Analýza aktivně deformovatelných křivek pro detekci zájmových oblastí z medicínských obrazů

Abstract

This bachelor thesis is focused on digital image processing, especially on one part of this process – region - based image segmentation. The theoretical part describes general parameters of pictures data and the process in which they are processed. The research part analyses image segmentation based on active contours – basic information, properties, and specific methods and models. The practical part deals with implementation of chosen segmentation methods on synthetic images from ultrasound phantom and on real medical images from CT (21 images), MRI (18 images), and US (18 images). These medical images are focused on pathological structures in human body (brain tumours – CT and MRI, benign breast tumours – US). The aim is to compare the quality and efficiency of two selected segmentation methods based on active deformable curves on image data. The evaluation is performed by comparing the results of segmentation methods with the ground truth. Various evaluation parameters are used (MSE, SSIM, correlation, accuracy etc.), and the results are shown in tables, that compares both methods on individual data sets – CT, MRI, and US.