Automatic Detection of Retinal Optic Disc using Vessel Inpainting

Aziah Ali, Mubdiul Hossain, Noramiza Hashim, Wan Noorshahida Mohd Isa, Zarina Che Embi, Shahbe Mat Desa

Abstract: One of the crucial landmarks within a fundus image is the optic disc. Automatic detection of the optic disc from a fundus image is a desired feature for a computer-aided retinal diagnosis system. Based on the location information of the detected optic disc, retinal information such as the disc size, the optic cup parameters, and retinal vessel width parameters can be efficiently estimated with reliable accuracy, thus contributing towards better retinal diagnosis. In this paper, a method to efficiently perform automatic detection of the optic disc from a retinal image input is proposed. The first main step after image pre-processing is simple thresholding to detect most retinal vessels on the fundus image. The resulting binary image is then used for the next inpainting step in which the segmented vessel pixels are replaced with pixels having values closer to the values of neighbouring pixels on the retinal background. This results in an image with most retinal vessels inpainted, hence increasing the general visibility of the optic disc. Applying Hough transform to the inpainted image then results in localising the optic disc centre as well as estimating the radius, making the optic disc parameters available for subsequent retinal analysis. The proposed method is applied to the MESSIDOR fundus image dataset and achieved a detection rate of 99.5% and segmentation accuracy of 99.61% with an average processing time of 0.81 seconds per image.