Segmentace prostaty z obrazů magnetické rezonance

Abstract

The aim of this bachelor thesis is to introduce the basic concept of medical image segmentation. The thesis analyzes the fundamentals of prostate anatomy and pathology, as well as the possibilities of diagnosing pathological conditions of the prostate. It also focuses on the principles of magnetic resonance imaging from a physical perspective. This is followed by a description of acquisition sequences and types of artifacts that may occur in magnetic resonance images. The acquisition of knowledge about recent magnetic resonance sequences and segmentation approaches for prostate images was carried out through a literature review. In the practical part, an implementation of a segmentation approach based on the convolutional neural network U-Net was performed. Subsequently, optimization and evaluation of the output were carried out. Robustness testing was performed using Rician and Salt and Pepper noise.