Detection and interactions of dynamic communities in brain functional connectivity imaging

Abstract

The way the brain connects its different areas to provide its unequaled level of efficiency is an aspect of the brain working that is currently not fully understood. A way to characterize it is to map the networks constituting functional connectivity owing to rest-functional Magnetic Resonance Imaging (rest-fMRI), a non-invasive imaging technique. Owing to connectivity metrics such as correlation or coherence calculated between rest-fMRI signals from different regions of the brain surface, Resting State Networks (RSNs) are constructed. Although functional networks share topological similarities with anatomical networks, they are not static as it has been assumed until 2010. The complexity of the problem is thus increased and methods are developed to extract dynamic properties from functional networks with a high enough time resolution. The main approach currently used to analyze dynamic functional connectivity identifies a finite set of connectivity states consisting of activity pattern reoccurring across time and subjects. However, the functional connectivity also varies spatially over time and the representation of dynamic functional connectivity as states is too restrictive. The aim of this master thesis is to develop a new approach to represent dynamic functional connectivity as workable networks respecting both spatial and temporal variability. It is inspired by community evolution mining in social networks and text topics and proposes a richer alternative to connectivity states. The functional networks are thus considered as dynamic communities that interact across time. These interactions are characterized at different levels by events and their analysis provides insights in the functional organization of the brain. A clean and robust representation of the dynamic functional connectivity and their interactions across individuals is thus established and applied to a population of 200 subjects. Six different dynamic communities are thus identified across the population. They share similarities with static RSNs and they are repeatable across subsets of the population . The temporal characteristics of their activation allow to detect recurrent pattern in their co-occurrence and they are characterized by events whose significance can be evaluated.

The way the brain connects its different areas to provide its unequaled level of efficiency is an aspect of the brain working that is currently not fully understood. A way to characterize it is to map the networks constituting functional connectivity owing to rest-functional Magnetic Resonance Imaging (rest-fMRI), a non-invasive imaging technique. Owing to connectivity metrics such as correlation or coherence calculated between rest-fMRI signals from different regions of the brain surface, Resting State Networks (RSNs) are constructed. Although functional networks share topological similarities with anatomical networks, they are not static as it has been assumed until 2010. The complexity of the problem is thus increased and methods are developed to extract dynamic properties from functional networks with a high enough time resolution. The main approach currently used to analyze dynamic functional connectivity identifies a finite set of connectivity states consisting of activity pattern reoccurring across time and subjects. However, the functional connectivity also varies spatially over time and the representation of dynamic functional connectivity as states is too restrictive. The aim of this master thesis is to develop a new approach to represent dynamic functional connectivity as workable networks respecting both spatial and temporal variability. It is inspired by community evolution mining in social networks and text topics and proposes a richer alternative to connectivity states. The functional networks are thus considered as dynamic communities that interact across time. These interactions are characterized at different levels by events and their analysis provides insights in the functional organization of the brain. A clean and robust representation of the dynamic functional connectivity and their