Design a implementace vybraných evolučních strategií pro optimalizaci regionálních segmentačních modelů s cílem identifikace objektů z medicínských obrazů

Abstract

The topic of this diploma thesis is testing the effectiveness of segmentation algorithms in the segmentation of medical image data, the acquisition of which was performed using MRI, fundus camera and ultrasound. In the second part of the work dealing with the segmentation of selected objects of interest, CT, MRI and ultrasound images were used. Noise in the image is an undesirable additive component that changes the brightness intensity of the pixels, and thus errors can occur when classifying pixels into individual segmentation regions. A pair of Fuzzy-ABC and F-FCM algorithms, which are based on the principle of fuzzy logic, were tested in this work. These algorithms overcome the problem of pixel misclassification caused by local statistical aggregation. Another pair of algorithms are the K-means and Otsu thresholding methods. These two algorithms are so-called conventional algorithms, and their segmentation efficiency was compared with the efficiency of both fuzzy algorithms. The theoretical part of the work is briefly devoted to the basic principles of image data segmentation and selected evolutionary strategies for image segmentation. A review of such evolutionary strategies used for image segmentation was also made. The main goal of the work was to analyze the effectiveness and robustness of segmentation methods in the context of variable deterministic noise (gaussian, salt&pepper, speckle) with dynamic intensity and subsequent comparative analysis and modeling of the effectiveness of segmentation of tested methods depending on the parameters of segmentation strategies. Objective evaluation methods were used to evaluate the results (corelation, MSE and SSIM).