| Biomedical imaging technologies that were originally developed for human medical diagnosis have been adapted to study the anatomy and metabolism of small animals. However, multimodality image registration algorithms, which have been used for analysis of human images, have not been utilized as effectively in the small animal imaging community. As with the human studies, there exist many applications where image registration is a useful task in small animal imaging applications: visualizing and quantifying changes in longitudinal studies, combining complementary information from different modalities such as to employ both anatomical (MR, X-ray CT) and functional (PET, SPECT or optical imaging) information, and identifying and labelling of specific anatomical structures using an atlas. In this thesis, I review the image registration methods commonly used in human studies and show examples where they can also be very effective in small animal imaging applications. I also describe a high resolution 3D digital mouse atlas, Digimouse, for use in the small animal imaging community, using CT, PET and cryosection data, in which I used image registration techniques reviewed earlier. I also show applications of Digimouse to generation of simulated data for use in the evaluation of PET and optical image reconstruction algorithms, and also to the automatic labelling of anatomical structures using X-ray CT images.; In the third chapter of this thesis, I describe a new method for reconstructing a surface model of the mouse from structured light data. This method has applications in 3D optical imaging of bioluminescence and fluorescence where the surface of the animal is needed to define a mesh for solution of the forward and inverse problems. I develop a method which can derive height maps from different views and integrate computed height maps into a single estimate of the surface of the animal using the system geometry and an iterative closest point algorithm (ICP). I validate my method by comparing the surfaces with those obtained from the corresponding X-ray CT surface of the mouse.; With increasing use of mouse models of human disease, many imaging studies involve lesions. This presents a challenge for automated labelling by registration to the mouse atlas, since the lesions are not present in the atlas. Even with human studies this is a challenging problem since no correspondence to a tumor, exists in an atlas of normal anatomy. I approach this problem by seeding the atlas image with an earlier stage of the tumor which I have formulated using level sets with a uniform growth assumption and applying the viscous fluid registration algorithm which can model large scale deformations due to the tumor.; In the final chapter (5) of the thesis, I describe a new method developed for the automated extraction of skull and scalp boundaries from T1-weighted MR images - in this with applications to human rather than small animal imaging. |
