Automated localisation of the optic disc and fovea to assist diabetic retinopathy screenings

The localisation of the anatomical structures in retinal fundus images is an important step in the automated analysis of retinal images. In this paper automated methods for the localisation of the optic disc and the fovea are proposed. For an optic disc localisation, the area of most vasculature loops is determined to locate an initial optic disc centre, followed by morphological operations and a circular Hough Transform to determine its boundary and the final centre. For fovea localization, foveal features and a model of geometric relations between the fovea and both the optic disc and the vasculature are used to locate its boundary and centre. Both methods were evaluated using two sets of images from different datasets, and their competitive performance indicate that they could be used for a computer-aided mass localisation of the optic disc and the fovea as part of an automatic screening programme.