MEG

Magnetoencephalography at the CCNi

Research In Depth

Localising Granger Causality

In recent years there has been significant effort in developing methods for identifying causality in information flow between brain areas. A family of such methods study the frequency characteristics of Multivariate Autoregressive Models (MAR) built on time-series of activated brain sources , with the most widely used being Partial Directed Coherence and Directed Transfer Function.

These methods operate on the derived coefficients of MAR models describing the linear interdependence between activation time series at locations inside the brain. These time-series are reconstructed from MEG sensor data through spatial filters derived commonly by beamforming algorithms . The large number of potential activation sources produced by such algorithms corresponds to a  large number of activation time-series, which is prohibitive for the derivation of robust MAR models.

In this work we investigate the derivation of MAR model directly on MEG sensor data and the projection of ONLY the model coefficients in the source space. By this method the modelling process is performed on the sensor space which has moderate dimensionality as compared to the high-dimensional source space.  This leads to greater model robustness as well as significantly reduced computation times.

It is also evident that the accuracy of this projection largely depends on the beamforming algorithm used to estimate the spatial filters. Thus we work towards improving existing methods for the estimation of spatial filters with main aims to reduce cross-talk noise, to prevent elimination of coherent sources in different locations and to reduce biasing in the smoothness of the spatial filters due to the brain volume geometry.

The Mechanisms of neural communication

How are the ~1011 neurons in the brain organized structurally and functionally to endow us with our remarkable information processing capabilities? What are the mechanisms balancing segregation and integration, excitation and inhibition? How does the brain achieve its highly efficient communication between different neural entities on different spatial scales? These fundamental and exciting questions are at the very heart of our research in the field of Neuroscience.
 
In summary, our research aims at unravelling some of the basic rules governing interneuronal communication (at different spatial scales) and focuses on the following interconnected neuroscientific aspects:
 
The Mechanisms of neural communication
 
Filtering and resonance
Filtering and resonance phenomena describe the dependence of neuronal activity on the frequency content of the input and are implemented at the level of synapses, single neurons and neuronal populations. They result in the transmission of selected aspects of information, enhance the sensitivity to a certain input frequency and are subject to plastic changes.
 
Structural connectivity
In general, neurons in the brain are tightly interconnected. The abundance of short-range connections is the structural basis for the emergence of functional local networks. A small percentage of long-range connections greatly reduces the minimal pathlength between any two neurons. This structural connectivity pattern is optimal for selectively biasing local and global network communication. Changing communication demands might lead to changes in anatomical connectivity.
 
Spatio-temporal pattern of excitation and inhibition
In general, inhibition is important for the emergence of oscillations in networks of principal cells and interneurons. The output of neurons with a balanced excitatory and inhibitory input is particularly sensitive to correlated input. Inhibitory interneurons also provide a possible mechanism for synchronization of the LFP oscillations of remote neural populations.
 
 
 
Although neural oscillations are prominent features in electrophysiological recordings their functional role is still unclear. I am interested in further exploring this aspect e.g. by focusing on information processing in primary sensory areas. In addition, our studies of pathological synchronization in movement disorders demonstrated the feasibility and importance of connectivity analysis using patient data.
 
Information from various measurement modalities as well as anatomical connectivity data should be combined with knowledge of biophysical properties of large-scale brain areas to design models of large-scale brain dynamics. It seems that (subcortico-cortical and cortico-cortical) loops will play a major role in these models. Comparison of model output with experimental data and systematic variations of model parameters will likely yield further insights into basic mechanisms of brain dynamics.
 
 
 

Identification of Causality between Brain Areas

 In recent years there has been significant effort in developing methods for identifying causality in information flow between brain areas. A family of such methods study the frequency characteristics of Multivariate Autoregressive Models (MAR) built on time-series of activated brain sources , with the most widely used being Partial Directed Coherence and Directed Transfer Function.

