Supervisors: Dr. Christian Baumgartner, Viktor Wegmayr, Prof. Konukoglu
One of the challenges in identifying the effects of Alzheimer's disease (AD) is that the development of the disease is occurring together with aging, i.e. subjects develop AD as they get older. The effects of aging and disease are observed together in the magnetic resonance imaging (MRI) scans of the human brain. Therefore, it is important to come up with algorithms that automate the identification and disentanglement of age and disease related effects at a subject-specific level. In this work, we propose an encoder-decoder architecture, called Age and Disease Modifier Network (ADMnet), to predict the subject-specific effects of aging and AD on human brain. ADMnet is able to modify the observed age and disease condition of the input image by feeding different age and disease condition values into the network. ADMnet achieves almost perfect disentanglement of the effects of aging and disease on the synthetic dataset and outperforms the state-of-the-art. ADMnet is also largely successful in disentangling age and disease related changes on the real 3D neuroimaging dataset, where its findings generally agree with the medical findings about the effects of aging and AD on human brain.