Super-resolution estimation of quantitative MRI parameters
6 September 2016
UAntwerpen, Campus Drie Eiken, Promotiezaal Q0.02 - Universiteitsplein 1 - 2610 Antwerpen-Wilrijk (route: UAntwerpen, Campus Drie Eiken
Organization / co-organization:
Department of Physics
Gwendolyn Van Steenkiste
J. Sijbers, B. Jeurissen & D. Poot
PhD defence Gwendolyn Van Steenkiste - Faculty of Science, Department of Physics
Magnetic resonance imaging (MRI) is a versatile non-invasive imaging modality. Although MRI is mostly used to acquire qualitative images, in which the signal intensities are arbitrary units, it can also be used to obtain quantitative parameters. These parameters can provide biomarkers that potentially allow the identification of various pathologic conditions. Despite the numerous applications of quantitative MRI (qMRI), many of them are seldom used in clinical practice since they require the acquisition of reliable data, which to this day still remains challenging. One reason is the need for a large number of images in order to estimate these parameters, which results in long scan times. Lying still for long periods of time often becomes uncomfortable for the patient and increases the chance of patient motion and consequently inaccurate measurements. Furthermore, each acquired MR image is corrupted by noise, i.e. random variations in the signal. The precision of the measurement linearly scales with the signal-to-noise ratio (SNR). One way to increase the SNR is by averaging over multiple acquisitions. However, this increases the acquisition time even more. Therefore, MR images are often acquired at a low spatial resolution since bigger voxels contain more signals and thus have a higher SNR. Unfortunately, this also introduces another tradeoff. Decreasing the spatial resolution increases the prevalence of partial volume effects, i.e. voxels contain multiple tissue types and structures. Consequently, small anatomical structures cannot be visualized.
Recent work has shown that the tradeoff between the spatial resolution, SNR and acquisition time can be improved by applying super-resolution reconstruction (SRR), a technique in which a high resolution image is estimated from a set of lower resolution images. Existing SRR techniques, however, do not take the parametric model into account. The aim of this thesis is to develop SRR methods for two qMRI techniques: T1 mapping and diffusion tensor imaging. The proposed methods substantially increase the spatial resolution of the parameter maps, while still respecting clinically feasible scan times and without compromising the precision and accuracy of the parameter estimates. Quantitative high resolution studies of the brain allow for better identification of changes in biomarkers, which can in turn improve the diagnosis of pathologic conditions, as well as the understanding of the brain and brain connectivity.