Identifikace a modelování tkání z medicínských obrazů na základě metod shlukové analýzy s prvky umělé inteligence

Abstract

This thesis deals with methods of cluster analysis with elements of artificial intelligence designed for regional segmentation of medical images. With the help of segmentation, we are able to break down and classify a certain area of interest of the image, which is crucial for us. Today's radiodiagnostic methods achieve high-quality image outputs, but a frequent phenomenon in obtaining an image is its influence on parasitic noise. Artificial intelligence methods using evolutionary and genetic algorithms are used in many fields to solve very complex optimization problems. By converting these algorithms into the context of image segmentation, we are able to achieve a better division of the image into individual segments and thus compensate for the shortcomings of conventional methods. This work includes a comparative analysis of individual methods in the context of variable image conditions. Specifically, these are the algorithms KM, FCM, GA, PSO, which were subjected to a thorough analysis in the test and simulation environment of the MATLAB software. In the next part of the work, the extraction and modeling of tissues from medical images also affected by parasitic noise is performed. The result is a global evaluation of all mentioned algorithms using objective evaluation parameters. In the end, all the resulting analyzes are evaluated and at the same time a graphical user environment was created for a better understanding and comparison of the analyzed methods.