98
IRUS TotalDownloads
Altmetric
Robust multi-structure segmentation of magnetic resonance brain images
File | Description | Size | Format | |
---|---|---|---|---|
Ledig-C-2015-PhD-Thesis.pdf | Thesis | 78.18 MB | Adobe PDF | View/Open |
Title: | Robust multi-structure segmentation of magnetic resonance brain images |
Authors: | Ledig, Christian |
Item Type: | Thesis or dissertation |
Abstract: | Magnetic resonance (MR) imaging is a powerful technique for the non-invasive in-vivo imaging of the human brain. In this thesis several robust techniques are developed that allow the fully-automatic analysis of MR brain images. In particular an approach is presented that quantifies the volume of more than 100 individual structures within the whole brain. This methodology is extended to measure structural volume changes based on images acquired at multiple time points. The possibility to quantify volumetric change of numerous brain structures simultaneously sets the method apart from many established approaches that measure change of individual structures only. The presented methodology is robust in the presence of disease-related atrophy or severe brain deformation. The proposed algorithms are evaluated on widely used reference datasets. The obtained results compare favourable to state-of-the-art techniques. The methods are employed to perform a cross-sectional and a longitudinal analysis of a large cohort of Alzheimer’s disease (AD) patients. Employing the extracted information (volume, atrophy) as features it is shown that AD related disease states can be classified with high accuracy. In addition a prospective cohort of traumatic brain injury (TBI) patients is studied. Structural volumes are extracted at the acute disease stage and the extent of structural change occurring within the months following the head trauma is quantified. The results show that cerebral white matter atrophy is increased in TBI and that the involvement of individual brain structures, such as the hippocampus or the thalamus, is particularly predictive for the outcome of the head injury. These results confirm the methods’ potential to support large scale imaging studies by robustly extracting markers with clinical interpretation from MR brain images. In addition, a novel evaluation measure is proposed to quantitatively assess the accuracy of automatically calculated image segmentations with respect to a reference segmentation. |
Content Version: | Open Access |
Issue Date: | Jul-2015 |
Date Awarded: | Dec-2015 |
URI: | http://hdl.handle.net/10044/1/28959 |
DOI: | https://doi.org/10.25560/28959 |
Supervisor: | Rueckert, Daniel |
Sponsor/Funder: | European Union |
Funder's Grant Number: | 270259 EP/K503733/1 |
Department: | Computing |
Publisher: | Imperial College London |
Qualification Level: | Doctoral |
Qualification Name: | Doctor of Philosophy (PhD) |
Appears in Collections: | Computing PhD theses |