Automatická analýza, modelování a časová evaluace retinálních lézí

Abstract

The main theme of this diploma thesis is the analysis and modeling of retinal lesions from clinical images from the RetCam3 system. In the thesis, a segmentation algorithm was designed based on time-iterative deformable curves, which are represented by a model of active contours. Active contours modify their geometric properties during the segmentation process, thereby gradually approximating the area of manifestation of retinal lesions. The result is a binary mathematical model reflecting the geometric parameters of retinal lesions. An important procedure of the proposed model is the differentiation of individual lesions. This classification procedure makes it possible to separate individual retinal lesions in order to identify and monitor a particular lesion. The model allows the extraction of geometric parameters of individual lesions in order to objectivize the manifestation of the lesion against the optical disk. Modeling of retinal lesions was tested on a large dataset containing 1015 records.