]
Klein R, Meuer SM, Myers CE, et al. Harmonizing the classification of age-related macular degeneration in the three-continent AMD consortium. Ophthalmic Epidemiol. 2014; 21: 14–23. [CrossRef] [PubMed]
Klein R, Davis MD, Magli YL, et al. The Wisconsin age-related maculopathy grading system. Ophthalmology. 1991; 98: 1128–1134. [CrossRef] [PubMed]
Ferris FL, III, Wilkinson C, Bird A, et al. Clinical classification of age-related macular degeneration. Ophthalmology. 2013; 120: 844–851. [CrossRef] [PubMed]
Jain N, Farsiu S, Khanifar AA, et al. Quantitative comparison of drusen segmented on SD-OCT versus drusen delineated on color fundus photographs. Invest Ophthalmol Vis Sci. 2010; 51: 4875–4883. [CrossRef] [PubMed]
Gregori G, Wang F, Rosenfeld PJ, et al. Spectral domain optical coherence tomography imaging of drusen in nonexudative age-related macular degeneration. Ophthalmology. 2011; 118: 1373–1379. [CrossRef] [PubMed]
Chen Q, Leng T, Zheng L, et al. Automated drusen segmentation and quantification in SD-OCT images. Med Image Anal. 2013; 17: 1058–1072. [CrossRef] [PubMed]
Farsiu S, Chiu SJ, Izatt JA, Toth CA. Fast detection and segmentation of drusen in retinal optical coherence tomography images. Ophthalmic Technologies XVIII. SPIE; 2008: 48–59.
Diniz B, Ribeiro R, Heussen FM, Maia M, Sadda S. Drusen measurements comparison by fundus photograph manual delineation versus optical coherence tomography retinal pigment epithelial segmentation automated analysis. Retina. 2014; 34: 55–62. [CrossRef] [PubMed]
Garzone D, Terheyden JH, Morelle O, et al. Comparability of automated drusen volume measurements in age-related macular degeneration: a MACUSTAR study report. Sci Rep. 2022; 12: 21911. [CrossRef] [PubMed]
Ko F, Foster PJ, Strouthidis NG, et al. Associations with retinal pigment epithelium thickness measures in a large cohort: results from the UK Biobank. Ophthalmology. 2017; 124: 105–117. [CrossRef] [PubMed]
Wintergerst MWM, Schultz T, Birtel J, et al. Algorithms for the automated analysis of age-related macular degeneration biomarkers on optical coherence tomography: a systematic review. Transl Vis Sci Technol. 2017; 6: 10–10. [CrossRef] [PubMed]
Amarasekera S, Samanta A, Jhingan M, et al. Optical coherence tomography predictors of progression of non-exudative age-related macular degeneration to advanced atrophic and exudative disease. Graefes Arch Clin Exp. 2021; 260: 1–10.
Ferrara D, Silver RE, Louzada RN, et al. Optical coherence tomography features preceding the onset of advanced age-related macular degeneration. Invest Ophthalmol Vis Sci. 2017; 58: 3519–3529. [CrossRef] [PubMed]
Mueller JE. An introduction to the hydraulic and topographic sinuosity indexes. Ann Assoc Am Geographers. 1968; 58: 371–385. [CrossRef]
Bhutto I, Lutty G. Understanding age-related macular degeneration (AMD): relationships between the photoreceptor/retinal pigment epithelium/Bruch's membrane/choriocapillaris complex. Mol Aspects Med. 2012; 33: 295–317. [CrossRef] [PubMed]
Kita Y, Naghizadeh F, Kita R, Tomita G, Holló G. Comparison of macular ganglion cell complex thickness to total retinal thickness ratio between Hungarian and Japanese eyes. Japan J Ophthalmol. 2013; 57: 540–545. [CrossRef]
Paulsen AJ, Pinto A, Merten N, et al. Factors associated with the macular ganglion cell-inner plexiform layer thickness in a cohort of middle-aged US adults. Optom Vis Sci. 2021; 98: 295. [CrossRef] [PubMed]
Perez CI, Chansangpetch S, Mora M, et al. Ethnicity-specific database improves the diagnostic ability of peripapillary retinal nerve fiber layer thickness to detect glaucoma. Am J Ophthalmol. 2021; 221: 311–322. [CrossRef] [PubMed]
Wang H, Kalloniatis M. Clinical outcomes of the Centre for Eye Health: an intra-professional optometry-led collaborative eye care clinic in Australia. Clin Exp Optom. 2021;1–10.
Hajian-Tilaki K. Sample size estimation in diagnostic test studies of biomedical informatics. J Biomed Inform. 2014; 48: 193–204. [CrossRef] [PubMed]
