Modeling of tubular structures and fibers in in vivo data: revealing asymmetry in human vasculature and white matter fiber tracts

Official URL: http://risc01.sabanciuniv.edu/record=b1640296 (Table of Contents)

In medical image analysis, asymmetric modeling of vasculatures, such as cerebro- or cardio-vessels, is a challenging task because of the n-furcated branching geometries. Detection of asymmetry in anatomical structures such as vessels is a signi cant step in the accurate modeling of vasculatures. Therefore, it is essential to present new computational methodologies to model the underlying asymmetries in anatomical structures. For the asymmetric modeling of vasculatures, which is the rst part of this thesis, we present a vasculature segmentation method that is based on a cylindrical ux-based higher order tensor (HOT). On a vessel structure, the HOT naturally models branching points as well as the tubular sections of the vessels. We demonstrate quantitative validation of the proposed algorithm on synthetic complex tubular structures, cerebral vasculature in Magnetic Resonance Angiography (MRA) datasets and coronary arteries from Computed Tomography Angiography (CTA) volumes. Capturing asymmetry in white matter (WM) bers is another open problem. Detection of asymmetry in bers is important in both the localization and the quantitative assessment of speci c neuronal pathways. More than 60% of WM ber populations make crossings in a voxel, therefore, it is natural to expect a substantial part of those to involve asymmetric crossings/junctions. However, most well-known white matter ber reconstruction methods assume symmetric signal acquisition that yield symmetric orientation distribution functions (ODFs) even when the underlying geometry is asymmetric. In the second part of this thesis, we employ inter-voxel ltering approaches through a cone model to reveal more information regarding the cytoarchitectural organization within the voxel. The cone model facilitates a sharpening of the ODFs in some directions while suppressing peaks in other directions, thus yielding an asymmetric ODF (AODF) eld. The feasibility of the technique is demonstrated on in vivo data obtained from the MGHUSC Human Connectome Project (HCP) and Parkinson's Progression Markers Initiative (PPMI) Project database. Quantitative Susceptibility Mapping (QSM) reconstruction is a recent technique for venous imaging. The reconstruction of QSM image volume is a challenging problem due to its long acquisition time, which causes several artifacts that need to be handled separately using a regularization term in the reconstruction. Prior knowledge such as smoothness and sparsity assumptions has been widely used as regularization. We hypothesize that incorporation of local orientation of vessels into regularization leads to an enhanced imaging of vasculatures. In the last part of this thesis, we present vessel orientation as a new regularization term to improve the accuracy of l1- and l2-norm regularized QSM reconstruction in cerebral veins. Using a multi-orientation QSM acquisition as gold standard, we show that the QSM reconstruction obtained with the vessel anatomy prior provides up to 40% Root-Mean-Square-Error (RMSE) reduction relative to the baseline l1 regularizer approach.