Topographic Measurement of the Subretinal Pigment Epithelium Space in Normal Aging and Age-Related Macular Degeneration Using High-Resolution OCT

Abstract

Purpose: A micrometer scale hyporeflective band within the retinal pigment epithelium basal lamina – Bruch's membrane complex (RPE-BL-BrM) was topographically measured in aging and age-related macular degeneration (AMD).

Methods: In a prospective cross-sectional study, 90 normal eyes from 76 subjects (range = 23–90 years) and 53 dry AMD eyes from 47 subjects (range = 62–91 years) were enrolled. Isotropic volume raster scans over 6 mm × 6 mm (500 × 500 A-scans) were acquired using a high-resolution (2.7 µm axial resolution) spectral-domain optical coherence tomography (SD-OCT) prototype instrument. Six consecutive optical coherence tomography (OCT) volumes were computationally motion-corrected and fused to improve feature visibility. A boundary regression neural network was developed to measure hyporeflective band thickness. Topographic dependence was evaluated over a 6-mm-diameter Early Treatment Diabetic Retinopathy Study (ETDRS) grid.

Results: The hyporeflective band thickness map (median of 4.3 µm and 7.8 µm in normal and AMD eyes, respectively) is thicker below and radially symmetric around the fovea. In normal eyes, age-associated differences occur within 0.7 to 2.3 mm from the foveal center (P < 0.05). In AMD eyes, the hyporeflective band is hypothesized to be basal laminar deposits (BLamDs) and is thicker within the 3-mm ETDRS circle (P < 0.0002) compared with normal eyes. The inner ring is the most sensitive location to detect age versus AMD-associated changes within the RPE-BL-BrM. AMD eyes with subretinal drusenoid deposits (SDDs) have a significantly thicker hyporeflective band (P < 0.001) than those without SDDs.

Conclusions: The hyporeflective band is a quantifiable biomarker which differentiates AMD from aging. Longitudinal studies are warranted. The hyporeflective band may be a useful biomarker for risk stratification and disease progression.

Age-related macular degeneration (AMD) causes severe visual impairment and can progress to central vision loss in older populations.^1,2 AMD affects the retinal pigment epithelium (RPE), photoreceptors, Bruch's membrane (BrM), and choriocapillaris^3,4 in the context of extracellular deposits on either aspect of the RPE. Although complement inhibitors have been recently approved for pre-existing atrophy, an effective treatment does not currently exist for early nonexudative AMD. 

Basal laminar deposits (BLamDs) accumulate between the RPE cell body and its RPE basal lamina starting in aging and contribute to soft drusen formation in AMD.^5,6 Previous studies utilizing near ultrastructural resolution histology⁶ show BLamD exists as thin patches (thickness of approximately 0.3 µm) in normal older eyes, whereas continuous and thicker (approximately 4–5 µm) BLamD is observed in AMD eyes. Furthermore, thick BLamD particularly under the fovea is a risk factor for disease progression.^5–7 Commercial optical coherence tomography (OCT) instruments with standard axial resolution (approximately 5 to 7 µm) cannot reliably resolve and quantify BLamD because it is so thin. Therefore, this feature of potential prognostic significance is often combined with the RPE-BrM complex. 

High-resolution spectral domain optical coherence tomography (SD-OCT) improves visualization of outer retinal layers,⁸ showing three distinctive bands in the RPE basal lamina (RPE-BL) – Bruch's membrane (BrM) complex (RPE-BL-BrM) in normal aged and AMD eyes.^8,9 Comparison of the hyporeflective band vi^6,9 within the RPE-BL-BrM with histology and ultrastructure images suggests this band is attributed to the accumulation of BLamD in AMD eyes, whereas, in normal eyes, it corresponds to the basal RPE and RPE basal infoldings, which degenerate with aging.^9–11 Because the BLamD is continuous and thick in AMD eyes, we refer to it as a hyporeflective band rather than a split. Our previous study used single, high-resolution B-scans and showed the hyporeflective band thickens and visibility improves in early AMD eyes.⁹ The previous study observed thin bands of the outer retina that are still under investigation and were thus named as band i to vii from anterior to posterior (described in fig. 2B of Chen et al.).⁹ The hyporeflective band within the RPE-BL-BrM, the focus of the current study, was named as a hyporeflective band vi. 