Ueda-Arakawa N, Ooto S, Tsujikawa A, et al. Sensitivity and specificity of detecting reticular pseudodrusen in multimodal imaging in Japanese patients. Retina. 2013; 33: 490–497. [CrossRef] [PubMed]
Trinh M, Khou V, Zangerl B, Kalloniatis M, Nivison-Smith L. Modelling normal age-related changes in individual retinal layers using location-specific OCT analysis. Sci Rep. 2021; 11: 1–16. [PubMed]
Tong J, Phu J, Khuu SK, et al. Development of a spatial model of age-related change in the macular ganglion cell layer to predict function from structural changes. Am J Ophthalmol. 2019; 208: 166–177. [CrossRef] [PubMed]
Mandrekar JN. Receiver operating characteristic curve in diagnostic test assessment. J Thorac Oncol. 2010; 5: 1315–1316. [CrossRef] [PubMed]
Lin AC, Lee CS, Blazes M, Lee AY, Gorin MB. Assessing the clinical utility of expanded macular OCTs using machine learning. Transl Vis Sci Technol. 2021; 10: 32–32. [CrossRef] [PubMed]
Wu Z, Ayton LN, Guymer RH, Luu CD. Second reflective band intensity in age-related macular degeneration. Ophthalmology. 2013; 120: 1307–1308.e1301. [CrossRef] [PubMed]
Wang J-W, Jie C-H, Tao Y-J, et al. Macular integrity assessment to determine the association between macular microstructure and functional parameters in diabetic macular edema. Int J Ophthalmol. 2018; 11: 1185. [PubMed]
Kim DY, Loo J, Farsiu S, Jaffe GJ. Comparison of single drusen size on color fundus photography and spectral-domain optical coherence tomography. Retina. 2021; 41: 1715–1722. [CrossRef] [PubMed]
Salmon AE, Sajdak BS, Atry F, Carroll J. Axial scaling is independent of ocular magnification in OCT images. Invest Ophthalmol Vis Sci. 2018; 59: 3037–3040. [CrossRef] [PubMed]
Lake S, Bottema M, Williams K, Reynolds K. The correlation between optical coherence tomography retinal shape irregularity and axial length. PloS One. 2019; 14: e0227207. [CrossRef] [PubMed]
Shen L, Liu F, Nardini HG, Del Priore LV. Natural history of geographic atrophy in untreated eyes with nonexudative age-related macular degeneration: a systematic review and meta-analysis. Ophthalmol Retina. 2018; 2: 914–921. [CrossRef] [PubMed]
Wong T, Chakravarthy U, Klein R, et al. The natural history and prognosis of neovascular age-related macular degeneration: a systematic review of the literature and meta-analysis. Ophthalmology. 2008; 115: 116–126. [CrossRef] [PubMed]
Brown MM, Brown GC, Stein JD, et al. Age-related macular degeneration: economic burden and value-based medicine analysis. Can J Ophthalmol. 2005; 40: 277–287. [CrossRef] [PubMed]
Venhuizen FG, van Ginneken B, van Asten F, et al. Automated staging of age-related macular degeneration using optical coherence tomography. Invest Ophthalmol Vis Sci. 2017; 58: 2318–2328. [CrossRef] [PubMed]
Lee CS, Baughman DM, Lee AY. Deep learning is effective for classifying normal versus age-related macular degeneration OCT images. Ophthalmol Retina. 2017; 1: 322–327. [CrossRef] [PubMed]
Rein DB, Wittenborn JS, Burke-Conte Z, et al. Prevalence of age-related macular degeneration in the US in 2019. JAMA Ophthalmol. 2022; 140: 1202–1208. [CrossRef] [PubMed]
Keel S, Xie J, Foreman J, et al. Prevalence of age-related macular degeneration in Australia: the Australian National Eye Health Survey. JAMA Ophthalmol. 2017; 135: 1242–1249. [CrossRef] [PubMed]
Vaghefi E, Hill S, Kersten HM, Squirrell D. Multimodal retinal image analysis via deep learning for the diagnosis of intermediate dry age-related macular degeneration: a feasibility study. J Ophthalmol. 2020; 2020: 7493419. [CrossRef] [PubMed]
Owsley C, Swain TA, McGwin G, et al. How vision is impaired from aging to early and intermediate age-related macular degeneration: insights from ALSTAR2 baseline. Transl Vis Sci Technol. 2022; 11: 17–17. [CrossRef] [PubMed]
