Quantifying the clinical evaluation of retinal fluid change in real-world OCT images using deep learning

Abstract

Purpose : To quantify and compare the clinical evaluation of intra- and subretinal fluid (IRF, SRF) change between eye care professionals with different degrees of experience in retinal diseases in spectral-domain optical coherence tomography (SD-OCT) images of neovascular age-related macular degeneration (nAMD) patients.

Methods : We included SD-OCT data of patients who were diagnosed with nAMD and treated at our retina department and are part of the Vienna Imaging Biomarker Eye Study (VIBES) registry. There were 28 to 120 days between two visits and anti-VEGF injections were either given between the two visits or on the day of the second visit. A fully automated algorithm based on deep learning was used to quantify the change in IRF/SRF between visits. Firstly, the segmentation performance and clinical utility of each scan was determined by manual inspection. Secondly, retina specialists, ophthalmology residents, ophthalmologists working in private practice and orthoptists were asked to grade the IRF/SRF change between visits, following a standardized questionnaire.

Results : SD-OCT volumes of 230 visit pairs were included in our analysis. Manual inspection of all scans revealed a detectable fluid change of approximately 5nl. We therefore excluded scans with a fluid volume of <5nl at the first visit and fluid change of <5nl to the follow-up visit.

Conclusions : Artificial intelligence allows a precise assessment of fluid change over time, and thus an optimized management of macular edema. Our study determines the variability in the clinical assessment of retinal fluid change and the significance of retina- and OCT related experience for an accurate clinical evaluation of macular edema.

This is a 2021 ARVO Annual Meeting abstract.