Abstract
Magnetic resonance imaging (MRI) is a medical imaging technique that generates excellent soft-tissue contrast and allows for investigating both anatomy and function of tissues noninvasively. In conventional MRI, direct HR acquisition requires long scan times to achieve adequate precision and spatial resolution of the resulting MR image. From a diagnostic perspective, long scan times increase the likelihood of motion artefacts, whereas, from an economical perspective, they reduce the throughput. In addition, long scan times cause discomfort for patients. Multi-slice super-resolution reconstruction (MS-SRR) has the potential to reduce this limitation, improving the inherent trade-off between resolution, SNR, and scan time. MS-SRR consists in estimating a 3D high-resolution (HR) image from a series of 2D multi-slice images with a low through-plane resolution.
Two strategies are conventionally adopted to acquire data for an MS-SRR experiment. The first consists in acquiring a set of multi-slice images with parallel orientations, where each image is shifted in the through-plane direction by a different, sub-pixel distance. The second consists in acquiring rotated multi-slice images, where each image is rotated around the frequency and/or phase encoding axis by a different rotation angle. These two strategies will be compared in terms of accuracy and precision of the reconstructed images.
MS-SRR estimation is generally an ill-posed problem and the use of regularization has an impact on the SRR estimated image. I will investigate a Bayesian SRR framework in which local correlation information is learnt from MRI images and used to stabilize the SRR estimate. An optimal experimental design framework will be developed in which the Bayesian Mean Squared Error (BMSE) of the MAP estimator is proposed as a performance criterion, to compare the two aforementioned acquisition strategies in the context of regularized MS-SRR.
We plan to validate the BMSE-based predictions on simulated and real data. Finally, we aim to extend the MS-SRR optimal experimental design framework to quantitative SRR (qSRR). In qSRR, a high resolution (HR) relaxation parameter map is estimated from a series of weighted multi-slice images with a low through-plane resolution. Each slice of each LR image can be acquired with different weighting settings, thus offering maximum flexibility to optimize the weighting settings for each slice individually. Aiming at the highest attainable precision for a given acquisition time, we will optimize the experiment design of the SRR framework by searching for the optimal acquisition parameters. This research is expected to further improve the trade-off between signal-to-noise, resolution, and scan time in qSRR, by for example allowing precise estimation of HR parameter maps from shorter scans.
Researcher(s)
Research team(s)
Project website
Project type(s)