Investigating new imaging and functional biomarkers is critical for evaluating potential therapeutic targets and understanding disease pathogenesis and progression. One example of a functional biomarker is rod-mediated dark adaptation (RMDA), which measures retinal sensitivity recovery after bright light stimulus.^12,13 Delayed RMDA is an early functional biomarker in AMD eyes and identifies normal eyes with increased AMD risk.^13–15 Several studies reported that delayed RMDA depends on the target locations, showing a larger impairment and longer recovery time at 3 degrees to 5 degrees compared to 10 degrees to 12 degrees (just outside of a 6 mm × 6 mm imaging field).^16–19 

Several studies showed the distributions of soft drusen and subretinal drusenoid deposits (SDDs) follow the distributions of cone and rod photoreceptors on the retina,^20,21 respectively. Further, as recently reviewed,²² the risk for AMD progression is highly concentrated within the 3-mm-diameter macula lutea, conveniently captured by the grading grid of Early Treatment of Diabetic Retinopathy Study (ETDRS). Thus, topographic measurement for understanding AMD pathogenesis and progression has taken on new importance. 

In this study, we measure the hyporeflective band topography with a high-resolution prototype OCT using isotropic volume raster scanning combined with computational motion correction and volume fusion,^23,24 and a custom-designed neural network. This study explores the topographic relation of the hyporeflective band thickness within the ETDRS grid in normal aging and AMD to elucidate associations with other imaging or functional biomarkers for therapeutic development and clinical trials.^22,25 

Methods

Study Design

Subjects of different ages with normal eyes and patients with AMD were enrolled at the New England Eye Center (Boston, MA) under a study protocol approved by the institutional review boards at the Tufts Medical Center and Massachusetts Institute of Technology. Written informed consent was obtained from all participants. Normal eyes were determined by medical history and fundus examination. AMD diagnosis and severity were determined by a retinal specialist based on color fundus photography.²⁶ The presence of SDDs in AMD eyes were assessed by en face OCT and B-scans. Exclusion criteria included exudative AMD, signs of neovascularization, presence of ocular pathology other than AMD, and significant ocular opacity. Participants with diabetes but without diabetic retinopathy were not excluded for the normal cohort. 

High-Resolution Prototype OCT Instrument and Scan Protocol

A high-resolution SD-OCT prototype⁸ was utilized for this study. Previous publications^8,9,27 used the term “ultrahigh-resolution” for axial resolutions of approximately 3 µm, however, we chose “high-resolution” to be consistent with terminology used by commercial instruments with approximately 3 µm axial resolution, which recently became available. The high-resolution OCT prototype has approximately 2.7 µm axial resolution and operates at 128 kHz A-scan rate. A volume raster scan protocol covering 6 mm × 6 mm with 500 × 500 A-scans, corresponding to an isotropic (in X-Y) 12 µm A-scan spacing, was used. Six volumes with alternating orthogonal fast scan directions were acquired during a single scan. To preserve axial resolution, a linear scale was used to display OCT images. Machine learning (ML)-based segmentation of BrM²⁸ was performed on each B-scan. 

Computational 3D Motion Correction and Volume Fusion

OCT volumes were computationally corrected for eye motion during post processing.²³ The motion correction technique utilizes consecutive OCT raster scanned volumes with alternating orthogonal (X-Y, horizontal, and vertical) fast scan directions, which have low motion distortion in the temporal-nasal and superior-inferior directions. Distortion from patient eye motion is small in individual OCT B-scans (the fast scan direction) and is progressively more pronounced between OCT B-scans which are acquired further apart (the slow scan direction). Computational motion correction assigns displacements to each A-scan in an OCT volume and estimates these displacements by optimizing similarity of the motion corrected OCT volumes subject to motion constraints. Our computational technique motion corrects the multiple OCT volumes with sub-pixel accuracy²³ (sub-micrometer-scale in the axial direction, and micrometer-scale in the transverse direction). 

