Purchase this article with an account.
Timothy P.H. Lin, Herbert Y.H. Hui, Poemen P. Chan, Yumeng Wang, Ruyue Shen, Tien Y Wong, Clement C. Tham, Carol Y. Cheung; Assessment of risk of normal tension glaucoma progression via deep-learning based retinal-vessel caliber measurement. Invest. Ophthalmol. Vis. Sci. 2021;62(8):1027.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Vascular insufficiency has been suggested to be of pathogenic role in normal tension glaucoma (NTG). This study evaluated the associations between baseline retinal-vessel caliber measured by an automated deep-learning system to the risk of NTG progression.
In this longitudinal cohort study, 390 eyes from 253 NTG patients with a follow-up period of ≥24 months were included. Central retinal arteriolar equivalent (CRAE) and central retinal venular equivalent (CRVE) were measured from retinal photographs by a validated deep-learning system (SIVA-DLS). We assessed the agreement of retinal vessel caliber measurements between SIVA-DLS and humans, and used the Cox proportional-hazards model to examine the relationships between baseline retinal vessel caliber and subsequent NTG progression. We further assessed the incremental value of adding the retinal vessel caliber for prediction of NTG progression beyond previously reported risk factors. NTG progression was defined as progressive retinal nerve fiber layer (RNFL) thinning over time on Cirrus HD-OCT detected by its built-in guided progression analysis.
Agreement of retinal vessel caliber measurement between SIVA-DLS and human was good to excellent, with intraclass correlation coefficients of retinal vessel caliber between SIVA-DLS and human ranging from 0.87 to 0.90. Sixty-nine eyes (17.7%) developed NTG progression over the follow-up period. Narrower CRAE (hazard ratio [HR] 1.36 [95%CI 1.05, 1.76]) and CRVE (HR 1.38 [95%CI 1.03, 1.84]) at baseline were independently associated with NTG progression in the multivariable model. Addition of baseline CRAE and CRVE improved the discrimination (C-statistic 0.703 vs 0.695, p=0.038 and 0.704 vs 0.695, p=0.038, respectively) of NTG progression risk beyond previously reported risk factors.
This study demonstrated significant associations between baseline retinal vessel caliber and risk of NTG progression. These findings provided evidence to support the hypothesis of vascular insufficiency in NTG pathogenesis and the utility of a high-throughput deep-learning system for risk assessment in NTG.
This is a 2021 ARVO Annual Meeting abstract.
CRAE and CRVE measurement using SIVA-DLS. The original retinal photograph captured by a 45 degree digital retinal camera (i). SIVA-DLS automatically generate heat maps based on the features of the retinal vessels for prediction and calculation of CRAE (ii) and CRVE (iii).
This PDF is available to Subscribers Only