Hyporeflective Core Drusen Cause an Artifactual Bending of Bruch's Membrane on OCT Images: Optical Mechanism and Proposed Quantitative Metrics

Abstract

Purpose

To investigate the optical mechanism and impact of the artifactual bending of Bruch's membrane (BM) beneath hyporeflective core drusen (hypoRCD) on optical coherence tomography (OCT).

Methods

This observational and experimental laboratory study analyzed volumetric OCT scans with hypoRCD from an age-related macular degeneration (AMD) cohort. The optical mechanism underlying BM bending was examined using tissue phantoms with different oil droplets to model different refractive indices (RIs) of hyporeflective cores (hypoRCs). Three metrics were proposed to quantify hypoRCD: (1) hypocore bending index (CoreBI), (2) drusen bending index (DruBI), and (3) maximum bending index (B_max). The impact of BM bending on choriocapillaris flow deficit (CCFD) analysis was assessed by comparing CCFD measurements before and after correcting BM segmentation under hypoRCD.

Results

BM bending was consistently observed beneath the core region of hypoRCD in both structural OCT and optical attenuation coefficient (OAC) B-scans. Phantom experiments demonstrated that the bending was associated with differences in the RIs of the hypoRCs. The degree of bending, quantified by the CoreBI, increased with the RI of the core material (P < 0.001). The CCFD percentage beneath hypoRCD was significantly reduced by 25.9% ± 7.7% (P = 0.005) after correcting the BM segmentation. Morphological changes associated with the bending could be quantified using the three proposed metrics.

Conclusions

BM bending induced by hypoRCD is an artifactual OCT feature that reflects the optical properties of the hypoRCs. This morphological characteristic may improve our understanding of hypoRCD, raise cautions in quantitation of CCFDs under hypoRCD, and provide additional insights into their role in AMD.

Soft drusen are the hallmark of age-related macular degeneration (AMD) and usually appear as yellow focal lesions on color fundus photography and as elevations of the retinal pigment epithelium (RPE) on optical coherence tomography (OCT) images.¹ However, in AMD, drusen can be classified into various phenotypes based on their size, shape, reflectivity, and homogeneity in OCT images, and these phenotypes include soft drusen, calcified drusen, cuticular drusen, or reticular drusen.^2–5 Recently, hyporeflective core drusen (hypoRCD) characterized by heterogeneous and low internal reflectivity^6–9 have emerged as a phenotype of particular interest. Several studies have suggested that hypoRCD may serve as a biomarker for increased risk of AMD progression.^7,8,10,11

However, the reported prognostic value of hypoRCD has been inconsistent. Ouyang et al.⁶ and Tan et al.⁷ found that hypoRCD were significantly associated with new onset of geographic atrophy at 6-month and 1-year follow-ups, respectively. Similarly, Veerappan et al.⁸ reported an association between low reflective core drusen and geographic atrophy development, although not with choroidal neovascularization. In contrast, Goh et al.⁹ found no independent association between hypoRCD and late AMD progression after adjusting for other risk factors, such as pigmentary abnormalities and drusen volume. Furthermore, Nittala et al.¹² likewise reported no significant association when using multivariate regression analysis. These inconsistent findings may stem from differences in how hypoRCD are defined and graded. The terms used to identify hypoRCD include hyporeflective foci within drusen,¹¹ low reflective core drusen,⁸ drusen with heterogeneous internal reflectivity,^6,7 drusen with hyporeflective cores (hypoRCs),¹³ and hypoRC drusen with at least 40 µm in maximum height.¹² Such variation suggests that the morphological features of hypoRCD in OCT images are not yet fully characterized. Incorporating additional morphological criteria may improve inter-study consistency in the identification and risk assessment of hypoRCD.

