Interpatient registration and analysis in clinical neuroimaging

The first part of this thesis deals with brain infarct patterns in patients with differing degrees of stenosis of the internal carotid artery. Stenosis or narrowing of the internal carotid artery (ICA) is a well-known cause of infarcts or strokes. Infarct maps are used to study differences in infarct pattern in patients with ICA stenosis. Infarct distribution maps for each stenosis category were created by mapping each CT image to an average brain image and applying this transformation to the manually segmented infarcts. The construction of the average CT brain image proceeded in two steps. First, a temporary average image was made, based on a small number of images. The temporary average image was used as reference image for the construction of the real average image. The use of a temporary average provides a better starting point than the use of a randomly selected image. Three different average images based on the same data were created. This was important for the evaluation of the registration consistency. The use of an average brain image as reference results in more consistent registration than the use of a single image. ICA stenosis was categorized into mild (0-49% stenosis), severe (50-99% stenosis), or occlusion (100% stenosis). In total, 80 patients with mild stenosis, 33 with severe stenosis, and 29 with occlusion of the ICA were included. Differences between the infarct maps were analyzed using a non-parametric randomization based technique. Subtle differences were found between mild stenosis and occlusion maps and between severe stenosis and occlusion maps, but not between mild and severe stenosis maps. Differences were located in the territory of the middle cerebral artery. Patients with ICA occlusion had larger infarct volumes than patients with mild or severe stenosis. Volumes between mild and severe stenosis did not differ. The second part of this thesis is focused on the impact of diabetes mellitus type 2 (DM2) on brain volume and on white matter lesion volume and pattern. Magnetic resonance images (Inversion Recovery (IR) and Fluid Attenuated Inversion Recovery (FLAIR)) of 99 patients with DM2 and 46 controls were used. The images were segmented into white matter, gray matter, cerebrospinal fluid, lateral ventricles and white matter lesion (WML) using an automated k-Nearest Neighbour classification algorithm. Linear regression analyses adjusting for intracranial volume, age, gender and level of education revealed a significant decrease in gray matter volume in DM2 patients and significant increases in the volume of the lateral ventricles and of the WMLs. These differences were more pronounced in women. Analyses of differences in WML location in a random subset of subjects (61 DM2 patients and 26 controls) were directed at the white matter area around the lateral ventricles. The analyses were performed by non-rigid mapping of segmentations of cerebrospinal fluid, including the lateral ventricles, to a reference lateral ventricles image. Evaluation of this method showed an accurate matching of the lateral ventricles to the reference image. This deformation is used for WML mapping. WML patterns are made by summation of the mapped WMLs of DM2 patients and controls separately. Analysis of differences in WML pattern showed more frequent lining and capping of the lateral ventricles in DM2 patients. A separate analysis of men and women showed significant differences only in women