Recent Bayesian approaches for analysis of neuroimaging data

Date: 
January 22, 2020
Time: 
3:00 to 4pm
Place: 
MH-2700

Raquel Prado, PhD, Department of Statistics, Baskin School of Engineeing, University of California Santa Cruz

We present some recent statistical models and related computational and inferential methods for analyzing different types of neuroimaging data.  We begin discussing Bayesian approaches for detecting activation and co-activation from complex-valued fMRI data.  We show how these approaches lead to more accurate detection of activation when compared to alternative methods based on magnitude-only data.  We illustrate our results in simulation studies and in human studies.  We then present an approach for magnitude-only data that make use of Bayesian tensor regression models for joint estimation of activation and connectivity.  The approach combines low-rank tensor decompositions and multiway stick breaking priors for inferring activation at the voxel level.  Connectivity is modeled at the region of interest level using a Gaussian graphical prior structure.  These models are illustrated in the context of analyzing multi-subject fMRI data from the balloon-analog risk-taking experiment.  Finally, we present new spectral, time-domain and time-frequency approaches for analyzing multi-channel electroencephalogram data.

 

Following: Social hour.

Event Type: 
Biostatistics and Bioinformatics Seminar