Another poorly understood aspect of hypoRCD is their composition. Histological analysis of cadaveric human eyes suggests that the heterogeneous internal reflectivity core within drusen may be composed of multilobular polycrystalline hydroxyapatite nodules.⁷ However, the composition, density and ratio of contents of the druse likely changes over time. Current binary grading systems, which classify lesions simply based on the presence of a hyporeflective core, do not account for these dynamic or compositional variations, limiting their ability to correlate OCT morphological features with underlying tissue composition. Thus, there is a need for improved characterization of hypoRCD using OCT images.

In this study, we introduce a previously unreported OCT finding: the appearance of Bruch's membrane (BM) bending beneath hypoRCD, hereafter referred to as the BM bending effect. We conducted a tissue phantom experiment demonstrating that this effect likely arises from the high refractive indices (RIs) of hypoRCs. In addition, we proposed three quantitative indices to characterize hypoRCD based on the BM bending effect. Finally, we investigated how segmentation errors of BM may impact choriocapillaris flow deficit (CCFD) analysis in regions containing hypoRCD.

Methods

Study Compliance Statement and Human Subjects

A retrospective review of a prospective observational swept-source optical coherence tomography angiography (SS-OCTA) imaging study of patients diagnosed with AMD at the Bascom Palmer Eye Institute was performed. This study received approval from the Institutional Review Board of the University of Miami Miller School of Medicine, and written informed consent was obtained from all participants before their inclusion. The tenets of the Declaration of Helsinki were followed, and the study adhered to the regulations of the Health Insurance Portability and Accountability Act of 1996.

Imaging Acquisition and HypoRCD Grading

OCT images were acquired using a SS-OCTA system (PLEX Elite 9000; Carl Zeiss Meditec, Jena, Germany). The SS-OCTA was running at a central wavelength of 1050 nm, a bandwidth of 100 nm, and a scanning rate of 100,000 Hz, providing an axial resolution of ∼5 µm in tissue and a lateral resolution of ∼20 µm. All subjects were imaged using a 6 × 6-mm scan pattern consisting of 500 B-scans, each comprised of 500 A-lines and repeated twice at each position. Scans with a signal strength less than 7, as determined by the instrument output, or those with significant motion artifacts were excluded from analysis. Optical attenuation coefficient (OAC) images were generated using previously described methods to improve the segmentation of physiological and pathological features of interest in this study.¹⁴ HypoRCD were defined as drusen containing foci with internal reflectivity as low as that of the outer nuclear layer on the same B-scans. On en face imaging, hypoRCD appeared as dark spots on a slab generated between the RPE and BM (Figs. 1A1, 1F1). To validate this finding, a grading experiment was conducted. Two graders (OEM, JK) manually outlined hypoRCD on B-scans by reviewing all 500 B-scans within a 6 × 6-mm volume. Each hypoRCD was delineated using a straight line drawn at the base of the druse, parallel to BM. HypoRCD that cast a hypotransmission defect (hypoTD) into the choroid were outlined separately and defined as calcified drusen with hyporeflective cores (hypoRC-CaD).^3,15 These annotations were then compiled to create a binary en face image representing hypoRCD. Independently, another pair of graders (MS, AB) outlined the dark lesions directly on the RPE to BM (RPE–BM) slab while simultaneously reviewing the sub-RPE slab, defined by inner and outer boundaries positioned at 64 µm and 400 µm beneath BM, along with their corresponding B-scans (Figs. 1A2, 1A3, 1F2, 1F3). HypoRC-CaD were outlined as well, represented as a dark spot on the RPE–BM en face image and casting a hypoTD on the sub-RPE slab. A second binary en face image was generated from their annotations. Upon superimposing the two binary images, there was perfect agreement between the two sets of outlines, confirming that the RPE–BM en face slab accurately represents hypoRCD.

Figure 1.

