Spectral Causality in Multivariate Signals: Beyond Linearity

• Activity Number: 14 - Translational Methods for the Assessment of Brain Function
• Type: Invited
• Date/Time: Sunday, July 29, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Hernando Ombao* and Abdulrahman Althobaiti
• Companies: King Abdullah University of Science and Technology and Rutgers University and King Abdullah University of Science and Technology
• Keywords: Time Series; Spectral analysis; Causality; Non-stationary; Brain Signals; Coherence

Abstract

There has been a strong interest in causality testing in brain signals. Most of the current methods are based on vector autoregressive models and the limitations include potential model misspecification and the ability to capture only linear associations. Our proposed approach will be to study causality via oscillatory activities (dependence between different frequency bands). Of prime interest here is the notion of ``spectral causality" which is broadly characterized as the extent to which an oscillatory activity in a population of neurons can predict various oscillatory activities in another region at a future time point. Using these oscillations, we will build a class of functional multivariate cross-frequency oscillatory models so that our method can capture potential non-linear dependence of the present and past oscillatory activity. The new framework will be illustrated to multichannel local field potential recorded during an induced stroke with the goal of differentiating the causality structure pre-stroke and post-stroke in rats. This experiment was performed at the Frostig neurobiology laboratory at UC Irvine.