Archana Venkataraman, Ph.D.
John C. Malone Assistant Professor
Department of Electrical & Computer Engineering
Johns Hopkins University
Mathematical Models for Functional Neuroimaging
Functional neuroimaging is field with all the difficulties that come from high dimensional data and none of the advantages that fuel modern-day breakthroughs in computer vision, automated speech recognition, and health informatics. It is a field of unavoidably small datasets, due to the costly acquisitions and environmental confounds, massive patient variability, and an arguable lack of ground truth information. My lab tackles these challenges by combining analytical tools from signal processing and machine learning with hypothesis-driven insights about the brain.
The first part of this talk will focus on resting-state fMRI (rsfMRI), which captures steady-state patterns of co-activity in the brain. One of the open challenges in this space is how to describe a heterogeneous patient cohort. As part of this effort, I will present a hierarchical Bayesian model to define patient-specific brain parcellations given an initial population atlas. I will demonstrate that our refinement procedure yields better task fMRI concordance in a population of brain tumor patients. I will then describe a joint optimization framework to predict clinical severity from rsfMRI data. Our model is based on two coupled terms: a generative non-negative matrix factorization and a discriminative linear regression. The final part of this talk will focus on EEG data. Here, I will develop a spatio-temporal model to track the spread of epileptic seizures. Our model relies on a latent network structure that captures the hidden state of each channel; the latent variables are complemented by an intuitive likelihood model for the observed neuroimaging measures. This effort takes a first step toward noninvasive seizure localization.