Quantitative reference map of retinopathy of prematurity improves diagnostic reliability and treatment decisions.

Abstract

Retinopathy of prematurity (ROP) zone classification guides disease severity assessment and Early Treatment for Retinopathy of Prematurity (ETROP)-based treatment decisions but remains subjective, particularly in the peripheral retina. We developed and validated an anatomically derived quantitative ROP zone reference map to standardise zone assessment in wide-field paediatric retinal images. A reference map was constructed using optic disc-fovea distance and multimodal anatomical data, then applied to RetCam panoramic images generated with a robust deep learning registration pipeline. Four ophthalmologists evaluated 85 ROP examinations with and without the reference map in a crossover reader study. Reader-reference and interobserver agreement for zone classification and zone-dependent ETROP decisions were assessed using kappa statistics. The reference map improved reader-reference agreement for zone classification from 0.58 to 0.92 and interobserver agreement from 0.47 to 0.86. In a Zone III versus non-Zone III analysis, reader-reference agreement improved from 0.40 to 0.88. In 20 zone-dependent ETROP observations, agreement improved from 0.85 to complete concordance within this subset. The quantitative ROP zone reference map improved reliability of zone classification and supported more consistent ETROP guideline application in this retrospective reader study.