Refined Four-Layer Peripapillary OCTA Enables Layer-Specific Localization of Microvascular Alterations in Retinal Vein Occlusion

Study Population

This retrospective cross-sectional study was approved by the Institutional Review Board of Zhongshan Ophthalmic Center (2025KYPJ153) and adhered to the tenets of the Declaration of Helsinki. The requirement for additional informed consent was waived by the Institutional Review Board because of the retrospective use of anonymized clinical data. Patients with unilateral RVO (including BRVO and CRVO) and age-matched healthy controls (HC) were retrospectively recruited from January 2020 to June 2025. Eyes were categorized into three groups: RVO-affected (RVO-A), RVO-fellow (RVO-F), and HC. Exclusion criteria were (1) concurrent retinal or optic nerve diseases (e.g., glaucoma, diabetic retinopathy); (2) high myopia (axial length > 26.0 mm or spherical equivalent < −6.0 D); (3) prior intraocular surgery or intravitreal injections; (4) significant media opacity preventing high-quality imaging; and (5) hemorrhage obscuring the peripapillary measurement area; (6) edema-related segmentation distortion that precluded reliable quantitative analysis; and (7) systemic diabetes mellitus, to minimize the potential confounding effect of diabetic microangiopathy.

All RVO-affected eyes were newly diagnosed and treatment-naïve at the time of OCTA imaging. Imaging was performed before any intravitreal anti-VEGF injection, laser treatment, or intraocular surgery. Disease duration was defined as the interval from symptom onset to OCTA imaging and was recorded in days.

Clinical Examination and OCTA Imaging

All participants underwent comprehensive examinations, including best-corrected visual acuity (BCVA, converted to logMAR), non-contact tonometry, and axial length measurement using the IOL Master 700 (Carl Zeiss Meditec, Jena, Germany). OCTA imaging was performed using a swept-source system (VG200; SVision Imaging, Luoyang, China; central wavelength: 1050 nm; A-scan rate: 200,000 scans/s; axial resolution: 3.8 µm). A 6 × 6-mm volumetric scan (1024 × 1024 A-scans) centered on the optic disc was acquired.

Image Quality Control and Segmentation Review

Scans were included only when the signal strength score was ≥7. All OCTA images were independently reviewed by two retinal specialists (ZJL and ZXJ) together with corresponding structural B-scans. Review criteria included en face OCTA image quality, motion or blink artifacts, hemorrhage masking of the peripapillary measurement region, four-layer segmentation boundaries, and DCP slab reliability. For major quality control decisions, such as image exclusion due to severe segmentation failure, both graders reached the same decision (resulting in the exclusion of four eyes specifically because of severe edema-related segmentation distortion; Fig. 1). For minor boundary uncertainties in three eyes (two with CRVO and one with BRVO), the two graders jointly re-evaluated the structural B-scans, resolved ambiguity through consensus discussion, and performed minor manual adjustments to ensure accurate quantitative analysis. Furthermore, 21 eyes were excluded because of dense peripapillary hemorrhage causing significant signal obstruction. Clinical subtype classifications (distinguishing BRVO vs. CRVO) were concordant between the two graders.

The proprietary three-dimensional projection artifact removal algorithm provided by the OCTA device manufacturer was applied automatically during image reconstruction to reduce projection artifacts in deeper vascular slabs. Because the exact implementation is proprietary, the algorithm itself was not independently validated in this study.

Eyes with retinal edema were retained only when segmentation boundaries were judged acceptable on both en face OCTA images and corresponding B-scans. Images were excluded when retinal layers could not be reliably identified or when edema-related boundary distortion precluded reliable quantitative analysis.

Hierarchical Anatomical Stratification

To evaluate the impact of anatomical granularity on peripapillary OCTA assessment, three hierarchical dimensions were defined according to the device-specific automated segmentation protocol. Dimension 1 represented the global inner retinal model. The inner retinal slab was defined from the internal limiting membrane (ILM) −5 µm to the inner nuclear layer/outer plexiform layer boundary (INL/OPL) +25 µm.

Dimension 2 represented the conventional two-layer model, including the superficial capillary plexus (SCP) and the DVC. The SCP slab was defined from ILM −5 µm to the one-third boundary of the ganglion cell complex (GCC 1/3). The DVC slab was defined from GCC 1/3 to INL/OPL +25 µm.