Vujosevic S, Smolek MK, Lebow KA, et al. Detection of macular function changes in early (AREDS 2) and intermediate (AREDS 3) age-related macular degeneration. Ophthalmologica. 2011; 225: 155–160. [CrossRef] [PubMed]
Roh M, Laíns I, Shin HJ, et al. Microperimetry in age-related macular degeneration: association with macular morphology assessed by optical coherence tomography. Br J Ophthalmol. 2019; 103: 1769–1776. [PubMed]
Jamous KF, Kalloniatis M, Hayen A, et al. Application of clinical techniques relevant for glaucoma assessment by optometrists: concordance with guidelines. Ophthalmic Physiol Opt. 2014; 34: 580–591. [CrossRef] [PubMed]
Ly A, Nivison-Smith L, Zangerl B, Assaad N, Kalloniatis M. Self-reported optometric practise patterns in age-related macular degeneration. Clin Exp Optom. 2017; 100: 718–728. [CrossRef] [PubMed]
Optometry Australia, Melbourne, Australia. Clinical practice guide for the diagnosis, treatment and management of age-related macular degeneration, [updated 2019 May; cited 2023 August 1]. Available from: https://www.optometry.org.au/wp-content/uploads/Professional_support/Practice_notes/AMD-Clinical-Practice-Guide-2019_final_designed_v5.pdf.
Group A-REDSR. A randomized, placebo-controlled, clinical trial of high-dose supplementation with vitamins C and E, beta carotene, and zinc for age-related macular degeneration and vision loss: AREDS report no. 8. Arch Ophthalmol. 2001; 119: 1417–1436. [CrossRef] [PubMed]
Lee AY, Butt T, Chew E, et al. Cost-effectiveness of age-related macular degeneration study supplements in the UK: combined trial and real-world outcomes data. Br J Ophthalmol. 2018; 102: 465–472. [CrossRef] [PubMed]
Muen WJ, Hewick SA. Quality of optometry referrals to neovascular age-related macular degeneration clinic: a prospective study. JRSM Short Rep. 2011; 2: 1–4. [CrossRef] [PubMed]
Ly A, Nivison-Smith L, Zangerl B, Assaad N, Kalloniatis M. Advanced imaging for the diagnosis of age-related macular degeneration: a case vignettes study. Clin Exp Optom. 2018; 101: 243–254. [CrossRef] [PubMed]
Tan CS, Cheong KX, Lim LW, Li KZ. Topographic variation of choroidal and retinal thicknesses at the macula in healthy adults. Br J Ophthalmol. 2014; 98: 339–344. [CrossRef] [PubMed]
Wagner-Schuman M, Dubis AM, Nordgren RN, et al. Race-and sex-related differences in retinal thickness and foveal pit morphology. Invest Ophthalmol Vis Sci. 2011; 52: 625–634. [CrossRef] [PubMed]
Corvi F, Corradetti G, Tiosano L, et al. Topography of choriocapillaris flow deficit predicts development of neovascularization or atrophy in age-related macular degeneration. Graefe's Arch Clin Exp Ophthalmol. 2021; 259: 2887–2895. [CrossRef]
Corvi F, Srinivas S, Nittala MG, et al. Reproducibility of qualitative assessment of drusen volume in eyes with age related macular degeneration. Eye. 2021; 35: 2594–2600. [CrossRef] [PubMed]
Finger RP, Wu Z, Luu CD, et al. Reticular pseudodrusen: a risk factor for geographic atrophy in fellow eyes of individuals with unilateral choroidal neovascularization. Ophthalmology. 2014; 121: 1252–1256. [CrossRef] [PubMed]
Christenbury JG, Folgar FA, O'Connell RV, et al. Progression of intermediate age-related macular degeneration with proliferation and inner retinal migration of hyperreflective foci. Ophthalmology. 2013; 120: 1038–1045. [CrossRef] [PubMed]
Lin J, Patel P, Szirth B, Khouri A. The use of handheld optical coherence tomography in urban community eye screening. Invest Ophthalmol Vis Sci. 2013; 54: 1461–1461.
Invernizzi A, dell'Arti L, Leone G, et al. Drusen-like deposits in young adults diagnosed with systemic lupus erythematosus. Am J Ophthalmol. 2017; 175: 68–76. [CrossRef] [PubMed]
De Fauw J, Ledsam JR, Romera-Paredes B, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med. 2018; 24: 1342–1350. [CrossRef] [PubMed]
Fieß A, Nickels S, Schulz A, et al. The relationship of ocular geometry with refractive error in normal and low birth weight adults. J Optom. 2021; 14: 50–57. [CrossRef] [PubMed]
Figure 1.
 