BM bending observed under hypoRCD in OCT images from two eyes. (A1, F1) RPE–BM slab en face images in the first and second eyes, respectively. Yellow arrows indicate the locations of hypoRCD illustrated in B and G. (A2, F2) Sub-RPE slab en face images of the corresponding eyes shown in A1 and F1, respectively. (A3, F3) Corresponding OCT B-scans indicated by the green dashed lines in A1 and F1, respectively. The segmentation boundaries used to generate the RPE–BM slabs are indicated by red and blue dashed lines. The segmentation boundaries used to generate the sub-RPE slabs are indicated by yellow dashed lines. (B, C) Structural OCT and OAC B-scan images, respectively, corresponding to the green dashed line in A. (D, E) Enlarged views of the hypoRCD from B and C, respectively. (G, H), Structural OCT and OAC B-scan images, respectively, corresponding to the green dashed line in F. The yellow arrow in G indicates the choroidal area under the hypoRC druse without shadow tail. (I, J) Enlarged views of a split-core hypoRCD from G and H, respectively. Yellow dashed lines in I and J indicate the BM segmentation. Note the two distinct BM bending locations (marked by blue and green arrowheads), which align with the hyporeflective areas in the drusen core (blue and green arrows).

HypoRCD Tissue Phantom Experiment

A tissue phantom experiment was conducted to investigate the optical mechanism responsible for BM bending beneath hypoRCD. Transparent agarose gel and three kinds of oils with high refractive indices were used to simulate homogeneous hypoRCs. Droplets of transparent agarose gel (with a RI close to the RI of water), olive oil, clove oil, and cinnamon oil (Fig. 2A) were embedded into a titanium dioxide (TiO₂) and 0.3% agarose gel mixture. This mixture provided background scattering surrounding the droplets. The agarose gel and background medium had a RI close to that of water (approximately 1.33), whereas the olive oil, clove oil, and cinnamon oil had RIs of approximately 1.44 to 1.47, 1.51 to 1.54, and 1.59 to 1.62, respectively.^16,17

Figure 2.

Tissue phantom experiment. (A) Olive, clove, and cinnamon oils were used as transparent materials with higher RIs compared to agarose gel. (B) Diagram of cross-sectional view of the tissue phantom, transparent agarose gel, or oil droplets were embedded inside a TiO₂ and 0.3% agarose gel mixture. The gel and droplets were positioned on top of a flat cover glass. (C–F) Representative OCT images of tissue phantoms with embedded cinnamon oil droplet (C), clove oil droplet (D), olive oil droplet (E), and transparent agarose gel (F). Red arrows indicate the bending of the cover glass surface beneath the high-RI droplets, demonstrating bending analogous to that observed in BM under hypoRCD. (G) Boxplot of CoreBI values of cinnamon, clove, and olive oil droplets calculated using OCT images of the tissue phantoms. The blue dashed line indicates a CoreBI of zero for the transparent agarose gel. ***P < 0.001 (n = 7 droplets per tissue phantom type).

To embed the droplets, 5 µL of transparent agarose gel and oil were dispensed onto the surface of cover glasses by slightly touching the micropipette tips to the glass surface. By this method, depending on how the micropipette tips touch the glass surface, multiple droplets with different sizes can be created. A warm TiO₂–agarose gel mixture was then poured over the droplets. The cover glasses were placed on a cold surface to rapidly solidify the gel mixture, preventing the oil droplets from floating onto the gel surface. At this step, the droplets were allowed to be divided into smaller ones. A diagram of the tissue phantom is shown in Figure 2B. The phantom samples were imaged using a customized SS-OCTA system operating at a central wavelength of 1050 nm and a bandwidth of 105 nm.¹⁸ This system exhibited imaging performance comparable to that of clinical SS-OCTA devices, ensuring the relevance and reliability of the phantom imaging data.

Quantitative Analysis of HypoRCD

Three indices were proposed in this study to quantitatively characterize hypoRCD and the associated BM bending: the core bending index (CoreBI), drusen bending index (DruBI), and maximum bending (B_max). These indices were designed to capture the optical properties, relative size, and absolute size of hypoRCs. For each A-line:

• •CoreBI was defined as the ratio of the bending subsidence (b) to the hypoRC height (h).
• •DruBI was defined as the ratio of the bending subsidence (b) to the drusen height (d).