Dimension 3 represented the refined four-layer model, including the RPC, SVP, ICP, and DCP. The RPC slab was defined from ILM −5 µm to the nerve fiber layer/ganglion cell layer boundary (NFL/GCL). The SVP slab was defined from the NFL/GCL boundary to GCC 1/3. The ICP slab was defined from GCC 1/3 to the midpoint of the inner nuclear layer (INL 1/2). The DCP slab was defined from INL 1/2 to INL/OPL +25 µm.

All automated segmentation boundaries were reviewed on corresponding structural B-scans. Manual correction was performed when necessary. Scans with hemorrhage obscuring the peripapillary measurement region, severe motion or blink artifacts, or edema-related distortion causing unreliable slab boundary identification were excluded from quantitative analysis.

Microvascular and Structural Quantification

Vessel density (VD, percentage of area occupied by vasculature) and perfusion area (PA) were quantified. VLD and VDI were derived via a skeletonization algorithm (VD ≈ VLD × VDI), decoupling vessel abundance from caliber.²⁰ These metrics were calculated within a 4 × 4-mm region of interest centered on the optic disc to minimize edge artifacts and ensure signal stability.

Mean retinal thickness (MRT) and optic nerve head morphometry were measured over the entire 6 × 6-mm scan area. FD,^18,19 representing global geometric complexity, was calculated using the box-counting method over the 6 × 6-mm area to capture the complete peripapillary network architecture.

Sample Size and Power Contextualization

Because this was a retrospective cross-sectional study based on eligible treatment-naïve RVO cases with high-quality OCTA images, no formal a priori sample-size calculation was performed. Effect estimates together with their 95% CIs were emphasized to characterize the magnitude, precision, and uncertainty of the observed associations.

Statistical Analysis

Statistical analyses were performed using R (v4.4.1) and SPSS (v26.0). Normality was assessed using the Shapiro–Wilk test. Continuous variables are presented as mean ± SD or median [interquartile range], as appropriate.

Demographics and Baseline Characteristics

The study flow, image quality control process, and analytical framework are shown in Figure 1. A total of 186 eyes from 114 participants were included: 72 RVO-affected eyes (RVO-A; 26 CRVO and 46 BRVO), 72 fellow eyes of patients with RVO (RVO-F), and 42 HC. All RVO-A eyes were newly diagnosed and treatment-naïve at the time of OCTA imaging.

Baseline characteristics are summarized in the Table. Age and sex distribution did not differ significantly among the RVO-A, RVO-F, and HC groups. Hypertension and hyperlipidemia were more prevalent in participants with RVO than in healthy controls. The median disease duration from RVO onset to OCTA imaging was 12.5 days (interquartile range [IQR], 7.75–20.0). RVO-A eyes showed worse visual acuity than RVO-F and HC eyes, with a mean logMAR BCVA of 0.67 ± 0.49. Axial length, signal strength index, and intraocular pressure were comparable across groups. RVO-A eyes showed increased mean retinal thickness compared with RVO-F and HC eyes, whereas no significant difference in mean retinal thickness was observed between RVO-F and HC eyes.

Anatomical Divergence and Signal Dilution

Representative en face OCTA images and vessel density heatmaps demonstrated depth-resolved peripapillary microvascular alterations in RVO-A eyes (Fig. 2). BRVO eyes showed more localized vascular involvement, whereas CRVO eyes showed more diffuse peripapillary abnormalities. Four-layer visualization allowed separate assessment of the RPC, SVP, ICP, and DCP and helped localize deep-layer OCTA abnormalities that were less apparent in global composite slabs. Representative en face OCTA images and corresponding structural B-scans illustrating acceptable segmentation are shown in Supplementary Figure S1.

Across the three anatomical dimensions, OCTA metrics showed greater layer-specific differences after refined stratification (Table; Supplementary Table S1). In Dimension 1, VD-IR was lower in RVO-A eyes than in HC eyes, whereas FD-IR did not differ significantly. In Dimension 2, VD-DVC was significantly reduced in RVO-A eyes. In Dimension 3, VD reductions were most evident in deeper peripapillary layers, including the SVP and DCP, whereas the RPC layer was relatively less affected in VD-based analyses. To further characterize composite-slab attenuation, VD-related measurements were decomposed into vessel length density and vessel diameter index. Global VLD did not differ significantly among groups, whereas DVC-VLD and DCP-VLD were significantly reduced in RVO-A eyes compared with HC eyes. In contrast, DCP-VDI remained stable across groups, suggesting that deep-layer OCTA abnormalities were more closely related to density- or length-related changes than to caliber alteration.