 

These methods operate on the derived coefficients of MAR models describing the linear interdependence between activation time series at locations inside the brain. These time-series are reconstructed from MEG sensor data through spatial filters derived commonly by beamforming algorithms . The large number of potential activation sources produced by such algorithms corresponds to a  large number of activation time-series, which is prohibitive for the derivation of robust MAR models.  

In this work we investigate the derivation of MAR model directly on MEG sensor data and the projection of ONLY the model coefficients in the source space. By this method the modelling process is performed on the sensor space which has moderate dimensionality as compared to the high-dimensional source space.  This leads to greater model robustness as well as significantly reduced computation times.

It is also evident that the accuracy of this projection largely depends on the beamforming algorithm used to estimate the spatial filters. Thus we work towards improving existing methods for the estimation of spatial filters with main aims to reduce cross-talk noise, to prevent elimination of coherent sources in different locations and to reduce biasing in the smoothness of the spatial filters due to the brain volume geometry. 

Neural mechanisms of sensory awareness

How does sensory information reach conscious access? According to Dehaene and colleagues (2006), sensory awareness relies on two factors: bottom-up stimulus strength and top-down attentional amplification. Therefore, environmental information will reach consciousness whenever it is sufficiently strong to induce reverberation in local sensory brain areas as well as being amplified by long-distance parieto-frontal loops. This project aims to investigate the oscillatory dynamics underlying sensory awareness. To this end, we are using a spatial cue paradigm with near-threshold stimulation to account for the influence of both bottom-up and top-down factors.

 

As early as the 19th century, Wilhem Wundt, the founder of experimental psychology defined psychology as the science of conscious experience. At that time the study of conscious perception was based on indirect behavioural measures. In the last years, with the arrival of neuroimaging techniques, we have gained the opportunity to look inside the brain in search of the neural mechanisms that allow us to be conscious of the world around us.

In the last decades, a growing body of study has aimed to characterize the neural correlates of sensory awareness (Babiloni et al. 2006; Boly et al. 2007; Hanslmayr et al. 2007; Kaiser et al. 2005; Linkenkaer-Hansen et al. 2004; Melloni et al. 2007; Palva et al. 2005; Sergent et al. 2005). However, the results of these studies are sometimes inconsistent; whereas some of them identify sensory regions as the neural substrate of conscious perception, others show that a parieto-frontal network is also involved.

Recently, Dehaene et al. (2006) have proposed that these apparently incongruent results might be explained by a unified framework: the global neuronal workspace model. According to this theory, sensory awareness relies on two factors. The first one of these is bottom-up stimulus strength. If the stimulus strength does not exceed a dynamic threshold information is not accessible, that is, subliminal. The second factor is top-down attentional amplification. If attention is absent, information is potentially accessible, although not yet accessed (preconscious state in Dehaene’s words). Therefore, environmental information will reach consciousness whenever it is sufficiently strong to induce reverberation in local sensory brain areas as well as being amplified by long-distance parieto-frontal loops giving rise to global synchrony.

 

In this project, we are using magnetoencephalography (MEG) to study the oscillatory dynamics underlying sensory awareness. More specifically, we are using a spatial cue paradigm with near-threshold stimulation to account for the influence of both bottom-up and top-down factors (see figure 1).

 

 

Neural dynamics underlying Individual and Joint Actions

 In our everyday interactions, we engage in individual and collaborative joint actions to achieve intended goals, e.g. moving a piece of furniture in and out of a room. We are interested in how joint actions might differ from individual actions, particularly in terms of the underlying neural processes: Do similar areas of the brain engage when you act on your own and with others? How do these brain regions interact with regards to e.g. your anticipation of others’ actions, and how other’s behaviour influences your own? 

A major challenge of such a neuroimaging study is the involvement of more than one participant or creating the effect that a participant is really engaging in a task jointly with another person. We hope to achieve this using a modified Simon-task paradigm.
 
Research Activity in Detail