Cuantificación de la sensibilidad de segmentación en OCTA: perfiles específicos del dispositivo en tres plataformas...

Optical coherence tomography angiography (OCTA) enables quantitative assessment of the retinal microvasculature structure, but measurements depend on accurate layer segmentation. We quantified how sensitive commonly used OCTA metrics are to small, systematic shifts of the boundary separating the superficial and deep capillary plexus slabs, and assessed whether the resulting changes are clinically meaningful. In this prospective cross-sectional study, 32 eyes from 16 healthy participants underwent 3 × 3 mm OCTA imaging on three commercial devices (Intalight DREAM, Zeiss Cirrus, Topcon Triton). Using each device's native segmentation tools, the shared inner plexiform layer/inner nuclear layer boundary was shifted generating five offset conditions (-10, -5, 0, + 5, and +10 μm); 0 μm represents the unshifted standardized definition. En face images were analyzed with the open-source OCTAVA pipeline to derive vessel area density, total vessel length, vessel length density, branching measures, and foveal avascular zone (FAZ) area in both plexuses. Linear mixed-effects models estimated the effect of offset on each metric, expressed as standardized slopes and relative change per 10 μm. Sensitivity to segmentation offset was strongly device- and layer-dependent. DREAM showed pronounced offset effects in both plexuses (e.g., superficial vessel area density +4.9% per 10 μm and deep vessel area density -5.2% per 10 μm), with marked deep FAZ enlargement (+15.5% per 10 μm). Cirrus exhibited moderate deep-plexus sensitivity (approximately -2% per 10 μm for vessel density/length metrics) with high FAZ sensitivity (+16.1% per 10 μm). Triton metrics were comparatively stable, except for deep FAZ area (+9.8% per 10 μm). Offsets in the single-digit micrometer range produced changes comparable in magnitude to meta-analytic differences reported for early diabetic microvascular impairment. Taken together, the marked device- and layer-specific offset sensitivities, often reaching effect sizes comparable to reported early diabetic microvascular differences, underscore that micrometer-scale slab definition changes can materially bias OCTA-derived endpoints. Standardized slab definitions, segmentation quality control, and transparent reporting of adjustments are therefore essential for reliable interpretation in longitudinal studies, multicenter research, and algorithm development. These quantitative sensitivity profiles may further inform practical tolerance thresholds for quality control and cross-platform harmonization in clinical studies and multicenter trials.