Segmentace papily zrakového nervu u dětí s ROP pomocí hlubokého učení

Abstract

This bachelor’s thesis focuses on the automatic segmentation of the optic disc in fundus images. This task is important for automated analysis of these images and may indirectly contribute to the diagnosis of ocular diseases. The aim of this thesis is to design and implement a system for accurate localization and segmentation of the optic disc using image processing and deep learning methods. The approach is based on a two-stage algorithm combining a localization and a segmentation convolutional neural network. First, an approximate position of the optic disc is determined using a probability map. Then, a region of interest is extracted from this area, and a U-Net-based segmentation network is applied. The system was tested on a dataset of neonatal retinal images. The method achieved a mean Dice coefficient of 84,96 \% (median 87,83 \%), with a precision of 87,54 \% and a recall of 83,46 \%. The results show that the method can reliably localize and accurately segment the optic disc. The approach may contribute to the automation of retinal image analysis and support clinical practice.