Metric-Specific Ocular Magnification Artifacts in Quantitative OCT Angiography: A Controlled Image-Based Simulation Study

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Background and Objectives: Optical coherence tomography angiography (OCTA) devices convert scan angle into retinal distance with a schematic eye, so eyes whose axial length (AL) differs from the assumed value are measured on an incorrect transverse scale. We quantified how this magnification error affects seven OCTA metrics and how well geometric and empirically calibrated corrections perform. Materials and Methods: In this controlled image-based simulation study, segmentation masks of 300 eyes from the public OCTA-500 dataset (6 × 6 mm scans) were re-rendered at ten simulated ALs (21.5 to 30.0 mm). Co-primary outcomes were seven metric-specific scaling exponents from per-eye log–log regression, tested against their geometric values (Holm-adjusted and equivalence tests). Secondary analyses separated field-content from pixel-sampling effects and assessed corrections in individual eyes by cross-fitting. Results: Between simulated ALs of 23.58 and 27.91 mm, uncorrected foveal avascular zone (FAZ) area changed by −30.8% and branch-point density by +37.4%. The FAZ-area exponent was equivalent to its geometric value (−1.985 vs −2), whereas skeleton-derived metrics fell short of theirs (branch-point density +1.72, vessel length density +0.91, caliber −0.74). A decomposition attributed the shortfall of the skeleton-derived exponents mainly to pixel sampling and the vessel density exponent to field content. Empirical exponents reduced the mean bias to 1.0% or less, but individual-eye errors exceeded 5% in 12% to 36% of eyes for vessel density, vessel length density, FAZ circularity and branch-point density. Exploratory analyses showed curved responses for all metrics, negligibly so for FAZ area, and category-dependent responses for vessel density, vessel length density and branch-point density. Conclusions: In this simulation, magnification artifacts were metric-specific: FAZ area followed geometric scaling, whereas skeleton-derived metrics also depended on pixel sampling and field content. Calibrated corrections apply to group means within the tested workflow and require validation in eyes with measured axial length.

This publication has 26 references indexed in Scilit: