Department of Image Processing

Eye diseases are a significant problem in modern society. Many of them can be diagnosed from visual information acquired by specialised medical devices. Glaucoma, one of the leading causes of blindness, can be diagnosed from retinal fundus images by optic nerve head analysis. This process can be automated by segmentation and feature detection algorithms, which reduce the workload of doctors and lead to more people getting the attention they need. We propose an optic disc divergence function suitable for optimisation and a thresholding algorithm utilising the function for the segmentation of optic discs. Further, we investigate the performance of gradient boosting in superpixel classification for optic cup segmentation and compare it with commonly used support vector machines on the same set of features. We evaluate our segmentation pipeline on publicly available datasets, analyse the behaviour of our optic disc divergence function and discuss the advantages and limitations of this approach.