Files in this item



application/pdfFU-DISSERTATION-2017.pdf (21MB)
(no description provided)PDF


Title:High-resolution full-vocal-tract dynamic speech magnetic resonance imaging
Author(s):Fu, Maojing
Director of Research:Liang, Zhi-Pei; Sutton, Brad
Doctoral Committee Chair(s):Liang, Zhi-Pei
Doctoral Committee Member(s):Do, Minh N.; Shosted, Ryan; Woo, Jonghye
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Speech imaging
Magnetic resonance imaging (MRI)
Abstract:Dynamic magnetic resonance imaging (MRI) holds great promise for speech-related studies because of its potential to investigate velopharyngeal motion and physiological properties jointly in real time. However, many applications of dynamic speech MRI are limited by the technical trade-offs in imaging speed, spatial coverage, spatial resolution and clinical interpretation. In particular, high-resolution dynamic speech MRI with full-vocal-tract coverage and phonetically meaningful interpretation remains a challenging goal for many speech researchers. This dissertation develops novel model-based dynamic speech MRI approaches to enable high-resolution, full-vocal-tract 3D dynamic speech MRI with quantitative characterization of the articulatory motion. Our approaches include technical developments in imaging models, data acquisition strategies and image reconstruction methods: (a) high-spatiotemporal-resolution speech MRI from sparsely sampled data is achieved by employing a low-rank imaging model that leverages the spatiotemporal correlations in dynamic speech motion; (b) a self-navigated sampling strategy is developed and employed to acquire spatiotemporal data at high imaging speed, which collects high-nominal-frame-rate cone navigators and randomized Cartesian imaging data within a single TR; (c) quantitative interpretation of speech motion is enabled by introducing a deformation-based sparsity constraint that not only improves image reconstruction quality but also analyzes articulatory motion by a high-resolution deformation field; and (d) accurate assessment of subject-specific motion as opposed to generic motion pattern is realized by using a low-rank plus sparse imaging model jointly with a technique to construct high-quality spatiotemporal atlas. Regional sparse modeling is further introduced to assist effective motion analysis in the regions of interest. Our approaches are evaluated through both simulations on numerical phantoms and in vivo validation experiments across multiple subject groups. Both simulation and experimental results allow visualization of articulatory dynamics with a frame rate of 166 frames per second, a spatial resolution of 2.2 mm x 2.2 mm x 5.0 mm, and a spatial coverage of 280 mm x 280 mm x 40 mm covering the entire upper vocal tract across 8 mid-sagittal slices. Deformation fields yielded from our approaches share an identical spatiotemporal resolution that characterizes accurate soft-tissue motion. With a high-quality atlas, the low-rank and the sparse components are reconstructed to reveal both subject-specific motion and generic speech motion across a specific subject group. The effectiveness of our approaches is demonstrated through practical phonetics investigations that include (a) integrative imaging and acoustics analysis of velopharyngeal closure; (b) understanding the formation and variation in a variety of languages, American English, North Metropolitan French, Brazilian Portuguese and Levantine Arabic; and (c) analyzing motion variability of a particular subject with respect to a specific subject group. The capabilities of our method have the potential for precise assessment of the oropharyngeal dynamics and comprehensive evaluation of speech motion.
Issue Date:2017-01-03
Rights Information:Copyright 2017 Maojing Fu
Date Available in IDEALS:2017-08-10
Date Deposited:2017-05

This item appears in the following Collection(s)

Item Statistics