The human brain is certainly the most complex biological system as it contains a network of more than l0 to the power of 11 individual nerve cells and interconnections. Fibre tractography using diffusion MR imaging is a promising non-invasive method for reconstructing the 3D fibre architecture of the human brain white matter in vivo. Despite the great potential, white matter tractography is relatively immature. At the current resolution of diffusion MR images, researchers agree that more than one third of imaging voxels in human brain white matter contain crossing fibre bundles. Generally, conventional diffusion tensor imaging (DTI) fibre tracking approaches have difficulties in crossing regions. Also, noise and other artefacts associated with diffusion MR data lead to uncertainty in the estimates of fib re orientation directions. Furthennore. each fib re tracking method has limitations due to the algorithmic approach that they follow and the assumptions they make. This thesis presents novel probabilistic based fibre tracking algorithms aiming to tackle a number of limitations of existing fibre tracking algorithms. Fibre clustering is a key step towards tract-based, quantitative analysis of white matter. Clustering algorithms analyse a collection of fibre curves in 3D and delineate them into anatomically distinct fibre tracts groups. In this thesis, a probabilistic framework is developed and the framework al lows for the clustering of sets of cunres In curve space. This thesis describes a number of original contributions to the field. First, a novel statistical framework is developed for improved fibre tractography and a quantitative analysis tool is introduced for probabilistic tracking methods using the statistical measures. The goal is to elucidate problems with existing detenninistic and probabilistic algorithms used to process diffusion MR images and propose solutions and methods through a new framework. Subsequently, random-walk and modelbased bootstrapping algorithms are developed using a two-tensor field to quantify the uncertainty of fibre orientation and probabilistic fibre tractography. A further problem tackled here is resolving crossing fibre configurations, a major concern in diffusion MR imaging, using data that can be routinely acquired in a clinical setting. Finally, a new probabilistic clustering algorithm is introduced using regression mixtures and the result of clustering is the probabilistic assignment of the fibre trajectories to each cluster. The tract geometry model is estimated using fitted parameters of the probabilistic clustering algorithm. Local reconstruction, tracking results, segmentation and quantitative analysis are shown on simulated datasets, on a hardware phantom and on multiple human brain datasets.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:592018 |
| Date | January 2012 |
| Creators | Ratnarajah, Nagulan |
| Publisher | University of Kent |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.003 seconds