Návrh a implementace segmentačního modelu s prvky umělé inteligence pro modelování retinálních lézí

Abstract

Retinopathy of premature primarily affects premature infants with low birth weight, manifested by pathological development of the retinal vascular system and bleeding spots, called retinal lesions. Approximately 8,000 premature babies are born in the Czech Republic annually, of which about 1,650 are born with weight below 1,000 g. The main purpose of the RoP screening is the early detection of the first signs of this disease. The practical part of the thesis is therefore devoted to an automatic algorithm with artificial intelligence elements for detection and classification of retinal lesions. Semi-supervised classification is used to enable online learning and automatic detection of retinal lesions without user input. For the inicialization of semi-supervised learning the image segmentation is used based on adaptive binarization and blob detection using the SURF feature extraction. It means, we talk about automatic segmentation without any human intervention, which can save a lot of time in the future. Individual differentiated retinal lesions are analyzed against the optical disc.