"Why" resting state functional connectivity must be restlessly dynamic?
Poster No:
1235
Submission Type:
Abstract Submission
Authors:
Mengiste Simachew1, Demian Battaglia2
Institutions:
1University of Strasbourg, Strasbourg, Bas Rhin, 2University of Strasbourg, Strasbourg, France
First Author:
Co-Author:
Introduction:
Neuroimaging recordings during resting state reveal that brain-wide functional connectivity networks are highly dynamic, undergoing a structured flow of stochastic-like reconfiguration, with events of transient stabilization of connectivity, intertwined with "leaps" with high and fast variability. Markers of dynamic functional connectivity are increasingly being used as biomarkers of various pathologies and cognitive performance.
Here we ask whether there should be some deep reason for Functional Connectivity to be as dynamic as it is observed to be. A first reason is merely mechanistic. Dynamic Functional Connectivity is a natural option because many more circuit dynamics regimes produce structural and fluid dFC than a statically quenched FC.
In another line of reasoning, we suppose that Functional Connectivity reflects in some indirect manner inter-areal information exchange. For such aim, a static network could be designed as well for the optimal dispatch of information. However, the rule of the game may change if we suppose that generating a functional connectivity link for a certain time has some cost (e.g. of metabolic nature). In this case, efficient static networks may be very costly requiring a large number of rarely used links, while randomization of links could provide a compromise allowing suboptimal but still efficient information diffusion at a smaller overall cost over time.
Considering both the extreme cases of omniscient information spreaders (upper bound to efficiency) and random information diffusion (lower bound), we explore and identify conditions in which the emergence of restlessly dynamic Functional Connectivity could be seen as the byproduct of simultaneously optimizing easiness of information diffusion, multi-scale local integration of information and the resources needed to achieve it.
Finally, we also compare the results of our theoretical models with observations of switching statistics in empirical fMRI resting state data. Specifically, we consider correlations in large rs fMRI datasets between cognitive scores in varied domains and the temporal network organization of the measured dFC, identifying marked correlations with cognitive performance.
Here we ask whether there should be some deep reason for Functional Connectivity to be as dynamic as it is observed to be. A first reason is merely mechanistic. Dynamic Functional Connectivity is a natural option because many more circuit dynamics regimes produce structural and fluid dFC than a statically quenched FC.
In another line of reasoning, we suppose that Functional Connectivity reflects in some indirect manner inter-areal information exchange. For such aim, a static network could be designed as well for the optimal dispatch of information. However, the rule of the game may change if we suppose that generating a functional connectivity link for a certain time has some cost (e.g. of metabolic nature). In this case, efficient static networks may be very costly requiring a large number of rarely used links, while randomization of links could provide a compromise allowing suboptimal but still efficient information diffusion at a smaller overall cost over time.
Considering both the extreme cases of omniscient information spreaders (upper bound to efficiency) and random information diffusion (lower bound), we explore and identify conditions in which the emergence of restlessly dynamic Functional Connectivity could be seen as the byproduct of simultaneously optimizing easiness of information diffusion, multi-scale local integration of information and the resources needed to achieve it.
Finally, we also compare the results of our theoretical models with observations of switching statistics in empirical fMRI resting state data. Specifically, we consider correlations in large rs fMRI datasets between cognitive scores in varied domains and the temporal network organization of the measured dFC, identifying marked correlations with cognitive performance.
Methods:
- To show the likelihood of emergence of a structured dynamic Functional Connectivity, we resort to connectome-based models, embedding generic human SC matrices and using a delayed rectified threshold-linear transfer function as local dynamics. Parameter sweeps are performed to quantify the relative extent of regimes with dynamic FC vs quenched FC in the parameter space.
- We define alternative temporal network ensembles to generate and study toy models of information diffusion. The key point is that generating a functional connectivity link of a certain weight has a certain cost per unit time of activation. If a link is activated for T time-steps its cost will thus be T times longer than a lin of equal weight activated for a single time-step. Through this procedure, we can compare in a fair manner information diffusion and integration on more or less static or dynamics networks.
- Information diffusion is simulated as a walk process on the dynamic networks along time-respecting paths, that can be driven either by an agent-based strategy (with some optimization goal) or by pure chance.
- We define alternative temporal network ensembles to generate and study toy models of information diffusion. The key point is that generating a functional connectivity link of a certain weight has a certain cost per unit time of activation. If a link is activated for T time-steps its cost will thus be T times longer than a lin of equal weight activated for a single time-step. Through this procedure, we can compare in a fair manner information diffusion and integration on more or less static or dynamics networks.
- Information diffusion is simulated as a walk process on the dynamic networks along time-respecting paths, that can be driven either by an agent-based strategy (with some optimization goal) or by pure chance.
Results:
- dFC is mechanistically more likely to arise than static FC
- dFC relative to a same cost static FC facilitates the income of information to a larger number of regions without affecting diffusion time.
- Empirical dFC networks are more efficient in info diffusion than their static averages, but also simultaneously optimize multi-scale information integration, unlike simpler toy models.
- Information diffusion and integration dynamic indicators correlate with cognitive performance on human rs fMRI.
- dFC relative to a same cost static FC facilitates the income of information to a larger number of regions without affecting diffusion time.
- Empirical dFC networks are more efficient in info diffusion than their static averages, but also simultaneously optimize multi-scale information integration, unlike simpler toy models.
- Information diffusion and integration dynamic indicators correlate with cognitive performance on human rs fMRI.
Conclusions:
Why the "d" in dFC? First "how": it is natural to get it. Second "what for": it has advantages on information integration and diffusion with respect to equal cost static FC.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
fMRI Connectivity and Network Modeling 2
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Computational Neuroscience
FUNCTIONAL MRI
Modeling
Systems
Other - dynamic Functional Connectivity
1|2Indicates the priority used for review