Abstract
Alzheimer's disease is the most common cause of dementia, and the disease can be characterized by aggregation of the protein Amyloid-beta Amyloid-beta is poisonous for the nerve cells and forms pathologic lesions that can be visualized by medical imaging techniques. Currently, Positron Emission Tomography (PET) is the imaging-based gold standard for detecting amyloid-beta deposits. However, PET is an expensive and time-consuming method and requires the administration of a radioactive isotope. Magnetic Resonance Imaging (MRI) is a non-invasive and more available alternative to PET. In MRI, amyloid depositions are associated with the formation of white matter hyperintensities (WMHs), and in Alzheimer's patients, a higher baseline of WMH is associated with a greater increase in Amyloid-beta. Accurate detection and quantification of WMH from MRI are therefore important in the diagnostic workup of Alzheimer's disease. Deep learning-based methods have proved powerful in a wide range of image segmentation tasks, and in this thesis, the aim is to test different deep learning approaches to automatically segment WMHs from MRI. A fully automated WMH segmentation tool is expected to aid the radiologist in the diagnostic workup in patients with early signs of Alzheimer's disease, speed up the diagnostic process, and may also reduce user bias.