Six OCT volumes are then fused to generate a single volume with improved signal to noise and feature visibility. Image fusion differs from image averaging and includes additional processing steps to prevent image brightness artifacts (illumination correction and inconsistent data removal).²⁴ A key step is compensating time- and location-dependent changes in the OCT signal (referred to as illumination correction) from corneal drying, changing vignetting, moving vitreous opacities, and other effects.²⁴ This corrects a combination of common image artifacts, such as quilting, stretching, and banding.^23,24 BrM segmentations were performed individually on the initial volumes and merged using the motion correction displacements from the corresponding OCT volumes. Flattening to BrM was incorporated into the final merging to minimize blurring of the layer boundaries during resampling. The resulting datasets were fovea-centered for topographic analysis. The foveal center was determined based on the inward rise of the external limiting membrane. A general factor of 1 mm = 3.5 degrees was used for conversion. Axial eye length information was not available. 

Depth-Map Regression Network

Building upon our previous ML-based segmentations of BrM and posterior boundary of RPE (pRPE),²⁸ we custom-designed a Depth-Map Regression Network (DMR-net)²⁹ to measure hyporeflective band thickness. We estimate a continuous value of layer boundary as a regression task, instead of voxel-wise layer segmentation. Our architecture uses a 3D OCT volume as an input and generates a 2D en face map of pRPE boundary locations located above the BrM, which was used to compute thickness between the 2 boundaries (pRPE-BrM). 

Initial ML-based segmentations of BrM and pRPE assuming continuous layer boundaries²⁸ were utilized and refined based on band visibility. The hyporeflective band can be axially very thin (approximately <5 µm) and is sometimes not visible to human readers. The reader interactively adjusted the layer boundaries that most precisely matched the reader's perception of overlaid boundaries. Including this visibility information in the training dataset was essential to differentiate very thin, but visible bands, from cases where the band was not visible, enabling the network output to better align with human perception. 

A total of 33 fused OCT volumes representing different age groups and AMD disease severity were used for training, and 9 volumes were used as a test set in DMR-net.²⁹ Utilizing the subset of the 42 representative volumes, corresponding to 16,500 B-scans for training and 4500 B-scans for testing the network, provided sufficient accuracy and repeatability.²⁹ Acquiring manual labels for the entire dataset (71,500 B-scans) was not feasible. All datasets used to train the network and presented in this study were acquired using the same OCT prototype and imaging protocol. The training was performed with a 5-fold cross-validation and mean squared error for the loss function. Root mean squared error and mean absolute error between the ground truth and prediction labels in the test set were measured to evaluate the network performance. Thickness greater than 25 µm, empirically chosen after reviewing OCT images, was masked if associated with drusen. The use of drusen masks was necessary in order to assess the BLamD thickness in AMD eyes. Otherwise, our measurements would be dominated by the presence of soft drusen, not the presence of BLamD. 

Statistical Analysis

Statistical analysis was performed using the Statistics Toolbox in MATLAB (R2023a, MathWorks). A P value < 0.05 was considered statistically significant. A two-sided Wilcoxon rank sum test was used to determine if the hyporeflective band thickness in early AMD eyes was different from age-matched normal eyes, and if the thickness in AMD eyes with SDDs was different from AMD eyes without SDDs. Two-way analysis of variance for unbalanced design, followed by the pairwise comparison from Tukey's honestly significant difference testing, were performed to determine the dependence of normal aging on the hyporeflective band measurements in the ETDRS grid. All analyses were performed using median measurements. 

Results

Participants

A total of 90 normal eyes from 76 subjects (age range = 23–90 years) and 53 nonexudative (dry) AMD eyes from 47 patients (age range = 62–91 years) were included in the study. Participant demographics and diagnoses are shown in the Table. Due to the difficulties in recruiting older subjects, eyes from subjects with diabetes mellitus without clinical retinopathy (28 eyes; 31%) were included in the normal group. Studies to date have not reported significant differences in the macula sub-RPE deposits in eyes of subjects with diabetes mellitus versus in age-matched normal eyes of subjects. 

Table.

