Shuo Chen, of the Division of Biostatistics and Bioinformatics and the Maryland Psychiatric Research Center at University of Maryland, School of Medicine, will present:

“Modeling Whole-brain Magnetic Resonance Spectroscopy Neurometabolite Imaging Data with Latent and Organized Dependence Structures”

Abstract: The recent development of magnetic resonance spectroscopy (MRS) imaging technology provides an opportunity to study the associations between neural metabolite concentrations in multiple brain areas and peripheral phenotypes. Understanding interactions between neural metabolite concentrations at distinct brain locations are both critical to study the underlying neurophysiological pathways and perform multivariate statistical inference. We show that the covariance structure of the multivariate MRS data shows a well-organized and latent topological pattern, however, it does not depend on the geometric distance. We propose a new autoregressive model with neighborhoods defined by a data-driven approach (l0 shrinkage), and develop more computationally efficient algorithms to estimate the dependence parameters. We apply our method to an MRS data of 204 study subjects and simulation data sets. The results show the accurately estimated covariance structure can improve statistical power and reduce false positive findings.

A social tea will be held at 3:00 p.m. in A434 Mayo. All are Welcome.