"In Canada, Alzheimer's disease (AD) and multiple sclerosis (MS) affect respectively 2.1 and 0.3% people. AD affects memory, day-to-day cognitive ability and represents 60 to 80 percent of dementia cases in the elderly while MS is the most common neurological disability in young adults. These two common neurodegenerative diseases have no cure, but approaches to help with early diagnosis, treat their symptoms, and delay their progression are currently under heavy international research investigation. While the neurodegenerative processes that occur during these diseases are quite different, they both can result in widespread and sometimes subtle anatomical changes over time. Neurodegenerative changes, such as loss of neurons or loss of other building blocks of the brain (e.g., myelin), can - over time - result in cerebral atrophy, defined as the macroscopic loss of brain parenchymal volume. Reliably quantifying these changes will provide new insights into disease progression with immense potential implications for disease prognosis, treatment strategy and monitoring therapeutic effects. Magnetic resonance imaging (MRI) enables the macroscopic visualization of brain structure and anatomy, and allows the non-invasive assessment of neurodegenerative phenomena (e.g., atrophy), as well as their temporal evolution. Although automatic image analysis methods for the measurement of anatomic changes in the brain have been used extensively for the estimation of morphological differences, they have been in large part limited by technological approaches that are inherently cross-sectional (single time-point). Recently, the emergence of numerous databases with longitudinal MR images provides the necessary data to develop, test and investigate the longitudinal structural changes of brain anatomy at a much larger scale. The main objective of this thesis was to develop robust, accurate and fully automatic measures of focal longitudinal changes in the brain on MRI. Such methodology will enable a more detailed understanding of the initiation and spread of neurodegeneration, with the overall goal to comprehend the dynamic spatio-temporal distribution of brain atrophy in neurodegenerative diseases. As registration-based methods to detect brain atrophy are confounded by image intensity changes due to focal lesions, the first aim was to automatically detect the presence of focal lesions. Second, we proposed a new filling strategy to remove these lesion-induced intensity irregularities that could potentially affect subsequent analysis. For the third aim, we proposed and assessed a new method to remove potential bias in pair-wise non-linear registration used to estimate brain atrophy. Finally, we developed a spatio-temporal image registration framework, which accounts for potential biases in longitudinal image analysis. The methods developed in this thesis improve longitudinal morphological measurements and can find application in monitoring and diagnosing neurodegenerative diseases. " --