CLASSIFICATION OF ALZHEIMER’S DISEASE FROM MAGNETIC RESONANCE AND PATIENTS’ METADATA

Alzheimer's disease (AD) is the most frequent type of dementia, that usually affects elderly people. It's a progressive disease that begins with modest memory loss and advances to the inability to talk and respond to the environment. Parts of the brain that control thought, memory, and language are involved in the injury that AD causes. It can have a significant impact on a person's capacity to carry out daily tasks. Since there is no completely efficient treatment for AD, it is useful to be detected in its early stage. This increases the chance of benefiting from treatment and increases its efficiency. The common way to do so is the use of Magnetic resonance imaging (MRI). MRI is being used in computer-aided diagnosis (CAD) to help the medical doctor in taking medical decisions. To increase the efficiency of AD detection using artificial intelligence-based CAD systems, the usage of patients’ metadata is being used combined with the MRI scans. The aim of this work is to study how this usage of metadata could affect CAD systems for the early detection of AD. This was done by investigating literature that combined the metadata with MRI scans. Those literature were selected by applying a two-steps criterion for selection; first step selection of 35 papers that contains the usage of the deep learning (DL) for AD diagnosis, while the second one resulted in the 7 papers that focus on the utilize of metadata for AD classification. The input of the final classification layer is almost the same in all the studies. This input is a features vector that contains two types of features. The first type is the hidden features extracted from MRI scans using DL networks. The other features are the metadata contained in the description of the scans data. This combination shows a slight increase in the overall performance of the classification problem. Some papers showed no significant change. On the other hand, introducing those metadata features results in problems such as bias of the classification model. This could be explained, because of the low number of studies. Therefore, it is advised to conduct more experiments in order to obtain uniform and generalized insights.