Prediction of OCT images of short-term response to anti-VEGF treatment for neovascular age-related macular degeneration using generative adversarial network

Abstract

Purpose : To generate post-therapeutic optical coherence tomography (OCT) images after anti-vascular endothelial growth factor (VEGF) therapy for typical neovascular age-related macular degeneration (nAMD) based on pre-therapeutic OCT images using generative adversarial network (GAN), and evaluate the predictive accuracy of synthetic OCT images.

Methods : A total of 476 pairs of pre- and 1-month post-therapeutic OCT images of patients with nAMD were included in training set, while 50 pre-therapeutic OCT images were included in test set, and their corresponding post-therapeutic OCT images were used to evaluate the synthetic images. The pix2pixHD method was adopted for image synthesis. No lesions or retinal and choroidal structures were labeled. The quality, authenticity of the synthetic images were evaluated by retinal specialists. The post-therapeutic macular status of wet or dry and macular wet-to-dry conversion after treatment in synthetic OCT images, which were essential in the management of nAMD, were evaluated.

Results : We found that 92% of the synthetic OCT images had sufficient quality for further clinical interpretation. Eight percent of the synthetic OCT images were excluded from the following evaluations because of insufficient quality, such as presence of two overlapped layers of neuroretina or chorioretinal coloboma. Among the 92% synthetic OCT images with sufficient quality, only about 26% to 30% synthetic post-therapeutic images could be identified as synthetic rather than real images by ophthalmologists, and the other synthetic images were difficult to be identified synthetic or real. The accuracy to predict post-therapeutic macular status of wet or dry was 0.85 (95% CI 0.74-0.95). And the accuracy to predict macular wet-to-dry conversion was 0.81 (95% CI 0.69-0.93).

Conclusions : Our results revealed a great potential of GAN to generate post-therapeutic OCT images with both good quality and authenticity. The present method to predict short-term response after anti-VEGF treatment shows relatively acceptable accuracy on prediction of post-therapeutic macular status and macular wet-to-dry conversion, which suggested the possibility that GAN might be used to assist decision-making in clinical practice. Further studies could focus on training GAN with labeled images and validating the performance using an external dataset.

This is a 2020 ARVO Annual Meeting abstract.