UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

Machine learning based computational models with permeability for white matter microstructure imaging

Hill, Ioana Diana; (2020) Machine learning based computational models with permeability for white matter microstructure imaging. Doctoral thesis (Ph.D), UCL (University College London). Green open access

[thumbnail of IoanaHill_PhDthesis_with_corrections.pdf]
Preview
Text
IoanaHill_PhDthesis_with_corrections.pdf

Download (36MB) | Preview

Abstract

Characterising tissue microstructure is of paramount importance for understanding neurological conditions such as Multiple Sclerosis. Therefore, there is a growing interest in imaging tissue microstructure non-invasively. One way to achieve this is by developing tissue models and fitting them to the diffusion-MRI signal. Nevertheless, some microstructure parameters, such as permeability, remain elusive because analytical models that incorporate them are intractable. Machine learning based computational models offer a promising alternative as they bypass the need for analytical expressions. The aim of this thesis is to develop the first machine learning based computational model for white matter microstructure imaging using two promising approaches: random forests and neural networks. To test the feasibility of this new approach, we provide for the first time a direct comparison of machine learning parameter estimates with histology. In this thesis, we demonstrate the idea by estimating permeability via the intra-axonal exchange time τ_i, a potential imaging biomarker for demyelinating pathologies. We use simulations of the diffusion-MRI signal to construct a mapping between signals and microstructure parameters including τ_i. We show for the first time that clinically viable diffusion-weighted sequences can probe exchange times up to approximately 1000 ms. Using healthy in-vivo human and mouse data, we show that our model's estimates are within the plausible range for white matter tissue and display well known trends such as the high-low-high intra-axonal volume fraction f across the corpus callosum. Using human and mouse data from demyelinated tissue, we show that our model detects trends in line with the expected MS pathology: a significant decrease in f and τ_i. Moreover, we show that our random forest estimates of f and τ_i correlate very strongly with histological measurements of f and myelin thickness. This thesis demonstrates that machine learning based computational models are a feasible approach for white matter microstructure imaging. The continually improving SNR in the clinical scanners and the availability of more realistic simulations open up possibilities of using such models as imaging biomarkers for demyelinating diseases such as Multiple Sclerosis.

Type: Thesis (Doctoral)
Qualification: Ph.D
Title: Machine learning based computational models with permeability for white matter microstructure imaging
Event: UCL (University College London)
Open access status: An open access version is available from UCL Discovery
Language: English
Additional information: © The Author 2020. Original content in this thesis is licensed under the terms of the Creative Commons Attribution 4.0 International (CC BY 4.0) Licence (https://creativecommons.org/licenses/by/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request.
UCL classification: UCL
UCL > Provost and Vice Provost Offices
UCL > Provost and Vice Provost Offices > UCL BEAMS
UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science
UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Chemical Engineering
URI: https://discovery.ucl.ac.uk/id/eprint/10093941
Downloads since deposit
133Downloads
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item