Multi Modal Detection of Alzheimer’s Disease Using Structural MRI Images

Alzheimer's disease (AD) is the most prevalent case of dementia and a progressive brain disorder. It is an irreversible neurodegenerative disease characterized by a decrease in cognitive and memory functions that are ultimately sufficiently severe to interfere with ordinary operations. The disease has no cure, but the related symptoms are managed by many therapy alternatives. While current treatments cannot prevent Alzheimer's progression, early detection can momentarily assist slow down the deterioration of dementia symptoms and enhance the quality of life for people with Alzheimer's disease and their caregivers. With clinical and neuroimaging data, attempts were produced to use multiple classical machine learning algorithms to automatically diagnose this disease. More recently, in-depth learning methods have been introduced for this purpose due to their superior efficacy. In this thesis, we suggest the use of the brain's structural magnetic resonance image (sMRI), acquired from the ADNI database, to construct a model for detecting Alzheimer's disease based on a profound convolutional neural network (CNN) ensemble. The proposed method relies on initial categorization of the brain data into three different categories: the full brain, the grey matter and the white matter. Three separate CNN models are built for each of these data types based on a pretrained network called VGGNet. After training, the decision (i.e Alzheimer’s patient or Healthy Control) from each of the models are then combined using a simple majority vote to obtain a single, final decision. This approach will give better and more accurate predictions than the use of a single model. Keywords: Alzheimer’s disease, Structural MRI, CNN, VGGNet, machine learning, Deep learning, ensemble network.