Machine learning for early prediction of the onset of Alzheimer’s dementia using MRI, PET and cognitive tests

Details

Supervisors

Macq, Benoît

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique, à finalité spécialisée

Abstract

Context: Alzheimer’s disease’s rising incidence poses a significant challenge for global healthcare systems, progressively impairing cognitive abilities and leading to dementia. Its prevalence increases with age, exacerbating concerns as global life expectancy rises. Objective: The thesis aims to develop a predictive model determining the onset of dementia in Alzheimer’s disease’s patients accurately. In addition, a secondary goal is to distinguish between patients who will progress to dementia and those who won’t. Methods: The research starts by meticulously analysing the relevant data sources, including MRI, PET scans and cognitive test data. Initially, the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm and Principal Component Analysis (PCA) will be used to visualise the data. Following this, the study will delve into diverse machine learning classifiers like Support Vector Machines (SVM) and Multilayer Perceptron (MLP) models. Results: The best model found used an MLP and achieved an accuracy of 0.6667 for predicting the onset of dementia. When trying to predict whether the patient will progress to dementia, an SVM achieved the highest accuracy with 0.9524. Conclusion: The models developed in this thesis have limitations due to data class imbalance, which affects the accuracy of dementia onset prediction. Addressing this imbalance is crucial to improve the accuracy of dementia onset prediction. However, the model for predicting whether a patient will progress to dementia is very accurate and shows promising results in this specific area.