For each hypoRCD, B_max was defined as the maximum bending amount of BM (B_max) observed across all A-lines.

$$\begin{array}{c}\hfill CoreBI=\frac{b}{h}\end{array}$$

$$\begin{array}{c}\hfill DrusenBI=\frac{b}{d}\end{array}$$

$$\begin{array}{c}\hfill MaxBending=B_{max}\end{array}$$

To calculate BM bending, an interpolated BM line was generated by removing the locally bending segment of BM. This interpolated line was treated as the presumed true physical location of BM. A schematic diagram illustrating the measurements is provided in Figure 3A.

Figure 3.

Examples of quantitative analysis of drusen in OCT images using the CoreBI, DruBI, and B_max. (A) Diagram shows a cross-sectional view of hypoRCD and the definition of BM bending (b), B_max, hypoRC height (h), and drusen height (d). The green solid line represents the bent BM as visualized on the OCT image, and the green dashed line indicates the interpolated BM position representing the true BM physical position. (B) A soft druse without underneath BM bending showed 0 for all three indices. (C–E) Three representative hypoRCD with the corresponding indices. Dashed green lines and dashed yellow lines indicate the true physical position of BM and the bent BM, respectively. Yellow shaded areas on the OAC images in C, D, and E indicate segmentation of the hypoRCs.

The hypoRC height was measured following manual segmentation of the hypoRCs. To minimize measurement uncertainty due to segmentation errors, only A-lines with BM subsidence greater than 2 µm and hypoRC height greater than 10 µm were included in the analysis. For the CoreBI and DruBI, the values were averaged across all included A-lines. In the tissue phantom experiments, only the CoreBI was calculated, as DruBI, and B_max measurements were not applicable.

CCFD Analysis

CCFDs were computed using a previously validated workflow.^19–21 Briefly, using the segmentation data output from the SS-OCTA machine, en face projections of the choriocapillaris (CC) microvasculature, superficial vessels, and CC structure were generated from OCT structural and OCTA angiographic volumes using a slab from 4 µm to 20 µm under BM. A compensation algorithm was applied to correct for signal loss, and the retina vessel projection artifacts were removed.

The CCFD map was generated by thresholding and binarizing the en face CC microvasculature image. In this study, the fuzzy C-means (FCM) thresholding algorithm was applied as previously described.^22–24 A binary mask of regions with hypoRCs was produced from an en face mean intensity projection of the RPE-to-BM slab by manually segmenting low-intensity areas, confirmed by inspection of corresponding B-scans (Fig. 4C). Regions with calcified drusen were masked out to avoid introducing artifacts into the CCFD quantification.¹⁹

Figure 4.

An example of CCFD measurement under hypoRCD before and after manual correction of BM segmentation under hypoRCD. (A) RPE–BM slab en face image. (B) Sub-RPE slab en face image. (C) Binary map of hyporeflective core areas. (D, F) Compensated CC flow images generated as a slab extending 4 to 20 µm below BM from OCTA data, shown before (D) and after (F) BM segmentation correction. (E, G) Corresponding binary CCFD maps of D and F. (H, I) Enlarged views of the areas outlined by blue boxes in E and G, showing CCFD maps before (H) and after (I) BM segmentation correction. (J, K) Cross-sectional views of OCT and OCTA, respectively, located at the green dashed lines shown in H and I. (L, M) Enlarged views of the areas outlined by orange boxes in E and G showing CCFD maps before (L) and after (M) BM segmentation correction. (N, O) Cross-sectional views of OCT and OCTA, respectively, located at the purple dashed lines shown in L and M. Yellow dashed lines (before) and red dashed lines (after) in J, K, N, and O indicate the CC slab before BM segmentation correction and after BM segmentation correction, respectively. Yellow arrows in A to G indicate the locations of hypoRCD.

