Alzheimer's Disease consists of a complex cascade of pathological processes, leading to the death of cortical neurons and development of dementia. Because it is impossible to regenerate neurons that have already died, a thorough understanding of the earlier stages of the disease, before significant neuronal death has occurred, is critical for developing disease-modifying therapies. The various components of Alzheimer's Disease pathophysiology necessitate a variety of measurement techniques. Image-based measurements known as biomarkers can be used to assess cortical thinning and cerebral blood flow, but non-imaging characteristics such as performance on cognitive tests and age are also important determinants of risk of Alzheimer's Disease. Incorporating the various imaging and non-imaging sources of information into a scientifically interpretable and statistically sound model is challenging. In this thesis, I present a method to include imaging data in standard regression analyses in a data-driven and anatomically interpretable manner. I also introduce a technique for disentangling the effect of cortical structure from blood flow, enabling a clearer picture of the signal carried by cerebral blood flow beyond the confounding effects of anatomical structure. In addition to these technical developments in multi-modal image analysis, I show the results of two clinically-oriented studies focusing on the relative importance of various biomarkers for predicting presence of Alzheimer's Disease pathology in the earliest stages of disease. In the first, I present evidence that white matter hyperintensities, a marker of small vessel disease, are more highly associated with Alzheimer's Disease pathology than current mainstream imaging biomarkers in elderly control patients. In the second, I show that once Alzheimer's Disease has progressed to the point of noticeable cognitive decline, cognitive tests are as predictive of presence of Alzheimer's pathology as standard imaging biomarkers. Taken together, these studies demonstrate that the relative importance of biomarkers and imaging modalities changes over the course of disease progression, and sophisticated data-driven methods for combining a variety of modalities is likely to lead to greater biological insight into the disease process than a single modality.