Retinal Vessel Segmentation from Fundus Images Using DeepLabv3+

Blood vessel segmentation from retinal images is crucial for identifying a range of eye diseases, including diabetic retinopathy and glaucoma. Therefore, research on automatic retinal blood vessels segmentation has sparked much attention. Numerous image processing techniques have been developed for segmenting retinal vessels from fundus images. In this paper, we propose a method that is based on deep learning. A semantic segmentation convolutional neural network (CNN) based on DeepLabv3+ was implemented. To allow for blood vessel segmentation, the network was modified to accept single-channel images and perform two-class pixelbased classification (vessel and non-vessel). Following segmentation, the output images are refined using morphological closing operation. The suggested technique was validated using images from the DRIVE dataset. The results show that it can achieve accuracy, sensitivity, specificity, precision, Jaccard, and Dice values of 0.9263, 0.8006, 0.9385, 0.5579, 0.4874, and 0.6551, respectively. We demonstrated that the proposed method could produce better results than those produced by other proposed methods.