Exercise 6, functional connectivity

Exercise: Functional connectivity

The data required in the assignment can be downloaded from http://neuro.hut.fi/~jjkujala/FBI_data_conn.zip. Matlab code designed to help in the assignment is given in another zip-file (http://neuro.hut.fi/~jjkujala/FBI_code_conn.zip).

The assignment consists of two parts, and of writing a report describing and showing the main findings.

1. Functional connectivity, real vs. spurious interactions

Demonstrate how the selection of an appropriate vs. inappropriate reference region affects the estimates of cortico-cortical coherence in MEG; principally, show how field spread/spatial leakage properties change when a reference region displays activity compared to a situation where it does not. First, identify the frequency showing the most prominent rhythmic activity and the source locations showing activity at that frequency, and show what kind of coherence estimates are obtained at the same frequency when one of the two brain regions active at that frequency is taken as a reference vs. cases where the reference is placed elsewhere. The data and related variables are contained in the file P1_connectivity_data.mat; the file contains the beamformer estimates of cortical time-series (funct_data) in 29 locations along a single axis (x-coordinates given in variable locations) and the sampling frequency of the data (sff). Two of the 29 locations contained simulated sources that were mutually coherent; in the rest of the regions, no activity was simulated.


-Spectral estimation can be conducted with the function pwelch (to be applied separately in each location). Frequency resolution of ~0.5Hz is sufficient, and for visualization it is best to focus on frequencies < 30 HZ. When you have identified the frequency of interest, the spatial profile of power will be clearer the power values are averaged e.g. within a 2-Hz wide window

-Compute coherence (with function mscohere) first from one of the identified active regions to all other regions, and the from at least one region were no activity was detected. Similarly to the power-mapping, the results will look clearer if the computed coherence values are averaged within a 2-Hz wide window.

2. Estimation of directed interactions using Granger Causality

Estimate the appropriate model order for Granger Causality analysis, and evaluate the directed influences two time-series have on each other. The main goal is to identify which time-series is the driving one, and at which frequency. The data (variable X containing 2 time-series) and its sampling frequency (s_freq) are given in P2_granger_data.mat.


-The appropriate model order can be identified using the function cca_find_model_order. The function returns the appropriate order based on both The Akaike and Bayesian information criterion. Either one can be used in the subsequent analysis. It is sufficient to consider model orders < 20.

-The spectral Granger causality estimation can be done using the function cca_pwcausal. The function can be called using the syntax [GW,COH,pp,waut,cons]=cca_pwcausal(X,Nr,Nl,nlags,Fs,freq, STATFLAG) (note the variable order), where the number of realizations is in this case 1 and STATFLAG can be set to 0. Regarding the examined frequencies, it is sufficient to consider frequencies below 50Hz. As regards the main output (GW), one should note that the terms are estimated such that the column variable is Granger Causing the row variable.