Quantitative Assessment of the Extent of Retinal Vascularization in Retinopathy of Prematurity

Objective: To evaluate pixel-based measurements of the minimum nasal and temporal extent of retinal vascularization (NERV and TERV) from RetCam images in retinopathy of prematurity (ROP). Design: Retrospective, multicenter diagnostic accuracy study. Subjects: Infants screened for ROP at 8 high-volume centers in Imaging and Informatics in ROP (i-ROP) consortium between July 2011 and December 2016. Methods: Widefield RetCam3 temporal and nasal fundus images showing the optic disc and vascular-avascular junction were selected. A reference standard diagnosis for zone, stage, and plus disease was assigned based on indirect ophthalmoscopy and expert panel review. Each eye was also assigned an artificial intelligence-derived vascular severity score (VSS) (scale 1-9). Manual stage segmentation was performed for each image, combined with automated optic disc localization, and the Euclidean distance from the disc center to the stage was calculated in pixels to derive the NERV and TERV. Intergrader reproducibility was assessed using intraclass correlation coefficient. Main Outcome Measures: Minimum NERV and TERV (pixels), comparison of NERV and TERV between zone I and zone II, nasal-temporal asymmetry, area under the receiver operating characteristic curve for zone I diagnosis, and relationship between VSS, stage, and NERV/TERV. Results: A total of 1832 images from 605 eyes of 350 infants met the inclusion criteria. Minimum NERV was shorter than minimum TERV in 98.3% of eyes, being a mean of 20.9% ± 7.9% lower. Both minimum NERV and TERV were significantly lower in zone I (276.5 ± 43.2; 345.0 ± 53.4 pixels) than zone II (360.8 ± 44.5; 452.6 ± 52.5 pixels; P < 0.0001). Area under the receiver operating characteristic curve for zone I was 0.91/0.90 for minimum NERV/TERV, respectively. The VSS was higher in eyes with a higher stage for a given extent of retinal vascularization (ERV) and higher within a given stage for eyes with a lower ERV. The intergrader reproducibility of ERV was excellent (intraclass correlation coefficient = 0.985, 95% confidence interval: 0.977 to 0.992). Conclusions: Objective measurement of the ERV in eyes with ROP demonstrates that the NERV is nearly always less than the TERV. These findings have implications for ROP diagnosis and classification. Financial Disclosure(s): Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.