The percentage of CCFDs under hypoRCD was then calculated as the ratio of CCFD pixels within hypoRCD regions to the total number of pixels in those regions. Because standard BM segmentation from OCT devices is typically smoothed, it may not accurately represent BM geometry beneath hypoRCD. To investigate the effect of BM bending on CCFD analyses, BM segmentation lines were manually corrected by two experienced graders (VHL, YZ) in regions under hypoRCD. These manual edits were guided by both structural OCT B-scans and OAC B-scans, which offered improved contrast at the RPE–BM complex.¹⁴ The BM lines were edited so that they reflected the bending under the hypoRCD. Examples of the BM lines before and after correction are shown as the red dashed line and the yellow dashed line, respectively, in Figures 4J and 4N. The CCFD map and CCFD percentages under hypoRCD were then re-computed using the same process described above, but with the corrected BM segmentation. Volumetric SS-OCTA scans from 10 eyes (10 patients) with hypoRCD were analyzed in this study.

Statistical Analysis

Data normality was examined using the Shapiro–Wilk test. Continuous variables were reported as mean ± standard deviation. Differences in CoreBI values among the three tissue phantom groups (cinnamon, clove, and olive) were evaluated using one-way analysis of variance (ANOVA). When a significant overall effect was detected, post hoc pairwise comparisons were performed using Tukey's honestly significant difference (HSD) test to assess differences between all group combinations. The Wilcoxon signed-rank test was used to assess the changes in CCFD percentages under hypoRCD before and after correcting BM segmentations. All statistical analyses were performed in RStudio (R 4.3.2; R Foundation for Statistical Computing, Vienna, Austria), with P < 0.05 considered statistically significant.

Results

BM Bending Under HypoRCD

Figure 1 shows two examples of hypoRCD lesions exhibiting BM bending in two different eyes. In the en face projection images of the RPE–BM slab, the hypoRCD appeared as dark areas (Figs. 1A1, 1F1, yellow arrows). Although the hypoRC druse in Figure 1A1 casts a shadow on the sub-RPE slab en face image (Fig. 1A2, yellow arrow), the hypoRC druse in Figure 1F1 does not cast a shadow on the corresponding sub-RPE slab en face image (Figs. 1F2, 1G, yellow arrow). This indicates that the hypoRC druse in Figure 1A1 represents hypoRC-CaD, whereas the one in Figure 1F1 does not. The bending of BM is visible in the corresponding cross-sectional OCT images (Figs. 1B, 1D, 1G, 1I). The BM bending is more distinctly visualized in the OAC images due to their enhanced contrast (Figs. 1C, 1E, 1H, 1J). Notably, in the case of split-core hypoRCD, the locations of BM bending aligned vertically with the hyporeflective regions within the drusen core (Figs. 1I, 1J, red dashed lines).

Tissue Phantom Experiment

As shown in Figures 2A and 2B, cinnamon oil, clove oil, and olive oil were embedded into a high-scattering medium composed of TiO₂ and 0.3% agarose gel on a flat cover glass surface, mimicking the BM layer observed in OCT. In OCT images, distinct downward bending of the cover glass surface was observed beneath each of the oil droplets (Figs. 2C–E, red arrows), whereas the surface remained flat in the surrounding areas and under the transparent agarose gel (Fig. 2F). Qualitatively, the subsidence beneath the olive oil droplet (Fig. 2E) appeared less pronounced compared to that beneath the cinnamon droplets (Fig. 2C) and the clove oil droplets (Fig. 2D). However, the difference in the degree of bending between the cinnamon and clove oil conditions was not visually distinguishable. Quantitative analysis of seven droplets per phantom type showed that CoreBI values differed significantly among the three oil droplet types (one-way ANOVA, P < 0.001). Post hoc analysis using Tukey's HSD test showed that the CoreBI of olive oil was significantly lower than that of both cinnamon oil (P < 0.001) and clove oil (P < 0.001). A significant difference was also observed between clove oil and cinnamon oil (P = 0.019) (Fig. 2G). These findings support that localized regions with a higher RI can induce optical distortions resembling the BM bending effect observed under hypoRCD in OCT imaging.

