Identifying and Utilizing Coupled Oscillatory Netwoks using the Complex Multivariate Normal Distribution

Group members involved in the project

Eric Maris, Roemer van der Meij

Background and aim of the project

Many studies have shown the relation between, on the one hand, the power in neuronal oscillations, and on the other hand, cognitive phenomena, stimulus variables, and motor behavior. However, very few studies have investigated the relation between, on the hand, the phases of distributed neuronal oscillations, and on the other hand, cognitive, stimulus and behavioral variables.

One of the few studies that did investigate the functional role of distributed phase configurations is the one by Canolty et al (2010), which shows that spikes may be strongly phase coupled to the oscillations in distributed neuronal populations, sometimes at a considerable distance from the spiking neuron. A methodology highly similar to the one used by Canolty et al (2010), can not only be used to investigate the relation of distributed phase configurations with neuronal spiking, but also its relation with cognitive, stimulus, and behavioral variables.

A convenient methodological tool for this type of study is a multivariate probability distribution for the distributed phases. Canolty et al (2010) use a distribution proposed by Cadieu & Koepsell (2010). This distribution distribution has a number of disadvantages: (1) it ignores the amplitudes of the oscillations (of which we know that they are related to both spiking activity and cognitive, stimulus, and behavioral variables), (2) its fit to the data is not motivated, and (3) parameters is difficult. I propose as an alternative the complex multivariate normal (CMN) distribution, which allows the same type of investigation as the distribution proposed by Cadieu & Koepsell (2010), but has none of its three disadvantages.

The CMN distribution is an excellent tool to build a BCI that uses the phase information in the neuronal signal in order to classify this signal. Making use of the phase information makes a lot of sense both when using invasive recordings and MEG. For the latter, this is because oscillating dipoles involve a phase difference of 180 degrees between the two lobes of the electromagnetic dipole. Moreover, classification on the basis of a probability distribution for the data is exceptionally easy because it does not require extraction of predictive feature. Instead, it just involves an application of the likelihood-ratio criterion.

Activities as a part of this stage project

This project involves method development and analysis of existing data (invasive and MEG). The following existing data sets are available: (1) MEG data collected in paradigms involving attentional orientation and sensory evidence accumulation, (2) rat ECoG data obtained in our own lab, and (3) human ECoG data collected by Michael Kahana.

For a number of analyses steps, the student will make use of FieldTrip, a MATLAB toolbox for EEG/MEG analysis developed at the Donders Institute.

The student is invited to join the weekly meeting of the Systems Neuroscience Journal Club and the Eletrophysiology Data Analysis (EDA) meeting.Â