Longitudinal PET Image Biomarkers for Conversions of Mild Cognitive Impairment to Alzheimer's Disease

• Activity Number: 346
• Type: Topic Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Yanming Li* and Bin Nan and Ji Zhu
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: Alzheimer's disease (AD) ; Cox propotional hazard ; Cure model ; Mild cognitive impairment (MCI) ; Positron emission tomography ; Ultra-high dimensional variable selection

Abstract

Mild cognitive impairment (MCI) is believed a precursor of Alzheimer's disease (AD). However, not all MCI patients will develop AD in their lifetime. It is of great interest to understand who will eventually become AD and when if they will. The positron emission tomography (PET) brain image is among the most promising biomarkers that may be useful in predicting AD conversion. There are two major challenges for analyzing such data. One is that the PET voxel level image data are ultra-high-dimensional; one is that patients may drop out from a study due to limited follow up time or other reasons, thus it is unclear if they would eventually develop AD. Motivated by the ADNI study, we consider a mixture cure survival model to investigate the association between longitudinal PET images and the conversion status of MCI patients. In such a modeling strategy, the MCI patient population is considered a mixture of AD converters and non-converters where the mixing proportion follows a logistic model, and the interval-censored time to AD conversion for converters is assumed to follow the Cox model. The elastic net method is implemented in estimating the high-dimensional parameters.