(A, B) The original foveal OCT B-scan was extracted and then binarized to obtain the (C, D) RPE and (E, F) Bruch's membrane (BM) segmentation line. Note differences in the RPE segmentation line for an eye with iAMD (left) and no AMD (right).
Figure 1.
 
(A, B) The original foveal OCT B-scan was extracted and then binarized to obtain the (C, D) RPE and (E, F) Bruch's membrane (BM) segmentation line. Note differences in the RPE segmentation line for an eye with iAMD (left) and no AMD (right).
Figure 2.
 
Comparison of RPE–Bruch's membrane (BM) sinuosity difference values between iAMD (orange circles) and non-AMD (blue circles) subgroups for whole vs. the central 1 mm of foveal B-scans (A) and foveal B-scan vs. multiple (five) B-scans within the vertical 1 mm spaced 120 µm or 240 µm apart (C, E, and G; left). The diagnostic accuracy of using the foveal B-scan, mean, median and maximum sinuosity to predict iAMD (B, D, F, and H; right).
Figure 2.
 
Comparison of RPE–Bruch's membrane (BM) sinuosity difference values between iAMD (orange circles) and non-AMD (blue circles) subgroups for whole vs. the central 1 mm of foveal B-scans (A) and foveal B-scan vs. multiple (five) B-scans within the vertical 1 mm spaced 120 µm or 240 µm apart (C, E, and G; left). The diagnostic accuracy of using the foveal B-scan, mean, median and maximum sinuosity to predict iAMD (B, D, F, and H; right).
Figure 3.
 
Scatter plots of the foveal RPE–Bruch's membrane (BM) sinuosity difference for non-AMD and normal aging subgroups (blue circles) and AMD subgroups (orange circles). Mean values (red dashed lines), CI (black brackets), and median RPE-BM sinuosity differences values are shown beneath subgroup labels.
Figure 3.
 
Scatter plots of the foveal RPE–Bruch's membrane (BM) sinuosity difference for non-AMD and normal aging subgroups (blue circles) and AMD subgroups (orange circles). Mean values (red dashed lines), CI (black brackets), and median RPE-BM sinuosity differences values are shown beneath subgroup labels.
Figure 4.
 
Diagnostic performance of RPE sinuosity for discriminating between subgroups including (A) any AMD, (B) early AMD, and (C) iAMD vs normal aging changes and non-AMD and (D) early vs iAMD.
Figure 4.
 
Diagnostic performance of RPE sinuosity for discriminating between subgroups including (A) any AMD, (B) early AMD, and (C) iAMD vs normal aging changes and non-AMD and (D) early vs iAMD.
Table 1.
 
Differences in Patient and Scan Characteristics Across Subgroups
Table 1.
 
Differences in Patient and Scan Characteristics Across Subgroups
Table 2.
 
Univariate and Multivariate Analysis of Factors Affecting RPE Sinuosity
Table 2.
 
Univariate and Multivariate Analysis of Factors Affecting RPE Sinuosity
×
×

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.

×