Author: Ursula Schmidt-Erfurth (Austria)
Co-authors: Bianca Gerendas, Martin Michl, Amir Sadeghipour, Hrvoje Bogunovic
To provide proof-of-principle of quantification of intra- (IRF) and subretinal fluid (SRF) volumes in OCT images from real-world management of neovascular age-related macular degeneration (nAMD) using an automated deep learning-based algorithm.
Retrospective analysis from the Vienna Imaging Biomarker Eye Study (VIBES) in patients with nAMD under routine anti-VEGF therapy from 2007-2018 at the Department of Ophthalmology, Medical University of Vienna.
Data from five databases (2 EHR, treatment database, 2 devices (Spectralis, Cirrus) including all patients treated in the out-patient clinic were matched. Filtering was performed for treatment-naive nAMD with a baseline (BSL) OCT image for automated IRF, SRF and CST segmentation and at least one follow-up examination. Visual acuity (VA) and OCT at BSL, month 1-3 and years 1-5, age, gender and treatment frequency were included. Main outcome measures were volumes of IRF and SRF in nanoliters (nl) as measured in the central 1, 3 and 6mm over time.
A total of 1127 eyes were eligible and included into the analyses. Mean CST with 358μm at BSL and decreasing to 280-303μm during follow-up demonstrated a representative cohort. Maximum volumes for IRF and SRF were seen at BSL with an IRF volume of 21.5/76.6/107.1nl, a SRF volume of 13.7/86/262.5nl in the 1, 3, 6-mm area respectively. The course of retinal fluid response in real-world management demonstrated recurrent disease activity following the loading dose. IRF volume decreased to a mean of 5nl during the loading dose in the central mm, increased to 11nl at year 1 and 16nl at year 5. SRF decreased to a mean of 4nl after initial loading in the central mm and remained low with less than 7nl until year 5. In contrast to CST and SRF values, IRF was closely reflecting the visual acuity change over time.
The automated AI-based fluid algorithm precisely detected and quantified IRF and SRF volumes in OCT images from clinical routine over long-term follow-up. Reliable and fast AI tools will allow to introduce efficient decision support indicating disease activity and therapeutic response for the real-world management of exudative macular disease, improving clinical outcomes while saving resources.
Genentech, Novartis, Roche, Retinsight - Consultant Kodiak, Apellis - Research support