Superpixel Segmentation Methods on Stereo Fundus Images and Disparity Map for Glaucoma Detection

Glaucoma is an optic neuropathy resulting in progressive vision loss. It is the leading cause of global irreversible blindness. The reported prevalence among the population in New Zealand is 2% over the age of 40 years. About 10 % of those over 70 years are diagnosed with this disease. Population-based studies report high rates of undiagnosed glaucoma with over 50 % of the population with glaucoma living in developed countries remaining undiagnosed and unaware of their disease. Clinical diagnosis rests on the detection of the characteristic optic disc signs. Stereo optic-disc imaging improves intra and inter-observer agreement in the detection of optic disc abnormalities. We propose a robust method to help the specialist in detecting some abnormalities in stereo optic-disc images using stereo vision and superpixel segmentation concepts. A stereo vision system produces a disparity map for the input stereo images of the retina in which abnormalities are more distinguishable. In the region of interest in the disparity map, we can clearly visually recognize all abnormalities because of changes in distances for abnormality parts in glaucoma patients. The produced disparity map is segmented using two different superpixel segmentation algorithms (simple linear-iterative clustering and simple noniterative clustering) to detect abnormalities. The original stereo images are also segmented using the same concept; results are compared with the segmented disparity map.

I. Introduction

Glaucoma is an optic neuropathy resulting in progressive vision loss [1]. It is the leading cause of global irreversible blindness [2]. In 2013, the population aged 40–80 years with glaucoma worldwide was estimated to be 64.3 million; this is projected to increase to 76.0 million in 2020 and 111.8 million in 2040 [3].