Novel methods for multi-modal MR image analysis with applications to Parkinson's

Parkinson’s Disease (PD) affects approximately 1% of the population above the age of 65.The morphological changes in PD are typically not obvious, even to the trained eyes. Thus computational methods are needed to determine the existence and potential significance of robust changes in brain morphometry in PD. In this thesis we present a number of novel signal and image processing methods, that when applied to different Magnetic Resonance Imaging (MRI) techniques, can be used for the diagnosis and monitoring of PD. We first motivate the research by measuring the extent of residual anatomical variability (RAV) after spatial normalization (Chapter 1). In Chapter 2, we present a novel brain morphometry method, based on the analysis of the deformation field required to map template regions of interest (ROIs) to the corresponding ROIs in the individual brains images to be assessed. Chapter 3 summarizes the development of an algorithm to improve registration of the midbrain region, the primary site of pathology in PD. We also propose a novel method for joint analysis of structural images and diffusion tensor images (DTI) using the Fukunaga-Koontz Transform (FKT), which jointly transforms anatomical and DTI data into the same spatial components, but with complementary loadings (Chapter 4). This allows selection of linearsubspaces which are heavily weighted on DTI changes relatively insensitive to residual morphological differences, and vice versa. All the proposed methods have been validated with clinic MRI data of both PD subjects and age-matched normal control subjects. To complement the above studies, we further designed a module-based work flow to extract time courses from function MRI (fMRI) signals, using automated segmentation of structural MRI data as masks (Appendix A). This has facilitated the analysis of fMRI studies in PD, such as examining the compensatory changes seen in PD.