An architecture integrating neuroimaging and cognitive testing for the early detection of Alzheimer's disease and the Braak staging

Details

Supervisors

Macq, Benoît

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en sciences informatiques, à finalité spécialisée

Abstract

The Alzheimer's disease (AD) is the predominant cause of dementia and present a diagnostic challenge that necessitate the detection of amyloid for indentification and the tau protein for Braak staging. As the prevalence of AD increases, the demand for early detection will rise while the equipment used for its early detection may not be sufficient to meet this need. Therefore, it is necessary to study this disease in order to identify as many biomarkers as possible using easily accessible equipment and technologies. During this thesis, the main goal is to develop an architecture capable of predicting preclinical cases of AD and Braak staging using primarly Magnetic resonance imaging (MRI) scan segmentations and cognitive test, aiming to construct a solid baseline for future studies willing to include additional medical examinations. We compared two algorithms to achieve this goal: Support Vector Machine and Random Forest. Additionally, we explored different strategies to build the datasets and the most significant features were retrieved using explainable artificial intelligence techniques to ensure models alignment with the existing literature. To predict preclinical case of AD, we achieved an AUC of 86% and to predict Braak staging, we achieved an F1 score of 66%. However, limitations such as class distribution imbalance need to be taken into account.