Examples of HypoRCD Quantification Using BM Bending

To demonstrate the proposed quantification method, representative hypoRCD and soft drusen were quantified. Figure 3B presents the analysis of soft drusen with no BM bending. Apparently, all three quantitative indices were zero for this soft druse. Analysis results for three hypoRCD of varying sizes are shown in Figures 3C to 3E. Although the DruBI remained relatively stable across the examples (0.214–0.246), B_max showed substantial variability among these representative hypoRCD cases (16–27 µm). Interestingly, the CoreBI value of the large hypoRCD (Fig. 3C) was lower than that for the two smaller hypoRCD (Figs. 3D, 3E), suggesting that the CoreBI may be more sensitive to variations in hypoRC segmentation of small hypoRCD.

Effect of BM Bending on CCFD Analysis

Because BM bending under hypoRCD is an optical imaging artifact, any analysis that depends on accurate BM segmentation should be interpreted with special caution in regions containing hypoRCD. Figure 4 illustrates the impact of correcting BM segmentation on CCFD analysis beneath hypoRCD areas. Before the segmentation correction, the en face CC flow image showed two dark areas corresponding to hypoRCD lesions (Fig. 4D, red arrows), which appeared as a strong flow deficit on the CCFD map (Figs. 4E, 4H, 4L). After BM segmentation correction for the BM bending artifact, these areas became brighter on the CC flow image (Fig. 4F), corresponding to reduced flow deficits (Figs. 4G, 4I, 4M). Notably, one area in the orange box (Figs. 4L, 4M) exhibited a more pronounced reduction in CCFD compared to another area in the blue box (Figs. 4H, 4I), suggesting that the CCFD beneath the hypoRC druse in the orange box was largely artifactual due to incorrect BM segmentation, whereas the hypoRC druse in the blue box may represent a partially true flow deficit.

In support of this, the cross-sectional OCTA images revealed decreased CC flow associated with the apparent BM bending artifact (Figs. 4J, 4K, 4N, 4O). Inconsistent with the qualitative observations, an analysis of CCFD under hypoRCD in 10 eyes showed that the mean CCFD percentages under hypoRCD decreased from 53.2% ± 10.9% (using the OCT device output segmentation) to 25.9% ± 7.7% after BM segmentation correction. The Wilcoxon signed-rank test confirmed that this reduction was statistically significant (P = 0.005; mean change, 27.2% ± 12.1%) (Fig. 5A). A scatterplot showed reductions in CCFD percentages in all 10 examined eyes (Fig. 5B).

Figure 5.

Comparison of the CCFD percentage under hypoRCD before and after BM segmentation correction for the BM bending under hypoRCD. (A) Boxplot of CCFD percentage under hypoRCD areas before and after the segmentation correction. (B) Scatterplot of CCFD percentage under hypoRCD areas before and after the segmentation correction. Number of eyes: 10.

Longitudinal Change in HypoRCD

Figure 6 depicts the longitudinal changes in OCT appearance and quantitative indices of three hypoRCD in an eye over a 3-year follow-up. One hypoRC druse progressed from an early stage at baseline to a larger lesion with a well-defined hypoRC by year 3 (Figs. 6E1–D4, orange dashed circle; Figs. 6A–D). For this lesion, all three indices were zero at the baseline and gradually increased over time (Figs. 6H–J, orange line).

Figure 6.

Longitudinal changes in hypoRCD. (A–D) RPE–BM slab en face images of an eye at baseline and at 1-, 2-, and 3-year follow-up visits, respectively. (E1–E4) OCT images at baseline and at 1-, 2-, and 3-year follow-up visits of the druse shown in the orange dashed circle in A. (F1–F4) OCT images at baseline and at 1-, 2-, and 3-year follow-up visits of the druse shown in the blue dashed circle in A. (G1–G4) OCT images at baseline and at 1-, 2-, and 3-year follow-up visits of the druse shown in black dashed circle in A. (H–J) Graphs of the CoreBI, DruBI, and B_max, respectively, of the three hypoRCD shown in A to D.

