Studies of the brain using magnetic resonance imaging (MRI) can provide
insights in physiology and pathology that can eventually aid clinical
diagnosis and therapy monitoring. MRI data acquired in these studies can
be difficult, as well as laborious, to interpret and analyze by human
observers. Moreover, analysis by human observers can hamper the
reproducibility by both inter- and intra-observer variability. These
studies do, therefore, require accurate and reproducible quantitative
image analysis techniques to optimally benefit from the valuable
information contained in the MRI data. In this thesis, we focus on the
development and evaluation of quantitative analysis techniques for brain
MRI data.
In the first part of this thesis, we focus on automatic brain tissue and
white matter lesion (WML) segmentation. We propose an automatic WML
segmentation method based on fluid-attenuated inversion recovery (FLAIR)
scans that can be added as an extension to brain tissue segmentation
methods. We optimize and evaluate a previously proposed automatic brain
tissue segmentation method in combination with the WML segmentation
extension. We compare the accuracy and reproducibility of this newly
developed segmentation framework to several other methods, some of which
are publicly available. Additionally, we compare two brain tissue
segmentation methods on the segmentation of longitudinal brain MRI data.
The second part of this thesis is about structural brain connectivity
based on diffusion MRI data. We propose a framework for analysis of
structural connectivity in large groups of subjects. Structural
connectivity is established using minimum cost paths based on the
diffusion weighted images and is summarized in brain networks. Using
statistical methods, we demonstrate that the obtained networks contain
information regarding subject age, white matter lesion load and white
matter atrophy. Finally, we evaluate the reproducibility of the proposed
brain connectivity framework.
No comments:
Post a Comment