An Automatic Active Contour Approach to Segment Retinal Blood Vessels

Abstract

Assessment of the blood vessel morphology in retinal fundus images is an indispensable method to diagnose diseases such as diabetic retinopathy and glaucoma. Ophthalmologists commonly evaluate fundus images with manual planimetry vessel extraction, which represents a clear bottleneck and is prone to human errors. Therefore, an automatic vessel segmentation tool can help clinicians perform this task, thus, improving the accuracy of the diagnosis. This work proposes a fully automatic segmentation framework based on the Chan-Vese active contouring algorithm for defining blood vessels in retinal images enhanced by matched filtering. Moreover, custom pre-processing workflows facilitate the subsequent segmentation depending on the analyzed images' intensity-based characteristics. The effectiveness of the proposed method was evaluated on the benchmark dataset STARE. Our framework outputs resemble much closer the ground truth images than other segmentation strategies, achieving an average accuracy of 94.37% and a Dice Similarity Coefficient of 0.7441.