Brain Connectivity Analysis: from Unimodal to Multimodal: Sergey Plis
Abstract: Cortical neurons form coherent networks that can either be identified anatomically by geodesic proximity on the cortex or functionally by coherent activations in the measurements. In the latter, the networks are surprisingly stable across subjects and conditions forming a set of functional units of the brain. Understanding the interactions among these units can help us to better understand brain function and, possibly, explain some mental disorders by disruption of the connectivity pattern and its effect on the interactions. Cross-correlation matrices are the most common way of assessing brain's connectivity (usually referred to as functional connectivity) but they have drawbacks (e.g. inability of distinguishing direct and indirect influence) that make us turn to modeling the connectivity via the directed graph of a Bayesian network (effective connectivity). The goal is to bring together multimodal information to improve effective connectivity estimates.
This talk covers our work on estimating the brain's effective connectivity from (electro-)magneto-encephalography, and functional MRI. Although, the results of unimodal connectivity analysis correspond to some of our understanding of brain function and provide new insight, focusing on a single modality can be misleading. An example will be presented where connectivity structures estimated from two modalities not only are different but may also lead to contradicting conclusions. Motivating this result further with an example of changes in apparent connectivity direction based on modality I will talk about our approach to this problem: multimodal connectivity analysis.