Novel Magnetic Resonance Acquisition and Processing Strategies for Biological Tissue Characterisation

Proton magnetic resonance techniques have become indispensable for characterising tissues non-invasively. These methods provide abundant information regarding metabolism, morphology and histology of the sample under study. While these techniques were more expensive in the past compared to radioactive methods, modern advances in hardware and methodology provide the potential to use magnetic resonance systems more efficiently and widely. In this context, this thesis explored innovative magnetic resonance technologies from three independent perspectives which are suitable for tissue characterisation, utilising techniques from a wide range of disciplines including physics, engineering, biology and medical sciences. One strategy relates to compressed sensing magnetic resonance imaging, seeking to recover detailed features at high undersampling rates. A data-adaptive sparse transform facilitated by principal component analysis was introduced as an alternative to the conventional pre-defined sparse transform. Moreover, the principal component analysis was used in a recognition algorithm for the reconstruction of undersampled data. The performances of these approaches were studied in cases of localised changes in the acquired images. The results demonstrated that the recognition reconstruction algorithm performed better than wavelet compressed sensing. This progress can be utilised to accelerate current state of the art imaging protocols at high magnetic field strengths. Furthermore, the prior knowledge contained in high resolution databases may enhance imaging capabilities of technologies at low magnetic field strengths. A second approach exploits nuclear magnetic resonance diffusion contrast instead of contrast agents for tissue characterisation. Microstructural information and global fractional anisotropy can be obtained from diffusion-diffusion correlation spectroscopy via a novel multi-dimensional gradient scheme. The concept was validated by random walk simulations and experiments of biological samples. Both correlation maps and global fractional anisotropy of in vitro healthy and tumour-bearing mouse brains were found to be different, thus providing a potential application of the proposed scheme in diffusion oncology. In addition, a threshold algorithm on the selection of a region of interest was implemented to minimise inter-observer variations. This technique was applied to a pilot study of diffusion weighted imaging data which were acquired from patients after x-ray mammography indicated lesions. The statistical analysis revealed an optimal threshold similar to values commonly used in positron emission tomography. Apart from selecting regions automatically, various data processing methods were implemented and compared with each other regarding their diagnostic accuracies. This field study provides opportunities for standardising procedures in diffusion weighted mammography, which may be integrated into clinical analysis in the future.