Flexible State-Space Models with Application to Brain Signals

• Activity Number: 587
• Type: Topic Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Hernando Ombao* and Zhe Yu and Raquel Prado
• Companies: UC Irvine and UC Irvine and UC Santa Cruz
• Keywords: brain signals ; state-space models ; coherence ; spectral analysis

Abstract

In this talk, we shall develop a novel family of state-space models that will be applied to estimating both local activation and between-region connectivity in both functional magnetic resonance imaging (fMRI) and electroencephalogram (EEG) data. In analyzing fMRI data, we shall model connectivity directly through the neuronal activation magnitude as represented by the amplitude of the blood oxygen level dependent (BOLD) response. As opposed to the purely exploratory approaches such as those based on cross-correlations and the vector auto-regressive (VAR) models, the proposed state space model offers an alternative philosophy to assess connectivity. It allows the activation magnitude to evolve over time, and models connectivity directly through the interconnections among activation across brain regions, instead of those among the unexplained noise in the observed signal.

The proposed family of state-space models can be adapted to a number of brain signals. The talk will focus on applications for fMRI but, as time permits, we shall also illustrate its applications to electroencephalograms.

This is work in collaboration with the joint UC Irvine - UC Santa Cruz Space-Time Modeling