Generalized Linear Model Estimates for Brain Metabolite Signatures in MRS Data

• Activity Number: 119
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Timothy W. Randolph*+ and Jaroslaw Harezlak and Madan Gopal Kundu
• Companies: Fred Hutchinson Cancer Research Center and Indiana University School of Medicine and Indiana University School of Medicine
• Keywords: functional data ; GSVD ; neuroimaging ; spectroscopy ; structured penalty ; regression

Abstract

In neuroimaging studies, magnetic resonance spectroscopy (MRS) is used to quantify concentrations of metabolites. Each sample is an MRS curve collected from various brain regions. Each curve is typically preprocessed to extract features and estimate metabolite concentrations. In clinical studies, a common goal is to model patient disease status as a function of metabolite profiles. In this case, the two-stage approach of feature extraction followed by regression limits the analysis to a few metabolites and introduces extraneous error. However, using entire MRS profiles as predictors in a logistic regression model is ill-posed and requires constraint. Our single-step approach forms a solvable regression problem by incorporating a penalty operator that is informed by MRS spectra. The estimate is derived from "partially empirical eigenvectors" obtained from the generalized singular value decomposition which reveals statistical properties of the estimator. The estimation may be carried out within the familiar linear mixed model framework. A simulation study and an application to the analysis of HIV patients' neurocognitive impairment as a function of MRS data are presented.