The second hypoRC druse showed growth in both druse and hypoRC size, along with increased BM bending (Figs. 6F1–F4, blue dashed circle; Figs. 6A–D). Although the B_max index steadily increased during the follow-up, the DruBI and CoreBI remained relatively stable (Figs. 6H–J, blue dashed line). The higher CoreBI value at baseline may reflect segmentation error due to the poorly defined hypoRC appearance at that time. The third hypoRC druse evolved from a mature hypoRCD into a donut lesion consistent with a large hypertransmission defect (hyperTD),^3,25 as indicated by clear choroidal hypertransmission under BM at the last visit (Figs. 6G1–G4, black dashed circle; Figs. 6A–D). Throughout the 3-year follow-up, all three indices for this hypoRC druse remained stable (Figs. 6H–J, black dotted line).

Discussion

In this study, we report on the bending of BM beneath hypoRCD in OCT images of eyes with intermediate AMD. We hypothesized that this apparent subsidence of BM is an imaging artifact induced by the high RI of the overlying drusen core. This hypothesis was supported by findings from a tissue phantom experiment. The discovery of this phenomenon has implications both for the characterization of hypoRCD and for the accurate interpretation of OCT-based analyses. We demonstrated that BM bending beneath hypoRCD can be quantitatively characterized using three novel indices: CoreBI, DruBI, and B_max. These indices may reflect the internal composition of the hypoRCs and the progression of hypoRCD over time. We also showed that BM bending can affect measurements such as CCFD, emphasizing the need for caution when interpreting OCT-derived metrics in affected regions.

OCT captures depth information by measuring differences in optical path lengths (OPLs) between the sample and reference arms.²⁶ Importantly, OCT directly measures OPLs, not physical depth. Because OPLs are proportional to the RIs of the media that the light passes through, retinal tissues are commonly assumed to have a uniform RI of 1.38 in OCT imaging.^27,28 This value is widely accepted, as it approximates the RI of water, the main constituent of retinal tissue. However, deviations from this assumption can lead to image misinterpretation when the actual RI of a structure differs significantly.

The elevated RI of the hypoRC is likely due to compositional changes that result in an optically denser medium relative to surrounding retinal tissues. A previous study suggested that drusen with heterogeneous internal reflectivity may contain nodules composed of hydroxyapatite,⁷ indicating that hypoRCD could represent drusen undergoing a calcification process. The RI of the hydroxyapatite has been reported to be approximately 1.636.²⁹ Although the exact values of the RIs of hypoRCs may vary, they are evidently higher than the assumed RI of retinal tissue.

Other imaging artifacts related to the high RIs of ocular structures have been reported as lensing effects by a number of groups.^16,30,31 Subretinal lipid globules can induce hypertransmission artifactual tails due to their hyporeflective spherical structure with a high RI.¹⁶ In addition, retinal microcysts were reported altering magnification of the photoreceptor mosaic in adaptive optics scanning light ophthalmoscopy images.³⁰ Similarly, Strampe et al.³¹ reported an interesting case in which the high RI of retained perfluorocarbon liquid on the surface of the retina induced a convex retinal surface appearance in OCT images.

To verify that BM bending beneath hypoRCD is not a true anatomical deformation, we reviewed studies with histological or electron microscopy images of hypoRCD or calcified drusen. Pilgrim et al.³² presented scanning electron microscopy images of large drusen containing multiple spherules enriched with calcium and phosphorus, confirmed by x-ray spectroscopy. Other studies have shown histological images of stage 4 drusen, corresponding to fully calcified drusen,³³ and calcific drusen with basal mounds.^7,34 In all cases, no physical bending or deformation of BM was observed, supporting our conclusion that the bending seen in OCT is an imaging artifact.

