Early detection of Alzheimer disease from path integration data using machine learning and statistical analyses

Details

Supervisors

Massart, Estelle

Faculty

Ecole polytechnique de Louvain

Degree label

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

Abstract

enEarly detection of Alzheimer’s disease (AD) is a critical challenge in modern healthcare. In fact, it has become a key factor in effective disease management. Identifying cognitive decline in its early stages maximizes the benefits of available therapeutic interventions, significantly improving the quality of life of patients. In this context, spatial navigation tasks have shown promising potential as diagnostic tools. This thesis continues the work initiated by Dr. Colmant et al., which explores the use of the Apple Game, a visual spatial navigation task, to distinguish between the effect of age and the effect of an increased genetic risk of developing sporadic AD on path integration ability. Building on that foundation, my work aims to reproduce and validate the original results, while also extending the analysis through additional metrics and machine learning techniques. Statistical analysis confirmed some previous findings while challenging others. Notably, the path length ratio revealed an effect of APOE status in the 61-70 age group under specific conditions. Machine learning models, including a random forest and an LSTM-based classifier, faced challenges due to class imbalance. And the LSTM-based autoencoder struggled with data variability. Overall, this work advances methodological approaches and helps guide future research on how AD affects path integration.