 

Subject Demographics and Summary

 

Improved Layer Visibility With Volume Fused High-Resolution OCT

Our 3D motion correction technique enables OCT volume fusion without image artifacts, improving retinal layer visibility compared with single (non-averaged) B-scans (Fig. 1). Figure 1D shows the hyporeflective band vi (red triangle) within RPE-BL-BrM, highlighting the benefits of flattening to BrM, linear display range, and axial stretching when examining axially thin features in the outer retina. 

Figure 1.

 

Visualizing and measuring the hyporeflective band vi within the RPE-BL-BrM complex. Multiple high-resolution SD-OCT volume raster scans are computationally motion corrected and fused in order to increase signal to noise and improve feature visibility. Data from a young healthy volunteer (25-year-old man) is shown. (A) Single B-scan displayed in standard logarithmic scale to show large dynamic range. (B) B-scan flattened to BrM and axially stretched to examine the outer retina. A linear scale display is used to preserve axial resolution. (C) B-scan extracted from the motion corrected fused volume and displayed in logarithmic scale. Signal to noise and feature continuity is improved compared with A. (D) B-scan flattened to BrM, axially stretched, and displayed in linear scale. Visibility of the hyporeflective band vi and other fine structure is improved compared with B. BrM = Bruch's membrane; COST = cone outer segment tip; EZ = ellipsoid zone; IS/OS = photoreceptor inner segment/outer segment junction; ROST = rod outer segment tip; RPE = retinal pigment epithelium; RPE-BL-BrM = RPE basal lamina-Bruch's membrane complex. This figure clearly shows bands originally called COST and ROST, and the latter is visible even in the fovea. Future work will address this terminology.

Volumetric Estimation of Hyporeflective Band vi Thickness

Figure 2 shows representative B-scans, overlaid with boundary labels generated from the DMR-net, and en face thickness maps of the hyporeflective band. In young normal eyes, the hyporeflective band was consistently visible in the 6-mm-diameter ETDRS grid, whereas it was less visible in the inner ring of ETDRS grid in older normal eyes. In AMD eyes, a thicker band, hypothesized to be BLamD, was frequently identified below the fovea and where the soft drusen were present.⁹ 

Figure 2.

 

Visualization and topographic measurement of hyporeflective band vi thickness in normal aging and AMD. All data are from motion corrected, fused volumes. Columns (left to right) show a B-scan displayed in logarithmic scale, B-scan flattened to BrM displayed in linear scale, axially stretched B-scan, same B-scan showing overlaid BrM and posterior RPE (pRPE) segmentations generated by Depth-Map Regression Network. Right column shows en face topographic map of the hyporeflective band thickness with white dotted line showing the B-scan position. (A) Young healthy subject (35-year-old man) with normal macula. (B) Older healthy subject (55-year-old woman) with normal macula. White arrows indicate visible hyporeflective band, and the yellow arrow indicates reduced visibility. (C) Early AMD eye of an 86-year-old man showing a thick hyporeflective band, which is hypothesized to be basal laminar deposit (BLamD). (D) Intermediate AMD eye of a 69-year-old man also showing a thick hyporeflective band. Drusen were masked in AMD eyes to focus on BLamD estimation in AMD eyes.

The root mean squared error and mean absolute error between ground truth and prediction layer segmentation in test datasets were less than the OCT optical resolution. To assess repeatability of the DMR-net, the hyporeflective band thickness maps of three intermediate AMD eyes (78-year-old woman and 74-year-old woman) were evaluated using repeated OCT acquisitions, with instrument re-alignment between the repeats. The median differences between two repeated hyporeflective band thickness measurements were 1.4 ± 1.4 µm, 0.3 ± 0.7 µm, and 0.9 ± 1.5 µm (Supplementary Fig. S1). In addition, two young normal eyes (in a 25-year-old woman and a 33-year-old woman) were imaged again after a several month interval (5 months and 6.5 months), assuming no significant changes in the hyporeflective band vi. The median thickness difference was 0.2 ± 0.2 µm for both eyes (Supplementary Fig. S2), again suggesting high repeatability of OCT acquisition, volume fusion, and the DMR-net. 

