Alzheimer's prognosis using a deep survival model

Details

Supervisors

Lee, John Aldo ; Gurpreet Singh

Faculty

Ecole polytechnique de Louvain

Degree label

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

Abstract

Alzheimer's Disease (AD) is progressive neurodegenerative disease that results in the loss of brain tissue. Predicting the behavior of patients diagnosed with MCI, a prodromal stage to AD, who are not known to eventually develop dementia, is a challenging problem in Alzheimer research. A lot of studies have addressed this issue in the recent years. To address this task, we propose here the use of a model using the characteristics of both survival analysis and deep learning, DeepSurv. The data consist of MRI images from the ADNI research center. To transform these high-dimensional data into a feature vector of reasonable size, we developed a novel clustering-based feature extraction technique. We reached AUC values up to 77% and a c-index of 71%.