EyeLoveGAN: Exploiting domain-shifts to boost network learning with cycleGANs

TL;DR

This paper introduces EyeLoveGAN, leveraging cycleGANs to generate artificial domain-shifted retinal images, thereby improving the training of neural networks for segmentation, classification, and localization tasks in ophthalmology.

Contribution

The novel use of cycleGANs to create domain-shifted images enhances training data diversity for retinal image analysis tasks.

Findings

Improved segmentation accuracy on retinal images.

Enhanced glaucoma classification performance.

Effective localization of the fovea using heatmaps.

Abstract

This paper presents our contribution to the REFUGE challenge 2020. The challenge consisted of three tasks based on a dataset of retinal images: Segmentation of optic disc and cup, classification of glaucoma, and localization of fovea. We propose employing convolutional neural networks for all three tasks. Segmentation is performed using a U-Net, classification is performed by a pre-trained InceptionV3 network, and fovea detection is performed by employing stacked hour-glass for heatmap prediction. The challenge dataset contains images from three different data sources. To enhance performance, cycleGANs were utilized to create a domain-shift between the data sources. These cycleGANs move images across domains, thus creating artificial images which can be used for training.