Which network caused the traffic? | A new approach to infer effective connectivity from fMRI signals
Description see below
Imagine you have access to a tape containing the conversations recorded at a family reunion. All family members are sitting at a large table, having dinner together and talking to each other in a language you don't understand. You might think that under these conditions the tape is of no use to you, as it is impossible to extract any information from this mess. However, what you hear is not just random noise. Individual speakers can be identified from the pitch of their voice or other features of their vocal tone. But what else can you hope to extract from such a recording? Can you maybe even infer the intricate social network characterizing this family?
If somebody speaks, you cannot really know who is listening, and whether the message sent was intended for any specific receiver. However, the fact that two different individuals raise their voices in a coordinated fashion, one after the other or simultaneously, might already reveal their involvement in a one-on-one conversation. To draw a conclusion like this, there is actually no need to understand the meaning of the messages they exchange.
However, just analyzing one pair of speakers at a time introduces some serious ambiguities. Imagine that the uncle living far away asks his sister's family for some impressions of their recent holiday trip. Several members of the addressed family may respond, one after another or simultaneously. Just listening to their voices and analyzing the temporal pattern of their speech might leave you with the wrong impression that they were talking to each other. However, the correct interpretation is that they were just responding to the same question. Another obvious issue is that the communication patterns may evolve over the course of time. Those who had an exchange with some family members earlier may tend to talk to different ones later.
What is the significance of this metaphor for brain research? It is now standard in clinical practice to record neuronal activity from the entire human brain using non-invasive techniques like functional magnetic resonance imaging, fMRI. Although we have no clue what information the activation or inactivation of a small brain volume conveys, we are still interested in inferring the communication channels between them. Knowing and analyzing the topology of brain networks is, in fact, considered a crucial step towards understanding brain function. It also appears that some cases of brain dysfunction are reflected in specific perturbations of these networks, so detecting them could be part of the diagnostic process.
An interdisciplinary team of researchers from the Bernstein Center Freiburg and the University Medical Center Freiburg has now made important progress by devising procedures that can infer networks from standard fMRI recordings. In contrast to existing methods, the new approach they developed can identify both the direction and the sign of effective interactions in large networks with high fidelity and reliability. In two new publications they explain the idea behind the procedure. They test it with the help of computer simulations where the underlying networks are actually known and report about first applications of the new method to a set of recordings in humans. Carolin Lennartz (Physics) and Jonathan Schiefer (Mathematics/Biology) have been pursuing this collaborative research as part of their respective doctoral theses.
Figure legend
Whole-brain effective connectivity inferred from human resting-state fMRI recordings. Shown here are positive (red) and negative (blue) directed connections among 45 areas in each brain hemisphere.
Original publications
Schiefer J, Niederbühl A, Pernice V, Lennartz C, LeVan P, Hennig J, Rotter S
From Correlation to Causation: Estimation of Effective Connectivity from Continuous Brain Signals based on Zero-Lag Covariance
PLOS Computational Biology 14(3): e1006056, 2018
Lennartz C, Schiefer J, Rotter S, Hennig J, LeVan P
Sparse Estimation of Resting-State Effective Connectivity from fMRI Cross-Spectra
Frontiers in Neuroscience 12: 287, 2018
Contact
Prof. Dr. Stefan Rotter
Professor of Computational Neuroscience
Managing Director, Bernstein Center Freiburg
Faculty of Biology
University of Freiburg
Hansastr. 9a
79104 Freiburg, Germany
Tel.: +49 (0)761 203 9316
E-mail: stefan.rotter@bio.uni-freiburg.de