Standardized disease contrast analysis further illustrated the effect-size variations across anatomical slabs (Fig. 3; Supplementary Table S2). For fractal dimension, the global inner retinal slab showed a small standardized contrast (d = −0.07; 95% CI, −0.31 to 0.46), whereas the conventional DVC and refined DCP showed larger contrasts of identical magnitude (d = −0.39; 95% CI, −0.78 to 0.00 for both). For vessel density, standardized contrasts were −0.57 (95% CI, −0.96 to −0.18) for the global IR, −0.51 (95% CI, −0.90 to −0.12) for the SCP, −0.83 (95% CI, −1.23 to −0.43) for the DVC, and −0.79 (95% CI, −1.19 to −0.39) for the DCP. These findings indicate that four-layer stratification did not increase the effect-size estimate beyond that achieved by the conventional deep slab, but rather provided improved anatomical localization.

OCTA-Based Discriminative Performance

OCTA-based discriminative performance improved with refined anatomical stratification (Fig. 4; Supplementary Fig. S2). For VD-based models, the apparent AUC increased from 0.636 (95% CI, 0.536–0.737) in Dimension 1 to 0.729 (95% CI, 0.637–0.822) in Dimension 2 and 0.826 (95% CI, 0.748–0.905) in Dimension 3. Compared with Dimension 2, Dimension 3 showed an AUC increase of 0.097 (95% CI, 0.007–0.186; paired DeLong P = 0.035). For FD-based models, the apparent AUC increased from 0.603 (95% CI, 0.495–0.711) in Dimension 1 to 0.688 (95% CI, 0.588–0.788) in Dimension 2 and 0.807 (95% CI, 0.726–0.888) in Dimension 3. The AUC difference between Dimensions 3 and 2 was 0.119 (95% CI, 0.022–0.216; paired DeLong P = 0.016). Bootstrap internal validation was performed for representative four-layer models and additional covariates (age, axial length, and image quality score). For the VD-based model, the estimated optimism was 0.038, resulting in an optimism-corrected AUC of 0.788. For the FD-based model, the optimism was 0.035, with an optimism-corrected AUC of 0.772, suggesting limited optimism within the current dataset.

Model fit also improved, as reflected by a decrease in AIC from 143.4 to 125.7. Reclassification analyses using continuous NRI and IDI further supported the incremental value of the four-layer model over the conventional two-layer model.

Subtype-stratified analyses showed that FD-based discriminative performance improved from Dimension 1 to Dimension 3 in both CRVO and BRVO subgroups (Supplementary Table S3; Supplementary Fig. S2). In CRVO eyes, the AUC increased from 0.838 to 0.985, whereas in BRVO eyes, the AUC increased from 0.641 to 0.811. The CRVO Dimension 3 estimate should be interpreted cautiously because of the limited subgroup sample size and potential model separation. In adjusted subtype-specific analyses, both DCP-VD and DCP-FD were lower in CRVO and BRVO eyes than in HC eyes, with a larger DCP-VD reduction observed in CRVO.

Associations Between OCTA Metrics and Visual Function

Refined anatomical stratification improved the explanatory performance of FD-based models for visual function (Fig. 5). In the global inner retinal model, FD showed limited explanatory power for logMAR BCVA (adjusted R² = 0.021), whereas DCP-FD in Dimension 3 showed higher explanatory performance (adjusted R² = 0.211). The improvement from Dimension 2 to Dimension 3 was significant for FD-based models (P = 0.038) but not for VD-based models (P = 0.882).

Scatter plots supported the association between deep-layer vascular network complexity and visual function. Because DVC-FD was also associated with BCVA, the added value of the four-layer model should be interpreted as improved anatomical localization and enhanced functional interpretability rather than as complete unmasking of an otherwise absent association. Relative importance analysis showed that DCP-related metrics contributed the largest proportion of explained variance in VD- and FD-based visual function models, accounting for 44.5% to 65.5% of the model-explained variance. Secondary metrics showed broadly consistent patterns for PA and VLD, whereas VDI appeared to be more strongly influenced by superficial-layer parameters (Supplementary Fig. S3).

