Software pro automatizovanou identifikaci a modelování pankreatu a tukové tkáně z MR obrazových dat

Abstract

This master thesis describes the physiology and pathology of the pancreas with emphasis on fat presence. It also describes the principles of magnetic resonance imaging and methods for identifying the pancreas from these images. The aim of this thesis is the pancreas segmentation from real magnetic resonance images and quantification of pancreas fat tissue. Images from the T1-VIBE DIXON sequence were used for the segmentation. A hybrid method was created, which uses U-Net neural network and the Sparse field method, which works on the principle of active contours. The quantification of adipose tissue is solved by calculating the fat fraction, which is calculated from the fat and water map provided by the Dixon sequence. For ease of use, a user interface has been created to allow segmentation of pancreatic images, as well as evaluation of segmentation quality and adipose tissue quantification. The implementation of all experiments was performed in the MATLAB environment.