Artificial magnetic resonance contrasts based on microvascular geometry: A numerical basis

Magnetic resonance imaging (MRI) is highly versatile, offering many contrast settings inherently sensitivity to tissue microstructure at the sub-voxel scale (below the imaging resolution). Since its invention, images produced with MRI have mainly been based on classical reconstructions, with contrast determined by the signal attenuation from local tissue and MRI sequence design. In the advent of machine learning becoming practical, wide availability of computational power and high-resolution imaging such as laser scanning microscopy, new processing techniques involving MRI interpretations based on comparisons with known signals and ground-truth microstructure can be explored. Data-driven signal classifications enable model-less predictions of tissue properties on a single voxel level, offering artificial MRI contrasts. In this thesis, groundwork is laid for the exploration of such contrasts and suitable MRI sequences, with a demonstration of the feasibility of such an approach based on transverse relaxation for brain tumor detection. The thesis is focused on the role of microvascular geometry on reversible transverse relaxation in the context of tumor imaging. Comprehensive quantifications of cancer-induced vessel remodeling are provided, and the effects thereof studied with MRI simulations. Consequently, a numerical framework was developed for correlations of MRI signal properties with underlying microstructure for further exploration of artificial contrasts