Magnetic Resonance imaging (MRI) is a powerful imaging tool but many important clinical applications are limited by long scan times and/or poor SNR. This proposal aims to improve the speed of MRI without losing SNR, through a Bayesian inference approach. Improvement in scan speed can enable new time-critical clinical and diagnostic MR applications, like cardiac imaging, time-resolved 4D coronary angiography, high-resolution volumetric brain imaging, dynamic contrast enhanced imaging, etc. A Bayesian framework for the reconstruction of raw MR data from multiple coils in parallel will be developed. This framework makes it possible to reduce the time taken during scanning multiple times by reducing the sampling rate of raw MR data. Our method will be generally applicable to most MR imaging modalities, targets and sampling schemes. Our method will then be validated and tested on the specific clinical application of volumetric structural brain imaging, which is an important procedure for the detection and diagnosis of neurodegenerative diseases, tumors, white matter lesions, measuring brain atrophy and hippocampal subfields, etc. The main goal of this project is to create a set of computational tools to perform the reconstruction of accelerated MRI data on arbitrary imaging targets, modalities and acquisition schemes, including random sampling schemes. Design of models to capture prior spatial information about images will be undertaken. Finally, the method will be validated on structural brain data in terms of metrics like SNR, partial voluming, test- retest repeatability, and the performance of subsequent processing steps like image segmentation.
This project has the potential to make clinical MR imaging much faster than currently possible. This will make many time-critical clinical applications of MRI more feasible, for instance real-time MRI of the heart. The resolving power of MRI to image finer, clinically interesting anatomical features will also increase, making more reliable diagnosis possible.
