Non-rigid image registration for enhanced localization in magnetic resonance brain imaging

Technical advances in magnetic resonance imaging (MRI) have lead to clinical and scientific applications with higher spatial and temporal resolution. One area of interest is in neuroscience and neuropathology studies towards revealing the ambiguities of human brain function and mental brain disorders. As more challenging patient populations are imaged and more advanced experiments are designed, there is an increasing demand for more accurate localization of brain function and structure. Localization of functional information and fusion of magnetic resonance images has been addressed through image registration techniques. Accurate rigid and affine registration techniques are now part of the standard brain image processing toolkits, which aim at correcting the effect of head motion and repositioning into a sub-voxel range. However, local spatial distortions may cause typical localization errors of up to 10 millimeters in magnetic resonance images acquired by Echo-Planar Imaging (EPI). This compromises the accuracy of localization and registration in functional MRI, Diffusion Tensor Imaging and Arterial Spin Labeling. Such spatial distortions are mainly caused by field inhomogeneity and susceptibility artifacts and cannot be compensated by rigid and affine registration techniques. Therefore, the main thrust of this dissertation has been the deployment of non-rigid registration for the correction of local spatial distortions in EPI. As a main challenge, quantitative in-vivo validation criteria have been devised to support the routine use of the developed techniques as part of a toolkit named NPTK, or Neuroimage Processing ToolKit. The outcome of this research makes up the core component of NPTK for enhanced localization. The technical developments on the optimization aspects of information similarity measures carried out as part of this research is generally applicable to high-dimensional optimization problems including non-rigid registration, information theoretic learning, pattern classification, and bioinformatics.