High-dimensional time-to-event prediction for Alzheimer's disease prognosis

High-dimensional time-to-event prediction for Alzheimer’s disease prognosis

Title: High-dimensional time-to-event prediction for Alzheimer’s disease prognosis

 

Type: Master Thesis

Student: Stefan Andraschko

Supervisor: Sebastian Pölsterl, Christian Wachinger

Collaboration with: Christian Böhm

Status: Finished on 04.07.2019

 

Abstract:

Predicting the conversion from mild cognitive impairment (MCI) to Alzheimer’s disease (AD) is among the clinically most relevant diagnostic tasks in the field of AD research. This thesis compares predictive accuracy of MCI conversion to AD to investigates risk factors and to determine which biomarker features are related to disease progression, lending support to clinical decision-making. It uses demographic information, cognitive measures, single-nucleotide polymorphism (SNP), magnetic resonance imaging (MRI), cerebrospinal fluid (CSF), and positron emission tomography (PET) features from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. For efficient selection of the subset of the most relevant predictor variables a Cox proportional hazard (PH) regression model with elastic net regularisation is used. The best model achieved a concordance index of 0.863 and selected the cognitive tests ADAS13 (C-index: 0.791; mean time-dependent AUC: 0.846) and CDRSB (C-index: 0.761; mean time-dependent AUC: 0.837) with a coefficient of 0.633 and 0.501 respectively as the features with the highest probability to predict the conversion from MCI to AD.