Topographic Representation of Hyporeflective Band vi in Normal Aging and AMD Eyes

Figure 3 shows the median band thickness in normal aging and AMD groups, overlaid with the central subfield, inner, and outer ring of the ETDRS grid. In the normal age groups, the map is radially symmetric from the foveal center, with a subtle but thicker band (11% thicker; 95% confidence interval [CI] = −1% to 23%) under the fovea compared to the inner ring. This eccentricity-dependence is also evident when the thickness is plotted radially at different angles (Fig. 3B). In AMD eyes, the hyporeflective band is dramatically thicker, particularly within the 3-mm ETDRS circle (Figs. 3A, 4). 

Figure 3.

 

Topographic map of the hyporeflective band thickness showing eccentricity dependence. (A) En face map of the median hyporeflective band thickness in different age groups of normal and AMD eyes. Age related thinning of the hyporeflective band is most evident in the ETDRS inner ring, with less pronounced changes in the central ETDRS subfield and outer ring. AMD eyes have thicker hyporeflective band, hypothesized to be basal laminar deposits (BLamD), and basal linear deposits within the ETDRS 3-mm circle. Increased thickness is observed in intermediate AMD (iAMD) compared with early AMD. (B) Median hyporeflective band thickness in different age groups of normal eyes versus foveal eccentricity. Measurements along radial directions at 5-degree angle intervals were performed. Age-associated differences are more pronounced in the ETDRS inner ring than the outer ring.

Figure 4.

 

Topography of hyporeflective band thickness in normal aging versus AMD. Box plots of median hyporeflective band thickness in (A), central 0.5 mm diameter circle, (B) ETDRS central subfield (1 mm diameter circle), (C) ETDRS inner ring (0.5 mm to 1.5 mm eccentricity from foveal center), and (D) ETDRS outer ring (1.5 mm to 3 mm eccentricity from foveal center) for subjects in different age groups and early AMD. Circles indicate box plot outliers. The hyporeflective layer thickness in the ETDRS inner ring can differentiate normal eyes (60-year-old or older) versus early AMD eyes. n.s. = not significant.

Among the central subfield, inner, and outer ring of the ETDRS grid, the greatest differences between the age-matched normal eyes and early AMD eyes occurred in the inner ring by a factor of 2.2 (95% [CI] = 70% to 290%), whereas differences were not significant in the outer ring (see Fig. 4). A statistically significant difference between early AMD eyes and age-matched normal eyes was observed up to a 2.1 mm eccentricity from the foveal center (P < 0.05), with greater statistical power (P < 0.0002) within the 3-mm ETDRS circle (Supplementary Fig. S3). This is largely due to much thicker deposits within the 3-mm ETDRS circle than the outer ring in AMD eyes, compared with normal eyes (also highlighted in Fig. 3). Consistent results were obtained using the datasets excluding diabetic eyes (Supplementary Fig. S4). 

In normal eyes, two-way analysis of variance showed that aging and ETDRS rings significantly affect the hyporeflective band thickness (P = 0.0026 and P = 0.0001, respectively). The median band thickness in the central subfield and inner ring was significantly higher in the 20s group compared to the 70s group (16% and 62% thicker in the central and inner ring, respectively). The thickness was statistically different among all the age groups within 0.7 to 2.3 mm from the foveal center (P < 0.05; see Supplementary Fig. S3). 

Continuously Thick Hyporeflective Band vi on AMD Eyes

SDDs are a known risk factor for disease progression in AMD,^30,31 and in eyes with SDDs the hyporeflective band is consistently thick and clearly visible across the 6-mm-diameter ETDRS grid (Fig. 5). This contrasts with the thick hyporeflective band concentrated in the 3-mm-diameter ETDRS circle in AMD eyes without macular SDDs (see Figs. 2C, 2D). Intermediate AMD eyes with SDDs showed a significantly thicker hyporeflective band in the ETDRS inner ring (P < 0.001; median of 64% thicker) and outer ring (P < 0.001; 43% thicker), compared with intermediate AMD eyes without SDDs. Although early AMD eyes with SDDs showed overall greater hyporeflective band thickness within the 6-mm-diameter ETDRS circle than early AMD eyes without SDDs, the difference was greatest in the outer ring (P = 0.014; 27% thicker). This is consistent with SDDs being more likely to be present in the outer ring.²⁰ 

Figure 5.