Sensitivity and Robustness Analyses

MRT-stratified sensitivity analyses were performed to evaluate whether DCP-related findings were mainly driven by retinal edema (Supplementary Table S4). Using an exploratory post-hoc threshold of 400 µm for MRT, both DCP-VD and DCP-FD remained significantly reduced in the MRT < 400 µm stratum. In the MRT ≥ 400 µm stratum, DCP-VD showed a larger reduction, whereas DCP-FD showed a similar effect estimate. After excluding eyes with MRT ≥ 400 µm, models incorporating DCP-VD or DCP-FD retained OCTA-based discriminative ability, with AUCs of 0.707 and 0.773, respectively. These results suggest that edema severity may amplify density-related DCP abnormalities, but DCP-related OCTA changes were not solely attributable to the most severely thickened eyes.

Disease-duration sensitivity analyses are shown in Supplementary Table S5. Disease duration was not significantly correlated with DCP-VD, DCP-FD, or logMAR BCVA. In multivariable models additionally adjusted for disease duration, DCP-FD remained significantly associated with BCVA, whereas DCP-VD and disease duration did not. These findings suggest that the association between DCP-FD and visual function was not primarily explained by disease duration in this newly diagnosed, treatment-naïve cohort.

Layer-Specific Peripapillary Alterations Beyond the RPC-Centric View

Traditionally, peripapillary OCTA studies have often focused on the radial peripapillary capillary layer because of its anatomical proximity to the optic nerve head and its relevance to optic nerve perfusion.^{5,12,13,21–23} However, the peripapillary vascular network is anatomically heterogeneous and includes multiple plexuses with potentially different susceptibilities to venous congestion and perfusion disturbance. In the present study, the RPC layer was relatively less affected in VD-based analyses, whereas deeper layers, particularly the DCP, showed more evident OCTA abnormalities.

These findings do not imply that the RPC layer is unimportant in RVO. Rather, they suggest that an RPC-centered or broad composite-slab approach may be insufficient to capture the full spectrum of peripapillary microvascular alterations. The value of four-layer stratification lies in its ability to localize abnormalities to specific plexuses rather than averaging them across anatomically heterogeneous slabs. This interpretation is particularly important because the DCP did not show a larger standardized disease contrast than the conventional DVC. Furthermore, DCP-related OCTA abnormalities may reflect a combination of microvascular impairment, signal attenuation, and segmentation-related effects in edematous eyes and should therefore be interpreted cautiously.^20,24,25

Anatomical Localization and Metric-Dependent Standardized Contrasts

The standardized disease-contrast findings require a metric-specific interpretation. For VD, moderate-to-large negative contrasts were observed in both the conventional DVC and the isolated DCP, with confidence intervals excluding zero. The standardized contrast was slightly larger in the DVC than in the DCP, with estimates of −0.83 and −0.79, respectively. For FD, the DVC and DCP showed identical point estimates of −0.39, and the corresponding 95% confidence intervals reached the null value. These findings do not support a monotonic increase in standardized effect-size magnitude with progressively finer anatomical segmentation. In particular, they should not be interpreted as demonstrating a stronger group-level effect in the isolated DCP than in the conventional DVC. Instead, the findings indicate that the value of four-layer stratification is primarily anatomical. The conventional DVC combines signals from the intermediate and deep portions of the retinal microvasculature, whereas the four-layer framework permits these components to be evaluated separately. Thus, even when the DVC and DCP produce similar standardized group contrasts, the DCP-resolved analysis may still provide more specific information regarding the anatomical layer contributing to the observed abnormality.

Geometric decomposition of VD into VLD and VDI provided additional context for this finding.²⁰ Global VLD did not differ significantly among groups, whereas DVC-VLD and DCP-VLD were reduced in RVO-affected eyes. In contrast, DCP-VDI remained relatively stable. These results suggest that layer-specific OCTA differences were more closely related to density- or length-related changes than to vessel caliber alteration. Importantly, this interpretation remains OCTA-based and does not establish histologic capillary loss. Rather, it supports the methodological value of separating plexus-level metrics when evaluating peripapillary vascular alterations in RVO.

