Automatická lokalizace a klasifikace jaterních lézí

Ing. Tomáš RYBA

Doctoral thesis

Automatic localization and classification of liver lesions

Abstract:

Computer-aided diagnostic (CAD) systems are widely used in technical and me\-di\-cal fields. Using the CAD systems in medicine allows the application of image processing methods as well as the methods of artificial intelligence. The purpose of the systems is to assists doctors. A radiologist needs to diagnose a great amount of image data, which is very focus-demanding work. Using a CAD system can support doctor's effectiveness in the sense of processing speed and/or accuracy. The goal of the thesis is to develop a CAD system for the automatic localization and subsequent classification of liver lesions. Liver cancer is mostly diagnosed from a differential diagnosis that consists of analysis of two serial CT screening. Because the screenings are taken at the time interval of several seconds, data registration needs to be performed. In both series the liver region is found using a fully autonomous method based on the Grow Cut algorithm and the results obtained are further refined by a localized active contour method. The liver region is then analyzed and searched for lesions. The localization of the lesion is performed by Markov Random Fields initialized with a combination of saliency maps. The lesions found are then paired-up and classified by a decision tree.

Abstract:

Computer-aided diagnostic (CAD) systems are widely used in technical and me\-di\-cal fields. Using the CAD systems in medicine allows the application of image processing methods as well as the methods of artificial intelligence. The purpose of the systems is to assists doctors. A radiologist needs to diagnose a great amount of image data, which is very focus-demanding work. Using a CAD system can support doctor's effectiveness in the sense of processing speed and/or accuracy. The goal of the thesis is to develop a CAD system for the automatic localization and subsequent classification of liver lesions. Liver cancer is mostly diagnosed from a differential diagnosis that consists of analysis of two serial CT screening. Because the screenings are taken at the time interval of several seconds, data registration needs to be performed. In both series the liver region is found using a fully autonomous method based on the Grow Cut algorithm and the results obtained are further refined by a localized active contour method. The liver region is then analyzed and searched for lesions. The localization of the lesion is performed by Markov Random Fields initialized with a combination of saliency maps. The lesions found are then paired-up and classified by a decision tree.

Keywords

diagnostic, liver, image processing, lesion, localization, classification, registration, saliency map, Markov random fields, active contours, decision tree

Language used: English

Date on which the thesis was submitted / produced: 23. 11. 2016

Accessible from:: 31. 12. 2999

Thesis defence

• Date of defence: 3. 10. 2017

University of West Bohemia

Faculty of Applied Sciences

Doctoral programme / field:
Applied Sciences and Computer Engineering / Cybernetics