In contrast to prior studies that assessed hypoRCD using binary classification based on the presence of features such as hyperreflective caps, split cores, or hypertransmission,^7,8 our proposed indices enable continuous, quantitative measurement of BM bending and the corresponding hypoRCD. This approach allows assessment of the degree of bending that may be correlated with changes in drusen composition, such as increasing calcification. Although soft drusen can progress to calcified drusen, they can develop hypoRCs.¹⁵ In this study, we observed hypoRCD both with and without the appearance of typical calcified drusen with choroidal hypoTDs. Based on the results of our grading experiment, hypoRCD can be broadly classified into two categories: non-calcified (non-CaD) hypoRCD and hypoRC-CaD. Non-CaD hypoRCD appear as discrete dark spots on the RPE–BM en face image, without accompanying hypotransmission defects on the sub-RPE slab. In contrast, hypoRC-CaD are characterized not only by their dark appearance on the RPE–BM slab but also by corresponding hypotransmission defects on the sub-RPE slab.³ Whether there is a prognostic difference between the two types of hypoRCD remains to be determined.

Given that BM bending under the hypoRCs is not a true anatomical feature, caution must be taken when interpreting OCT-derived measurements that depend on accurate BM segmentation, such as CC thickness measurements, CCFD analyses, or drusen height. In this study, we demonstrated that correcting the BM segmentation led to a significant reduction in the mean CCFD percentage under hypoRCD, from 40.3% ± 11.9% to 22.0% ± 7.1%. This corrected value closely matches the previously reported CCFD percentage value of 23.63% ± 6.97% for regions under soft drusen, measured using the same analysis method.¹⁹ By comparison, a prior study reported a much higher CCFD percentage of 59.8% under hypoRCD,¹³ which may reflect the artifactual bending of BM.

Our study highlights the critical need to account for apparent BM bending artifacts when interpreting such metrics. Although previous studies and expert guidelines have recommended manual BM segmentation for accurate CCFD measurements,^13,20 additional scrutiny is warranted in regions with hypoRCD. In addition, the visibility of BM under the hypoRCD may vary, making it difficult to be delineated, especially in cases where the OCT signal is strongly attenuated by an overlaying RPE layer. Because the OAC images are known to offer improved contrast at the RPE–BM complex,¹⁴ we used OAC images together with OCT images to facilitate the segmentation of BM. In challenging cases where segmentation accuracy is uncertain, excluding the affected region from CCFD analyses may be necessary.

This study has several limitations that should be considered when interpreting the results and applying the methodologies in future research. First, the number of OCTA scans with hypoRCD was limited and may not have captured the full spectrum of hypoRCD. Second, all clinical OCT images analyzed were obtained using SS-OCTA systems; no spectral-domain OCT images were included. The visibility of BM bending may vary with system resolution and image display settings, although the underlying optical mechanism should be consistent across platforms. Third, although the tissue phantom experiment produced results consistent with clinical observations, the oils used do not accurately replicate the RI of the hypoRCD in the human retina. A more refined phantom model would be necessary to obtain more accurate estimates of hypoRC RI values. Finally, quantitative analysis required manual segmentation of BM and hypoRCs, which is time consuming. This is particularly important for small hypoRCD, where CoreBI and DruBI values may vary significantly with minor segmentation adjustments. Development of robust, automated segmentation tools is essential for future large-scale studies.

In summary, we have described for the first time, to the best of our knowledge, the appearance of BM bending beneath hypoRCD in OCT images. Using a tissue phantom experiment, we demonstrated that this feature is an imaging artifact caused by the high RIs of the hypoRCs. We proposed three quantitative indices—CoreBI, DruBI, and B_max—for characterizing the hypoRCD using the bending artifact. These indices may serve as potential OCT biomarkers for monitoring AMD progression. Importantly, we emphasized the need for caution when interpreting OCT-based metrics, particularly CCFD measurements, in regions affected by this optical artifact. Further studies are needed to validate these indices and to explore their relationship with disease progression.