Advances in the Role of Quantitative NMR in Medicine: Deep Learning applied to MR Fingerprinting and Trabecular Bone Volume Fraction Estimation through Single-Sided NMR

Barbieri, Marco (2020) Advances in the Role of Quantitative NMR in Medicine: Deep Learning applied to MR Fingerprinting and Trabecular Bone Volume Fraction Estimation through Single-Sided NMR, [Dissertation thesis], Alma Mater Studiorum Università di Bologna. Dottorato di ricerca in Fisica, 32 Ciclo. DOI 10.6092/unibo/amsdottorato/9236.
Documenti full-text disponibili:
[img] Documento PDF (English) - Richiede un lettore di PDF come Xpdf o Adobe Acrobat Reader
Disponibile con Licenza: Creative Commons Attribution Non-commercial No Derivatives 4.0 (CC BY-NC-ND 4.0) .
Download (52MB)

Abstract

Nuclear Magnetic Resonance (NMR) has been a powerful and widespread tool since its birth thanks to its flexibility in assessing properties of physical systems without being invasive and without using ionizing radiations. Although applications of NMR for medical purposes have rapidly developed since the introduction of MR imaging (MRI), most of the clinical protocols retrieve qualitative information about biological tissues. Being able to retrieve also quantitative information with NMR may be beneficial to identify biomarkers for understanding and describing the pathophysiology of complex diseases in many tissues. However, established quantitative MRI (qMRI) methods require long scan times that not only can represent more exposure to image artifacts and more discomfort for the patient, but they also increase the costs of MRI protocols. To improve the clinical feasibility of quantitative NMR, one can focus on optimizing qMRI protocols to increase data acquisition efficiency, i.e. minimizing the acquisition times and maximising the number of retrieved information. Alternatively, one can focus on the application of low-cost, portable and low maintenance NMR devices in the medical field, such as single-sided devices. This Ph.D thesis presents studies that aim to advance the role of quantitative NMR in medicine using the two directions stated above. The first part of the thesis proposes a deep learning approach based on deep Fully Connected Networks (NN), for pixel-wise MR parameter prediction task in Magnetic Resonance Fingerprinting (MRF) as a solution to overcome the curse of dimensionality affecting the gold standard dictionary approach. The second part proposes a methodology to assess the trabecular bone-volume-to-total-volume (BV/TV) ratio using single-side NMR by means of NMR relaxometry measurements. Nowadays there are not well-established methodologies to assess trabecular BV/TV that are suitable for wide screening campaigns of the population at risk of bone fractures related to diseases such as osteoporosis.

Abstract
Tipologia del documento
Tesi di dottorato
Autore
Barbieri, Marco
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
32
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
NMR, MRI, qMRI, MR Fingerprinting, Deep Learning, Trabecular Bone, single-sided NMR, Bone Volume Fraction
URN:NBN
DOI
10.6092/unibo/amsdottorato/9236
Data di discussione
16 Marzo 2020
URI

Altri metadati

Statistica sui download

Gestione del documento: Visualizza la tesi

^