Domain Agnostic Pipeline for Retina Vessel Segmentation

TL;DR

This paper presents a simple yet effective pre-processing pipeline for retina vessel segmentation that achieves near state-of-the-art results across various datasets and image qualities without complex models.

Contribution

The authors introduce a novel pre-processing pipeline that enables robust retina vessel segmentation without relying on complex neural networks or training routines.

Findings

Achieves high segmentation performance across multiple datasets

Maintains accuracy on poor quality and pathological images

Demonstrates effectiveness with a simple pipeline

Abstract

Automatic segmentation of retina vessels plays a pivotal role in clinical diagnosis of prevalent eye diseases, such as, Diabetic Retinopathy or Age-related Macular Degeneration. Due to the complex construction of blood vessels, with drastically varying thicknesses, accurate vessel segmentation can be quite a challenging task. In this work we show that it is possible to achieve near state-of-the-art performance, by crafting a careful thought pre-processing pipeline, without having to resort to complex networks and/or training routines. We also show that our model is able to maintain the same high segmentation performance across different datasets, very poor quality fundus images, as well as images of severe pathological cases. Code and models featured in this paper can be downloaded from http://github.com/farrell236/retina_segmentation. We also demonstrate the potential of our model at http://lazarus.ddns.net:8502.