Abstract
Purpose :
To describe the development and validation of a deep learning algorithm (DLA) for the detection of neovascular age-related macular degeneration (AMD).
Methods :
A deep learning system for the automated detection of neovascular AMD was developed and validated on a set of 142,275 non-stereoscopic, retinal images (n=4,188 with neovascular AMD). Retinal photographs included in the development and internal validation datasets (n=56,113) were derived from a variety of clinical settings in China. Gold standard grading of these images was assigned when consensus was reached by 3 individual ophthalmologists. For external validation, we tested the efficiency and diagnostic accuracy of the DLA on 86,162 non-mydriatic images (76% Anglo-Celtic origin) derived from Melbourne Collaborative Cohort Study (MCCS). Images from the MCCS were labelled by professional graders. Referable AMD was defined as neovascular AMD and/or ungradable outcome in one or both eyes. Area under the receiver operating characteristic curve (AUC), sensitivity and specificity.
Results :
In the internal validation dataset, the AUC, sensitivity and specificity of the DLA for neovascular AMD was 0.995, 96.7%, 96.4%, respectively. Testing against the independent external dataset achieved an AUC, sensitivity and specificity of 0.967, 100% and 93.4%, respectively. More than 60% of false positive cases in the internal and external validation datasets displayed other macular pathologies. Among the false negative cases (internal validation dataset only), over half (57.2%) proved to be undetected detachment of the neurosensory retina or RPE layer.
Conclusions :
This DLA shows robust performance for the detection of neovascular AMD amongst retinal images from a multi-ethnic sample and under different imaging protocols. Future efforts will focus on investigating the effectiveness of our deep learning algorithm for neovascular AMD as a screening tool in high risk populations, and what impact the introduction OCT imaging has as a second-line screening tool amongst screen-positive patients.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.