Measurement variability in deep retinal layers remains an important consideration. Previous studies have shown that deep-layer OCTA metrics may be influenced by projection artifacts, reduced signal quality, and segmentation uncertainty.^26,27 In our study, the proprietary projection artifact removal algorithm was used as part of image preprocessing, but it was not independently validated. Therefore DCP-related findings should be interpreted with caution. The persistence of DCP-VD and DCP-FD differences in MRT-stratified analyses and after excluding eyes with MRT ≥ 400 µm suggests that the findings were not solely driven by eyes with marked retinal thickening (MRT ≥ 400 µm), but residual attenuation and segmentation effects cannot be completely excluded.

Fractal Dimension as a Complementary Marker of Network Complexity

FD provides information regarding vascular network complexity and spatial organization beyond area-based metrics such as VD.¹⁸ In the present study, FD-based models showed improved functional explanatory performance after refined anatomical stratification. Specifically, DCP-FD showed higher explanatory performance for BCVA than global FD, and the improvement from the two-layer model to the four-layer model was significant for FD-based models but not for VD-based models.^18,28

These findings suggest that FD may provide complementary information regarding visual function in RVO. However, FD should not be interpreted as universally superior to VD. VD and FD likely capture different aspects of OCTA-derived vascular alteration: VD reflects area- or density-related perfusion information, whereas FD reflects vascular network complexity and organization. Therefore the main implication is that FD may enhance functional interpretation when used alongside density-based parameters, particularly in plexus-resolved analyses.

Subtype-Specific Findings in BRVO and CRVO

Subtype-stratified analyses showed that FD-based discriminative performance improved from Dimension 1 to Dimension 3 in both CRVO and BRVO. CRVO showed higher AUC values than BRVO, and the CRVO Dimension 3 estimate should be interpreted cautiously because of the smaller subgroup sample size and potential model separation. In adjusted subtype-specific analyses, DCP-VD and DCP-FD were lower in both CRVO and BRVO eyes than in healthy control eyes, with a larger DCP-VD reduction observed in CRVO.

These findings indicate that the pooled RVO results were influenced by subtype heterogeneity. Nevertheless, the improvement observed in BRVO suggests that refined anatomical stratification may also be useful when vascular alterations are more localized. Larger subtype-specific cohorts are needed to validate these findings and to determine whether the incremental value of four-layer analysis differs between BRVO and CRVO.

Clinical Implications and Translational Value

From a translational perspective, the present findings suggest that peripapillary OCTA analysis may benefit from plexus-resolved anatomical stratification rather than relying solely on global, superficial, or broad deep slabs. The primary contribution of the four-layer framework was more specific anatomical localization of abnormalities within the conventional deep vascular complex. Although DCP isolation did not further increase the standardized disease contrast relative to the DVC, refined stratification provided metric-dependent incremental information, particularly for FD-based visual function models. These results support the potential value of DCP-resolved and layer-specific OCTA measurements as research tools for characterizing the anatomical and functional heterogeneity of RVO-related microvascular alterations.

However, the present cross-sectional design does not establish that these metrics can monitor disease progression or predict treatment response. Therefore the clinical role of four-layer peripapillary OCTA should be considered hypothesis-generating. Future longitudinal studies are needed to determine whether plexus-resolved OCTA metrics can provide prognostic information for visual outcomes, macular edema evolution, or post-treatment vascular remodeling.

Limitations

This study has several limitations. First, the cross-sectional design precludes definitive causal conclusions regarding the chronological sequence of layer-specific damage. Second, because previously treated eyes were excluded, the findings should be generalized primarily to newly diagnosed and treatment-naïve RVO eyes and may not fully reflect post-treatment vascular remodeling in routine clinical practice. Third, although automated projection artifact removal, strict image quality control, and B-scan-based segmentation review were applied, residual projection artifacts, hemorrhage-related shadowing, signal attenuation, and edema-related segmentation errors cannot be completely excluded. The proprietary projection artifact removal algorithm was not independently validated in this study; therefore DCP-related OCTA findings should be interpreted cautiously, particularly in eyes with substantial edema. Fourth, although subtype analyses were performed, the CRVO subgroup was relatively small and may be susceptible to model separation. Finally, because the discriminative models were developed and evaluated in the same dataset, the apparent AUC, AIC, NRI, and IDI estimates may be optimistic. This concern is particularly relevant to models containing a larger number of predictors and to subgroup analyses with limited sample sizes, including the CRVO analysis in which the Dimension 3 AUC reached 0.985. Internal validation can quantify optimism within the current cohort but does not replace external validation. Therefore the incremental discriminative performance of the four-layer models should be confirmed in independent multicenter cohorts before clinical application.