 

AMD eyes with subretinal drusenoid deposits (SDDs). (A, B) (left) En face OCT whole projection, (center) B-scan at the dotted white line position displayed in logarithmic scale, and axially stretched, flattened B-scan displayed in linear scale. (Right) Hyporeflective band thickness map over 6 mm × 6 mm. (A) Early AMD eye with macular SDDs (81-year-old man). (B) Intermediate AMD eye with macular SDDs (62-year-old woman). Compared with AMD eyes without macular SDDs (Figs. 2C and 2D), SDDs are associated with increased hyporeflective band/basal laminar deposit (BLamD) thickness even outside the ETDRS 3-mm circle. (C) Box plot of median hyporeflective band thickness in different ETDRS grid regions, stratified by AMD severity and presence of SDDs.

Discussion

High-resolution OCT is becoming commercially available, and multiple groups are investigating potential clinical applications.^32,33 Therefore, studies of high-resolution features and potential markers of disease are of increasing interest. By acquiring sequential, orthogonal raster scanned volumes and using computational 3D motion correction with volume fusion, it is possible to increase signal to noise and improve visibility of fine, micrometer scale features, without blurring from eye motion. Whereas this study utilized 6-volume fusion, 4 volumes may be sufficient for clinical applications assessing global thickness measurements. 

A deep learning-based 3D boundary regression method enabled topographic measurements of the hyporeflective band vi in the RPE-BL-BrM complex. Convolutional neural networks are widely used in OCT research as well as in medical imaging datasets for various segmentations.³⁴ These approaches assign a discrete class or label to every pixel (or voxel), such that layer boundaries are derived from pixel- (or voxel-) based segmentation results, often assuming smooth, continuous network outputs. The DMR-net is designed to predict a single, distinct boundary position for each axial scan. It does not require an additional step of estimating boundaries from pixel-based classification by directly handling continuous boundary labels and predicting boundary positions, thereby improving accuracy over pixel-wise labeling architectures. Furthermore, human perception-derived visibility was integrated into the training datasets to predict boundary positions that better coincide with what humans perceive as a visible hyporeflective band. 

The hyporeflective band was thicker under the fovea and in the ETDRS outer ring in normal eyes. A population study of a similar age range (range = 30–95 years) using SD-OCT also showed the thickest RPE (likely RPE-BrM complex) in the central subfield, but decreasing RPE thickness toward the outer ring, without significant age-associated changes.³⁵ In this study, we observed ETDRS grid-dependent age-associated differences in the hyporeflective band vi, particularly decreasing band thickness in the inner ring with aging. This suggests finer delineation of outer retinal bands will be critical to resolve age-associated changes in RPE cells, photoreceptor outer segments contacting RPE apical processes, and deposits between the RPE-BL and BrM. 

The general trend of decreasing hyporeflective band thickness with normal aging is consistent with the disappearance of RPE basal infoldings with aging.^10,11 The increased visibility of the hyporeflective band at the temporal and nasal region in B-scans of normal eyes was previously observed,⁹ and is consistent with the greater thickness within the outer ring shown in this study. Furthermore, the topography revealed a thicker hyporeflective band under the fovea than the inner ring of ETDRS grid in normal eyes. Histological studies show higher incidences of subfoveal BLamD in older normal eyes,⁶ which may explain why a consistently thick hyporeflective band is observed under the fovea, whereas the thickness decreases in the inner ring with aging. 

A statistically significant thickness difference between normal older eyes and early AMD eyes was observed throughout the central 4.2 mm diameter, with higher statistical power within the 3-mm diameter ETDRS circle. In normal eyes, statistical differences with aging occurred mostly in the ETDRS inner ring. Collectively, the ETDRS inner ring is the most sensitive location to detect both age- and AMD-associated structural changes within the RPE-BL-BrM complex. 

