A Neural Network-based Retinal Imaging Interface for Optic Disc Localization in Ophthalmic Analyses

An automatic detection of the position of Optic Disc (OD) is a fundamental step in the analysis of human retina to examine the severity of some diseases or their pathologic progression. A precise localization of Optic Nerve and OD in retinal images reveals unavoidable, but the until now developed solutions providing a unique position seem to collapse when retinal images showing artefacts are analyzed. In previous works, the idea of evaluating multiple pixel determinations of the position of OD on captured fundus images by a multiple Processor has been adopted. In this work, a NN-based Positioning Interface, constituted by a Retinal Imaging System, a Neural Validity Classifier and a Positioning Processor for an accurate localization of the Reference point of OD, is developed. More in detail, the locations of multiple candidates are accurately validated by synthesizing a Neural Network behaving as a Classifier of Validity for regular/abnormal candidate reference points. Then, a Positioning Processor, which considers only validated midpoints, adopts the most suitable point as the Reference point of the OD for subsequent ophthalmic analyses. Simulation results are reported on selected fundus oculi images.