BetaEntity Annotation Prototype
← Back to diseases

Annotated abstract

#ECR-Paper-03 VESPAR: quantitative analysis of the retinal vasculature using a novel segmentation algorithm to enable gene-phenotype association studies in a cohort of 3,944 patients with inherited retinal diseases

bmjophth · 2025-10-10 · canonical JSON source

16 visible annotations · policy: published · automated confidence ≥ 75.00%

Document resource

Objectives To develop a vessel quantification pipeline (VESPAR) for fundus autofluorescence (FAF) and infrared (IR) imaging in inherited retinal diseases (IRDs) to enable gene-phenotype association studies and explore vascular changes in these conditions.Methods Retrospective IRD patient data (55-degree blue-FAF and 30-degree IR from Heidelberg Spectralis, 2004–2019) were gathered from Moorfields Eye Hospital (MEH), the Royal Liverpool Hospital (RLH) and Oxford Eye Hospital (OEH). Manual segmentation of retinal blood vessels was undertaken in 127 FAF images from 69 patients representing 28 distinct genes by two expert graders, as well as 16 IR images from 16 patients representing 11 distinct genes by a third grader, from the RLH. A U-Mamba model trained on 72 FAF scans and tested on 23 FAF and 16 IR scans, segmented 42,268 FAF scans (3,616 patients, 179 genes) and 77,090 IR scans (4,307 patients, 198 genes) from MEH. Quantitative metrics were then derived using reti-py.Results Model-grader Dice was 0.68 on FAF (intergrader Dice 0.66) and 0.59 on IR. VESPAR’s fractal dimension strongly correlated with manual values, while fractal dimension and vessel density showed the best repeatability across intra- and inter-visit cohorts.Retinitis pigmentosa (RP) and Leber congenital amaurosis (LCA) exhibited the lowest fractal dimensions at baseline (1.30 - 1.40), whereas optic atrophy (OPA1) and choroideremia had the highest (1.49 and 1.46). Dominant RP genes showed higher vessel density (11.8%) and fractal dimension (1.421) than X-linked (8.2% and 1.351) and recessive genes (9.1% and1.369); while tortuosity density (0.737) was significantly lower than in recessive RP (0.748). In PRPH2, the RP phenotype had lower vessel density (12.4%) than the macular dystrophy phenotype (15.6%). In ABCA4, the most severe phenotype and genotype groups had significantly lower vessel density than the least severe groups. Moreover, vessel metrics correlated with VA and age and, revealed distinct progression rates across different genes and phenotypes.Conclusions VESPAR paves the way for robust gene-phenotype associations by facilitating cross-sectional and longitudinal analyses of vascular changes in IRDs.