Interestingly, the greatest hyporeflective band differences between normal older eyes and early AMD eyes within the 3-mm ETDRS circle (0 degrees to 5.3 degrees) resemble the test location dependence of RMDA measurements in AMD eyes, where slower RMDA and greater dysfunctional rod vision was observed at 3 degrees to 5 degrees compared with 10 degrees to 12 degrees.^{17,18,22,36,37} The baseline data from the Alabama Study on Early Age-Related Macular Degeneration (ALSTAR2) also showed greater statistical power in RMDA measurements at 5 degrees than 12 degrees for differentiating normal older eyes versus AMD eyes.¹⁹ These previous findings, combined with the current study, again suggest the 3-mm ETDRS circle as the most sensitive region affected by aging and AMD. With imaging and functional measurements of RMDA in the same population over time, it may be possible to determine which changes occur earlier. Topographic measurement of the hyporeflective band within the RPE-BL-BrM is candidate for an early structural biomarker that may also be associated with rod vision impairment at 3 degrees to 5 degrees, the earliest functional biomarker in AMD.^13,38,39 

One hypothesis is that measurement of the hyporeflective band can detect very early changes in BrM, RPE, and deposits, such as BLamD, located between those layers, which may impair photoreceptors and metabolic exchange across the RPE-BL-BrM and choriocapillaris.^39,40 This is consistent with previous studies that showed OCT intensities located near the RPE-BL-BrM band, ellipsoid zone, and interdigitation zone^41–43 within the outer retinal bands, were highly predictive of longer rod intercept time (RIT), indicating delayed RMDA. The hyporeflective band might have been included in previous measurements of overall drusen volume. However, the hyporeflective band in its thinnest form cannot be resolved with standard OCT devices and measuring topography which has a 0 to 5 µm range of thickness would be challenging. 

Consistently thicker deposits under the fovea were observed in most AMD eyes, which agrees with our previous study using high A-scan density, high-resolution B-scans⁹ and with histological evidence that advanced BLamD morphologies also cluster under the fovea.⁶ As expected, a thicker hyporeflective band is observed when drusen are present. AMD eyes with SDDs have a thicker hyporeflective band than those without SDDs, also supported by a recent study reporting thick BLamD in eyes with early-onset SDDs.⁴⁴ However, 17% of AMD eyes (n = 9; and 6 eyes from patients with early AMD) had a median hyporeflective band thickness less than 5 µm over a 6 mm × 6 mm region, suggesting potential variations with disease severity and AMD phenotypes. 

The high-resolution OCT and advanced computational methods presented in this study can be extended to detect other deposits or fluid in the sub-RPE space (Supplementary Fig. S5). There has been increasing interest in detecting thick BLamD using commercial OCT instruments, with the goal of differentiating BLamD from double layer sign (DLS) or shallow irregular retinal pigment epithelium elevation (SIRE), suggestive of choroidal neovascularization.^6,45–48 SIRE is defined as <100 µm thickness with >1 mm transverse length.⁴⁵ A recent study measured a median thickness of 67 µm in irregular elevation of the RPE (including DLS and SIRE) in 268 eyes with and without nonexudative macular neovascularization⁴⁹; the hyporeflective band in this study is overall much thinner. Developing standardized methods for differentiating thick BLamD from DLS or SIRE for stratifying progression risk will require improvements in en face mapping of the hyporeflective band vi thickness and layer visibility using high-resolution OCT, as well as the availability of corresponding OCT angiography (OCTA) data. 

Some commercial SD-OCT instruments have approximately 3 µm axial image resolution and, in principle, could measure the hyporeflective layer as well as other micrometer scale features. We elected to use volume raster scanning with computational motion correction and volume fusion because this gives data for volumetric feature analysis, providing a platform for future analysis of features in addition to the hyporeflective band. Computational motion correction is more accurate than hardware eye tracking and enables volumetric image fusion. By contrast, most commercial instruments average repeated B-scans using eye tracking hardware that compensates transverse, but not axial eye position. Furthermore, there is a latency time for detecting and compensating saccadic motion or tremor, so B-scans that are displaced out of plane might be averaged. Thus, B-scan averaging can result in blurring of fine features and introduce measurement variance. However, if commercial instruments save individual repeated B-scans, it should be possible to design new image averaging algorithms that detect out of plane motion and more precisely motion correct the repeated B-scans. This could generate averaged images which better preserve micrometer level features and enable quantitative measurements. The use of a linear display scale and flattening images to BrM, which enhances visibility of fine structures such as the hyporeflective band, can also be readily translated to commercial instruments. 

