Retinal Vasculature Segmentation Using Local Saliency Maps and Generative Adversarial Networks For Image Super Resolution

TL;DR

This paper introduces a GAN-based super resolution method for retinal images that uses local saliency maps to improve image quality and segmentation accuracy, especially for small or blurred features.

Contribution

It presents a novel saliency loss in GANs leveraging local saliency maps, enhancing super resolution quality and segmentation performance in retinal imaging.

Findings

SR images have perceptual quality close to original images

Segmentation accuracy with SR images approaches that with original images

Method outperforms competing super resolution techniques

Abstract

We propose an image super resolution(ISR) method using generative adversarial networks (GANs) that takes a low resolution input fundus image and generates a high resolution super resolved (SR) image upto scaling factor of $16$. This facilitates more accurate automated image analysis, especially for small or blurred landmarks and pathologies. Local saliency maps, which define each pixel's importance, are used to define a novel saliency loss in the GAN cost function. Experimental results show the resulting SR images have perceptual quality very close to the original images and perform better than competing methods that do not weigh pixels according to their importance. When used for retinal vasculature segmentation, our SR images result in accuracy levels close to those obtained when using the original images.