Reproducibility And Diagnostic Performance Of Swept-Source As-Octa In Corneal Neovascularisation
Published 2026 - 44th Congress of the ESCRS
Reference: POCOR482 | Type: Free paper | DOI: 10.82333/5wmf-qe98
Authors: Dr Valentin Jünger* 1 , Andrea Margarella 2 , Dr Giacom Visioli 3 , Dr Fabio Scarinci 3 , A/Prof. Carlo Nucci 4 , Dr Rossella Anna Maria Colabelli Gisoldi 3 , A/Prof. Augusto Pocobelli 3
1Department of Experimental Medicine and Ophthalmology,University of Rome Tor Vergata,Rome,Italy;Department of Ophthalmology,Rome Eye Bank, San Giovanni Addolorata Hospital Rome Eye Bank, San Giovanni Addolorata Hospital,Rome,Italy, 2Department of Clinical Medicine, Public health, Life and Environmental Sciences,University of L’Aquila,L’Aquila,Italy;Department of Ophthalmology,Rome Eye Bank, San Giovanni Addolorata Hospital Rome Eye Bank, San Giovanni Addolorata Hospital,Rome,Italy, 3Department of Ophthalmology,Rome Eye Bank, San Giovanni Addolorata Hospital Rome Eye Bank, San Giovanni Addolorata Hospital,Rome,Italy, 4Department of Experimental Medicine and Ophthalmology,University of Rome Tor Vergata,Rome,Italy
Purpose
Objective quantification of corneal neovascularisation (CoNV) remains challenging in both clinical and research settings. Swept-source anterior segment optical coherence tomography angiography (AS-OCTA) enables non-invasive vascular imaging, yet the reproducibility and diagnostic performance of quantitative CoNV metrics are not fully established. This study evaluated inter-rater agreement, test–retest repeatability, and discriminatory performance of AS-OCTA-derived percentage of corneal area covered by CoNV (%CoNV) using different segmentation strategies.
Setting
In this cross-sectional pilot study, 18 (6 controls; 12 CoNV eyes) eyes from 10 participants underwent 400-kHz full-range swept-source AS-OCTA imaging. Across 32 visits, 52 scan acquisitions (including same-visit repeat scans) generated 255 quantitative %CoNV measurements.
Methods
Three region-of-interest (ROI) strategies were analysed: fixed 6.5-mm circular ROI (RR), limbus-to-limbus ROI (LL), and manual corneolimbal margin contouring (mCM). Two independent graders performed measurements. Inter-rater and test–retest reliability were assessed using two-way random-effects intraclass correlation coefficients (ICC). Absolute agreement was further quantified using coefficient of variation of differences (CVdiff%) and within-subject coefficient of variation (wCV%). Linear mixed-effects modelling assessed the influence of method, rater, rescan, and disease severity on %CoNV. Diagnostic discriminatory performance between control and CoNV eyes was evaluated using receiver operating characteristic (ROC) analysis.
Results
Inter-rater reliability was excellent across all ROI strategies (LL ICC=0.958; mCM ICC=0.949; RR ICC=0.955). Inter-rater variability (CVdiff%) ranged from 10.8% (RR) to 32.1% (mCM). No systematic grader effect was detected (p=0.73). Test–retest repeatability was strong (LL ICC=0.942; mCM ICC=0.962; RR ICC=0.856). Within-subject variability (wCV%) ranged from 14.5% (RR) to 18.8% (mCM). A small rescan effect on %CoNV of −1.0 percent point was observed (p=0.009). Mixed-effects modelling demonstrated segmentation dependency (p<0.001) and a strong association with CoNV severity (p<0.001), confirming biological validity. ROC analysis demonstrated good discrimination between control and CoNV eyes (mCM AUC=0.846; LL AUC=0.840; RR AUC=0.731).
Conclusions
Swept-source AS-OCTA enables reproducible quantification of %CoNV. All ROI strategies demonstrated excellent inter-rater agreement and good test–retest repeatability. Automated ROIs minimize variability, whereas anatomy-guided contouring maximize diagnostic performance. These findings support %CoNV derived from AS-OCTA as a reliable quantitative biomarker for corneal vascularization assessment and highlight the importance of standardized ROI methodology when balancing reproducibility and diagnostic performance.