There are several limitations of the study. Structural OCT cannot yet differentiate BLamD from basal linear deposits, and thus our thickness measurements might overestimate BLamD in AMD eyes. At the same time, if a hyporeflective band vi is not resolvable with high-resolution OCT, our thickness measurements might underestimate BLamD. There is lack of corresponding functional measurements and OCTA. Although axial eye length was not available, only one eye each in the age-matched normal and the AMD group had moderate refractive error (refractive error of more than 3D). Analysis in a cohort with larger refractive errors would require using a conversion factor which depends on axial eye length. The methods used in this study are not currently available on commercial instruments, however, they are software based and can be applied to high resolution OCT commercial instruments in the future because hardware modification is not required. Finally, this is a cross-sectional study and limited by the progression rates in early and intermediate AMD. We are currently collecting longitudinal data to assess the role of the hyporeflective band in disease progression. 

It is important to highlight that unlike functional measurements that use a limited number of prespecified target locations to reduce measurement time, the hyporeflective band can be rapidly measured topographically over the entire ETDRS grid. If hyporeflective band thickness is associated with known markers for AMD progression risk, such as RMDA, high-resolution structural measurements could facilitate design of larger clinical studies to assess early AMD progression as well as assist in testing therapeutic interventions. 

Summary Statements

This study utilizes a high-resolution OCT prototype instrument, advanced image processing methods, and neural networks to quantify the sub-RPE space in normal aging and AMD eyes. The ETDRS inner ring showed the most sensitive age- and AMD-associated change in the sub-RPE space within a 6-mm ETDRS circle. In AMD eyes, significantly thicker sub-RPE space was observed within the 3-mm ETDRS circle, likely dominated by BLamDs. 

Acknowledgments

The authors thank Muhammad U. Jamil, Jessica M. Girgis, Kenneth Lam, and Joseph Woo from the New England Eye Center for their help with recruitment and Ajay Y. Manicka from Massachusetts Institute of Technology for his help with image processing. The authors acknowledge all the staff and physicians in the Retinal Clinic at the New England Eye Center for their help in subject recruitment and clinical assistance. 

Supported by the National Institutes of Health grants R01EY011289 and R01EY034080, the Beckman-Argyros Award in Vision Research (Irvine, CA), and Greenberg Prize to End Blindness to J.G.F. National Institutes of Health grants R01EY029595 and R01028282 to C.A.C. and Deutsche Forschungsgemeinschaft project 508075009 to S.B.P. The sponsor or funding organization had no role in the design or conduct of this research. 

Meeting Presentation: Presented at the 2024 Annual Meeting of the Association for Research in Vision and Ophthalmology – Imaging in the Eye. 

Disclosure: J. Won, None; H. Takahashi, None; S.B. Ploner, None; W. Karbole, None; O. Abu-Qamar, None; A. Yaghy, None; A. Marmalidou, None; S. Kaiser, None; Y. Hwang, None; J. Lin, None; A. Witkin, None; S. Desai, None; C.R. Baumal, None; A. Maier, None; C.A. Curcio, Heidelberg Engineering (F), Apellis (C), Astellas (C), Boehringer Ingelheim (C), Character Biosciences (C), Osanni (C), Annexon (C), Mobius (C), Ripple (C), and Genentech Roche (C), all outside the current work; N.K. Waheed, Apellis (C), Nidek (C), Boehringer Ingelheim (C), Carl Zeiss Meditec (F), Heidelberg Engineering (F), Nidek (F), Optovue (F), Topcon (F), Regeneron (F), all outside the current work, Ocudyne (I), Beacon Therapeutics (S); J.G. Fujimoto, Royalties from intellectual property owned by MIT and licensed to Optovue (R), Topcon (F), all outside the current work