Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.731145
Title: Imaging biomarkers in paediatric brain resection MRI
Author: Spiteri, Michaela
ISNI:       0000 0004 6494 5433
Awarding Body: University of Surrey
Current Institution: University of Surrey
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
High resolution brain magnetic resonance (MR) images acquired at multiple time points across the treatment of a patient allow the quantification of localised changes brought about by disease progression. The aim of this thesis is to address the challenge of performing automatic longitudinal analysis of magnetic resonance imaging (MRI) in paediatric brain tumours. The first contribution in this thesis is the validation of a semi-automated segmentation technique. This technique was applied to intra-operative MR images acquired during the surgical resection of hypothalamic tumours in children, in order to assess the volume of tumour resected at different stages of the surgical procedure. The second contribution in this thesis is the quantification of a rare condition known as hypertrophic olivary degeneration (HOD) in lobes within the brain known as inferior olivary nucleii (ION) in relation to the development of posterior fossa syndrome (PFS) following tumour resection in the hind brain. The change in grey-level intensity over time in the left ION has been identified as a suitable biomarker that correlates with the occurrence of posterior fossa syndrome following tumour resection surgery. This study demonstrates the application of machine learning techniques to T2 brain MR images. The third contribution presents a novel approach to longitudinal brain MR analysis, focusing on the cerebellum and brain stem. This contribution presents a technique developed to interpolate multi-slice 2D MR image slices of the brain stem and cerebellum both to infill gaps between slices as well as longitudinally over time, that is, in four-dimensional space. This study also investigates the application of machine learning techniques directly to the MR images. Another novel method developed in this study is the Jacobian of deformations in the brain over time, and its use as an imaging feature. Unlike the previous contribution chapter, the third contribution is not hypothesis-driven, and automatically detects six potential biomarkers that are related to the development of PFS following tumour resection in the posterior fossa. The limited number of patients considered in each study posed a major challenge. This has prompted the use of multiple validation techniques in order to provide accurate results despite the small dataset. These techniques are presented in the second and third contribution chapters.
Supervisor: Lewis, Emma ; Guillemaut, Jean-Yves ; Windridge, David ; Avula, Shivaram Sponsor: Rabn Ezra Trust
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.731145  DOI: Not available
Share: