Decision support system for early detection of Glaucoma in retinal images

Abstract

Glaucoma is the second leading cause of blindness worldwide. With lack of symptoms, Patients affected with the disease remain completely unaware and by the time they realize it, it is already too late as the damage done is irreversible. Glaucoma damages the Optic Nerve and this damage is permanent. Early detection of Glaucoma is very helpful because by detecting the disease at an early state, it can be treated and its progression can be stopped. Glaucoma is detected from retinal fundus images by looking for abnormalities in the Optic Nerve Head. These abnormalities occur in the form of increased Optic Cup to Optic Disc ratio and Neuroretinal Rim loss. The project aimed towards development of an automated Glaucoma diagnosis system using image processing and machine learning techniques. The end goal was to drive state of the art approach for accurate Glaucoma detection which can facilitate the general population for Glaucoma screening. Glaucoma when detected at an early stage can help the patient in avoiding blindness. The work done was divided in to multiple modules so that the project can be made extensible and these modules can aid in the development of automated diagnosis of other diseases like other as Diabetic Retinopathy, and Macular Edema. Optic Disc was extracted from the retinal fundus image by first applying preprocessing to the image to negate the effect of uneven illuminations and the presence of pathologies. Then a novel Polar transform based Optic Disc segmentation technique is applied that is able to segment the Optic Disc with a very high accuracy and in very little time. An in-depth comparison of proposed methodology with other techniques in the literature is also made and from the results it is clear that proposed methodology beats other techniques in segmentation accuracy and does it in a very low time. For the segmentation of Optic cup, specific preprocessing was applied in order to highlight the Optic Cup region from the Optic Disc region. Optic Cup was segmented using a Supervised Pixel classification approach. In the last step, Glaucoma detection was performed by applying machine learning classifiers on the CDR and ISNT features extracted from the Optic Cup and Optic Disc masks. The methods achieved much better accuracy in segmenting Optic Cup and Optic Cup and thus in the detection of Glaucoma.