Detection of Laser Marks from Retinal Images for Improved Diagnosis of Diabetic Retinopathy

Abstract

Eye diseases such as diabetic retinopathy may cause blindness. It affects the central vision of the person and in some cases causes severe blindness. Diabetic retinopathy is a progressive disease so at the advance stages of diabetic Retinopathy further disease progression is stopped using laser treatment. Laser treatment leaves behind marks on the retinal surface that causes misbehaviors in automated retinal diagnostic system. These laser marks hinders the further analysis of the retinal images so it is desirable to detect laser marks and remove them to avoid any unnecessary processing. In this paper we have proposed a method that uses techniques from image processing and machine learning to help segment out the laser marks from the retinal images. Our method uses techniques to remove uneven illumination from the images using various morphological operations. The system uses Minimum distance classifier using a feature set extracted from the laser marks in retinal images. The evaluation of the proposed system is done on a locally gathered dataset of patients suffering from different Retinal diseases. The result of our system are based on various parameters like accuracy, specificity and sensitivity. A fair comparison with any other technique is not possible due to limited literature on laser marks detection from Fundus images. The results of our method has shown that the laser marks from retinal images are detected with good accuracy though there are some failure cases but the result is still acceptable.