Abstract
Purpose :
To evaluate the relationship between zone, stage, and extent of stage 3 retinopathy of prematurity (ROP) and vascular severity using a deep learning (DL) based vascular severity score.
Methods :
We previously reported the development and evaluation of a deep learning-based ROP vascular severity scale that ranges from very mild (1) to very severe (9) and demonstrated that the vascular severity score correlates with the International Classification of ROP (ICROP) disease severity. In this study, we further explored the Imaging and Informatics in ROP (i-ROP) cohort which documented the “extent” of stage 3 disease in ROP as determined by the examining ophthalmologist using indirect ophthalmoscopy. We analyzed 6354 posterior pole Retcam (Natus medical incorporated, Pleasanton, CA) images of babies who underwent ROP screening without a history of treatment. We compared the vascular severity as a function of ROP stage (0,1,2,3) and as a function of number of quadrants with stage 3 disease (0-3 clock hours, between 3-6 clock hours, or > 6 clock hours), in zone I and zone II. All comparisons were done using analysis of variance in Stata v15.
Results :
The Table summarizes the demographics and ICROP classifications for the dataset. The figure demonstrates the ROP vascular severity scale as a function of ROP stage, extent of stage 3, and zone I and zone II disease. The vascular severity score increased with increasing stage of disease in zone I (top left, P<0.001) and zone II (top right, P<0.001), and with increasing extent of stage 3 in zone I (bottom left, P=0.03) and zone II (bottom right, P<0.001).
Conclusions :
A higher ROP vascular severity score was associated with decreasing zone, increasing stage, and increasing extent of stage 3 disease. These results suggest the potential utility of an objective metric for ROP disease severity that complements the existing ICROP classification schema.
This is a 2020 ARVO Annual Meeting abstract.