Automatic Detection of Optic Disc and Fovea from Colour Fundus Images

Abstract

Diabetic Retinopathy (DR) is a chronic eye disease. More specifically, it is the manifestation of diabetes mellitus in the retina. DR is diagnosed by ophthalmoscopy or fundus imaging. Photographs of the retina are obtained using a fundus camera consisting of a low magnification microscope to which a digital camera is mounted. Three important anatomical areas are of great importance for the automated detection of DR: blood vessels, optic nerve head (ONH) region, and the central macular area termed fovea (fov). This work presents a methodology to detect and segment the ONH based on multispectral analysis and its amount of entropy, and a method to localise the fov centre based on level contour maps. As a pilot study, this methodology is tested using two different databases: APEC and MESSIDOR, with a total of 93 fundus images. Based on these we conclude that ONH and fov centres were successfully detected automatically, with low errors in average of 0.17% and 0.30% for ONH and fov respectively. An average p value of 0.5189 from a paired t-student test shows that the manual vs the automatic distance between ONH and fov centers is not statistically significantly different. These preliminary results show that the used methodology is promising. A more thorough evaluation has to be made using larger databases.