Blood Vessel Segmentation in Retinal Images Using Neural Networks

Retinal imaging has been widely used to diagnose vascular and non-vascular pathology in the medical community. Retinal imaging provides information about changes in retinal vascular structure, which are common in diseases such as diabetes, occlusion, glaucoma, hypertension, cardiovascular disease, and stroke. Manual segmentation of blood vessels in retinal images is both prone to error and time consuming, even for experienced physicians. This paper proposes an improved nested U-Net network to achieve the automatic segmentation of blood vessels in retinal fundus images. The performance of the algorithm proposed in this paper has been tested on two publicly available datasets, Retinal Images for Vessel Extraction (DRIVE) and Structured Analysis of the Retina (STARE). The experimental results show that by increasing the complexity, the nested U-net is able to recover fine details such as the very small blood vessels. The model achieved a general accuracy of 0.975, an average area under the Receiver Operating Characteristic (AU-ROC) curve of 0.988 and an average Dice score of 0.813.