Smartphone-Based Automated Quantification of Bulbar Conjunctival Hyperemia Using Vessel Density Mapping

Purpose: To develop a smartphone application (Conjunctiva VD) that objectively quantifies bulbar conjunctival hyperemia by estimating the conjunctival vessel density (VD) within a user-defined region of interest (ROI). Methods: A total of 100 conjunctival images from 2 public datasets were analyzed. An external imaging technician selected images meeting predefined quality criteria and delineated an inferior bulbar conjunctival ROI within the app. After ROI selection, the application automatically resized each ROI to 2100 × 1800 pixels and performed on-device vessel segmentation using a fixed computer-vision pipeline to compute the VD (%). Three expert ophthalmologists independently graded hyperemia on the technician-defined ROI using the scale (0–4) and provided a subjective percentage estimate of the VD. The inter-rater agreement (the quadratic-weighted κ for Efron grading and the intraclass correlation coefficient for the subjective VD) and associations between the VDs and the expert assessments (Spearman’s ρ) were evaluated. Results: The inter-rater agreement was high for the Efron grading (quadratic-weighted κ = 0.854–0.885), with a similarly high reliability for the subjective VD estimation (intraclass correlation coefficient = 0.910; 95% confidence interval, 0.877–0.935). The application-derived VD correlated strongly with Efron grades across graders (Spearman’s ρ = 0.780–0.807; all P < 0.001) and with the expert-estimated VD (ρ = 0.814–0.843; all P < 0.001). Conclusions: A lightweight smartphone application can provide an objective VD metric that shows a strong association with an expert clinical assessment of conjunctival hyperemia. Translational Relevance: A portable, widely available smartphone-based tool for objective hyperemia quantification may lower barriers to point-of-care use and enable scalable applications in outpatient triage, longitudinal monitoring, and teleophthalmology.