Artery Venous Classification in Retinal Images

Abstract

Vessel segmentation and artery vein classification is often the first step in automated retinal image analysis. The numerous visual features and anomalies in retinal images can indicate the onset of diseases and their phase of progression at different times of examination. The experts and ophthalmologists must spend hours on manual detection of blood vessels apprehending the extensive data gathered by different techniques. All this leads to heavy workload for an ophthalmologist with manual diagnosis using extensive number of patients’ data. These challenges can be minimized by using digital imaging techniques in capturing eye fundus images and automating screening capabilities using image analysis algorithms. The more number of patients can be monitored and screened by using automated retinal image analysis and if needed they can be referred to expert/ ophthalmologist for detailed examinations. In this work, an automated vessel segmentation and classification system using deep learning is presented that is able to achieve state of the art results. Semantic segmentation, a category of deep learning is employed for segmentation and classification of blood vessels. Publicly available datasets are used for training and evaluation. The results prove that the proposed methodology performs really well and is able to segment and classify the retinal vessels accurately even in the presence of pathologies and uneven illuminations in the images. This was a major problem previously as uneven illuminations and retinas affected with disease pose a major problem for automated methods. Custom preprocessing needed to be developed specifically to solve the problem of uneven illumination. Pathologies present in the image also caused problems. These issues are handled in proposed methodology using deep learning based semantic segmentation architectures. All preprocessing required is being handled by the layers of the neural network model in order to produce accurate results.