Thursday, May 26, 2022 • 4:50–6:05 pm (CT) • Light Hall, Room 202
Organizer and chair: Sharmistha Guha, Texas A&M University
On novel multi-object regression and efficient computation
Rajarshi Guhaniyogi, Texas A&M University
Rene Gutierrez, University of California, Santa Cruz
Aaron Scheffler, University of California, San Francisco
Clinical researchers often collect multiple images from separate modalities (sources) to investigate fundamental questions of human health that are inadequately explained by considering one image source at a time. Viewing the collection of images as multiple objects, the successful integration of multi-object data produces a sum of information greater than the individual parts. This talk focuses on multi-modal imaging applications where structural information from grey matter (GM) and brain connectivity information in the form of a brain connectome network from functional magnetic resonance imaging (fMRI) are available for multiple subjects. The inferential goal in this application resides upon developing a regression model to characterize relationship between a language score used as a measure to assess the degree of primary progressive aphasia (PPA), and these image objects that capture complementary anatomical and network connectivity information. This talk will discuss a novel multi-object regression model that exploits linkage between different objects to draw inference and prediction of a language score used to measure PPA. The proposed method belongs to the class of structured regression models. We will also briefly discuss efficient computational strategies in the structured regression models.
Examining the association between depression outcomes and EEG characteristics via functional regression with missing data
Adam Ciarleglio, George Washington University
presented by Yuhao Zhang, George Washington University
Electroencephalography (EEG) is a neuroimaging modality that is used to collect information about brain function by measuring electrical activity at electrodes placed on the scalp. The data derived from EEG are rich with different kinds of information that may be useful as potential biomarkers for depression severity or antidepressant treatment response for those diagnosed with major depressive disorder. In this talk, we will examine the association between EEG-derived current source density curves and the outcomes of depression severity and antidepressant treatment response via a functional regression modeling approach using data from a clinical trial of placebo vs. sertraline. In the data set, some subjects are completely missing their EEG data but have other relevant variables available. In previous work, we have implemented a multiple imputation approach to handle such missing functional data. We extend this approach here by allowing for more flexible imputation models. We present results from a simulation study of the performance of these new imputation procedures as well as results from applying the procedures to the clinical trial data.
A multi-subject Bayesian model for dynamic functional connectivity with time-varying covariate-dependent transitions
Michele Guindani, University of California, Irvine
Jaylen Lee, University of California, Irvine
Isaac Menchaca, University of California, Irvine
Aaron R. Seitz, University of California, Riverside
Xiaoping Hu, University of California, Riverside
Ryan Warnick, Microsoft
Marina Vannucci, Rice University
Megan A. K. Peters, University of California, Irvine
Time Varying Functional Connectivity (TVFC) aims at investigating how the brain’s functional networks evolve through the course of an fMRI experiment. In this talk, we consider a multi-subject multivariate Bayesian framework where the networks are estimated through the classification of latent neurological states, leading to sparse connectivity networks in an integrated framework that borrows strength over the subjects as well as over the entire time course of an experiment. In practice, not only the connectivity networks but also the probability of transitioning between states may depend on subject- and experiment- specific information, recorded simultaneously with the fMRI data. Our model overcomes the typical assumption of stationary transition kernels by allowing for transitions to depend on time-varying exogenous variables, leading to an improved characterization of subjects’ dynamics during the experiment. We evaluate the performance of our algorithm via a simulation study and further apply our modeling framework to the analysis of a handgrip task experiment with concurrently-recorded pupillometry data.