Caracterisation of Dementia by 3D Analysis of Cerebral Structures

Abstract

Dementia is one of the main health challenges facing our generation. It is the third leading cause of death, after heart disease and cancer. Detection of dementia from neuroimaging data such as MRI through machine learning has been a subject of intense research in a recent year. Although many works of literatures have developed many approaches to characterize this disease and to classifier its stages automatically, currently there is no specific technique to confirm with certainty the diagnosis of dementia. Brain regions are important for prediction of dementia stages. Some of the researchers were focused on the segmentation of the grey matter, others are focused on the cortical thickness. This paper investigates how to characterize and to predict dementia stages from a 3D MR image Database. Data provided by OASIS BRAIN includes a cross-section of 175 subjects aged 60 to 96 years. The approach relies on many steps. It started with applying data preprocessing which includes registration and segmentation of the hippocampus with a Bayesian probabilistic approach based on the Markovian modeling. The second step is to extract some features from the hippocampus. Finally, the prediction of dementia stages is done with a supervised learning algorithm: Artificial Neural Network. The results evince that the proposed approach can be used to characterize and predict the stages of dementia with over 95% accuracy, considerably higher than that of conventional basic methods. This study confirms that to extract geometric descriptors from the hippocampus can effectively solve the dementia characterization problem.