Deep Learning Based Vessels and Optical Disc Segmentation in Retinopathy Images

Abstract

artificial intelligence-based algorithms. Eye diseases such as glaucoma is based on segmentation of optical disc and blood vessels in retinopathy images. Diabetic retinopathy is a radical eye disease and it causes of blindness at adverse level. Optical disc and blood vessels commence nurturing at early stage of diabetic retinopathy recognized as proliferative diabetic retinopathy. The correct segmentation of blood vessels and optical disc help the medical specialist and ophthalmologists in primary recognition of vision related diseases like glaucoma, hypoxemia, diabetic retinopathy etc. The segmentation of retinal images is much dependent on image quality and illuminations. The image acquisition stage can create non-uniform illumination of the fundus images which can make the retinal vessel pixel closer to the background. The conventional schemes are much dependent image processing techniques to enhance the image prior to the segmentation process which require much time and processing cost. Deep learning is famous to help the computer vision task with accuracy and reliability. Therefore, in this study, we propose a new deep learning-based method for the segmentation of optic disc and blood vessels using convolutional neural network. The intensive segmentation task is carried out by semantic segmentation which enable the network to perform the reliable segmentation without the overhead of pre-processing. The experiments include both retinal vessel and optical disc segmentation using publicly available datasets. The vessel segmentation experiment is performed with famous DRIVE dataset, whereas the optical disc segmentation experiment is performed with MESSIDOR dataset. The experimental results show the fine segmentation performance of proposed method for both vessel and optical disc segmentation in order to support the diagnosis in retinal diseases.