Detekce a lokalizace retinálních patologií

Abstract

The aim of this work is to design, develop and implement a system for detection and localization of pathological findings in the retina of the eye using modern image processing and machine learning methods. To achieve effective treatment, it is important to diagnose the disease and its condition correctly and early. For this purpose, a system based on the UNet architecture is proposed. The learned model can then predict binary images containing potential segmented pathologies with some confidence. The proposed algorithm then determines their size and distance to physiological points of the retina. The feasibility of the chosen solution is verified by implementing an experiment with annotated data. This work presents new possibilities in the field of retinal disease diagnosis using trained models and a web application. If further developed, this approach has the potential to be used by ophthalmologists in the future to diagnose and determine the extent of pathological changes in the retina of the eye.