BORIS Theses

BORIS Theses
Bern Open Repository and Information System

Improving the Clinical Use of Magnetic Resonance Spectroscopy for the Analysis of Brain Tumours using Machine Learning and Novel Post-Processing Methods

Da Silva Mendes Pedrosa de Barros, Nuno Miguel (2018). Improving the Clinical Use of Magnetic Resonance Spectroscopy for the Analysis of Brain Tumours using Machine Learning and Novel Post-Processing Methods. (Thesis). Universität Bern, Bern

[img]
Preview
Text
18barros_n.pdf - Thesis
Available under License Creative Commons: Attribution-Noncommercial-No Derivative Works (CC-BY-NC-ND 4.0).

Download (54MB) | Preview

Abstract

Magnetic Resonance Spectroscopy (MRS) provides unique and clinically relevant information for the assessment of several diseases. However, using the currently available tools, MRS processing and analysis is time-consuming and requires profound expert knowledge. For these two reasons, MRS did not gain general acceptance as a mainstream diagnostic technique yet, and the currently available clinical tools have seen little progress during the past years. MRS provides localized chemical information non-invasively, making it a valuable technique for the assessment of various diseases and conditions, namely brain, prostate and breast cancer, and metabolic diseases affecting the brain. In brain cancer, MRS is normally used for: (1.) differentiation between tumors and non-cancerous lesions, (2.) tumor typing and grading, (3.) differentiation between tumor-progression and radiation necrosis, and (4.) identification of tumor infiltration. Despite the value of MRS for these tasks, susceptibility differences associated with tissue-bone and tissue-air interfaces, as well as with the presence of post-operative paramagnetic particles, affect the quality of brain MR spectra and consequently reduce their clinical value. Therefore, the proper quality management of MRS acquisition and processing is essential to achieve unambiguous and reproducible results. In this thesis, special emphasis was placed on this topic. This thesis addresses some of the major problems that limit the use of MRS in brain tumors and focuses on the use of machine learning for the automation of the MRS processing pipeline and for assisting the interpretation of MRS data. Three main topics were investigated: (1.) automatic quality control of MRS data, (2.) identification of spectroscopic patterns characteristic of different tissue-types in brain tumors, and (3.) development of a new approach for the detection of tumor-related changes in GBM using MRSI data. The first topic tackles the problem of MR spectra being frequently affected by signal artifacts that obscure their clinical information content. Manual identification of these artifacts is subjective and is only practically feasible for single-voxel acquisitions and in case the user has an extensive experience with MRS. Therefore, the automatic distinction between data of good or bad quality is an essential step for the automation of MRS processing and routine reporting. The second topic addresses the difficulties that arise while interpreting MRS results: the interpretation requires expert knowledge, which is not available at every site. Consequently, the development of methods that enable the easy comparison of new spectra with known spectroscopic patterns is of utmost importance for clinical applications of MRS. The third and last topic focuses on the use of MRSI information for the detection of tumor-related effects in the periphery of brain tumors. Several research groups have shown that MRSI information enables the detection of tumor infiltration in regions where structural MRI appears normal. However, many of the approaches described in the literature make use of only a very limited amount of the total information contained in each MR spectrum. Thus, a better way to exploit MRSI information should enable an improvement in the detection of tumor borders, and consequently improve the treatment of brain tumor patients. The development of the methods described was made possible by a novel software tool for the combined processing of MRS and MRI: SpectrIm. This tool, which is currently distributed as part of the jMRUI software suite (www.jmrui.eu), is ubiquitous to all of the different methods presented and was one of the main outputs of the doctoral work. Overall, this thesis presents different methods that, when combined, enable the full automation of MRS processing and assist the analysis of MRS data in brain tumors. By allowing clinical users to obtain more information from MRS with less effort, this thesis contributes to the transformation of MRS into an important clinical tool that may be available whenever its information is of relevance for patient management.

Item Type: Thesis
Dissertation Type: Cumulative
Date of Defense: 2 February 2018
Subjects: 600 Technology > 610 Medicine & health
Institute / Center: 04 Faculty of Medicine > Department of Radiology, Neuroradiology and Nuclear Medicine (DRNN) > Institute of Diagnostic and Interventional Neuroradiology
Depositing User: Hammer Igor
Date Deposited: 20 Oct 2020 15:12
Last Modified: 20 Oct 2020 15:12
URI: https://boristheses.unibe.ch/id/eprint/2266

Actions (login required)

View Item View Item