Title: Segmentation of Fetal 3D Brain MRIs and Spectral Brain Matching
Other Titles: Segmentierung fetaler 3D Gehirn MRTs und spektrale Gehirzuordnung
Language: English
Authors: Weiser, Paul 
Qualification level: Diploma
Keywords: Deep Learning; Spectral Graph Theory; Fetal Brain Development; Magnetic Resonance Imaging
Advisor: Langs, Georg 
Issue Date: 2020
Number of Pages: 103
Qualification level: Diploma
Abstract: 
Medical research is a very diverse field and recently machine learning has become a part of it. One area in which it can be applied particularly well is the analysis of medical image data. The segmentation of this data is a difficult task in which a lot of research is done.In this thesis, a method for the automatic segmentation of magnetic resonance images (MRI) of fetal brains is presented. In comparison, the automatic segmentation of the adult brain is already well advanced and there are several interesting results. In this case, the quantity and quality of the data as well as the complex structure of the fetal brain are a major challenge for any automatic segmentation program. A popular method for segmentation is deep learning. In this process, artificial neural networks are trained on pre-segmented data. With the experience gained further data can be evaluated independently. Convolution and the U-Net architecture are used to particularly improve the quality of the image analysis of neural networks. In the course of this thesis, experiments are carried out to find a suitable structure for a neural network. Inspired by others, several techniques such as sequencing neural networks or hierarchical structures are evaluated and implemented. To improve the spatial information of the artificial neural network, spectral coordinates are applied and a topological loss function supports the identification of the cortex. This techniques improve the automatic segmentation and lead to promising results. Especially the spectral coordinates and the topological loss function increase the performance of the network in the cortex.
URI: https://doi.org/10.34726/hss.2020.63962
http://hdl.handle.net/20.500.12708/16331
DOI: 10.34726/hss.2020.63962
Library ID: AC16084829
Organisation: E505 - Studiendekanat für Technische Mathematik 
Publication Type: Thesis
Hochschulschrift
Appears in Collections:Thesis

Files in this item:

Show full item record

Page view(s)

20
checked on Feb 21, 2021

Download(s)

22
checked on Feb 21, 2021

Google ScholarTM

Check


Items in reposiTUm are protected by copyright, with all rights reserved, unless otherwise indicated.