Calculate functional connectivity

Here we describe a procedure to calculate seed-based functional connectivity.

Morph volumetric brain data onto the cortical surface

The original fMRI data has four dimensions: three in space and one in time. Here we transformed this 4D data into a 2D representation: the 3D brain volume is converted onto cortical surface locations.

This "volume to surface" procedure has been a common practice in analyzing fMRI signal over the cortex at the cost of ignoring brain activity in deep brain structures and cerebellum.

Because the brain volume from each participant has been spatially "normalized" (to the MNI152 space) but keeping the image voxel size during data collection (3 x 3 x 3.4 mm), we first transformed the brain volume to the MNI305 space with 2-mm isotropic resolution by this Matlab script.

Then, we used the same transformation to convert the signal for each time point and each participant to a target subject (fsaverage in FreeSurfer) using this script.

The results are two STC files, each of which contains a two dimensional matrix (cortical location x time points). STC files can be read by Matlab script or Python script.

Group time series at the cortical surface into different regions

We use the Destrieux Atlas to define 75 brain regions over the cortical surface at each hemisphere. Two "annotation files", each of which is for one hemisphere, are provided with FreeSurfer. These files are here and here if you dont' have FreeSurfer.

Calculate the functional connectivity between regions

We first remove the temporal fluctuations not related to brain activity. These signals include the head motion (rotation and translation) and the time series in the head that has minimal neural activity, such as ventricle and white matter. This Matlab script can get the time series at the white matter and ventricles for each participant, whose data were in the MNI305 space with 2-mm isotropic resolution. At this step, a registration matrix is needed to transform the volume (in MNI305 space) to the cortical surface (in MNI305 space). Therefore, an identity matrix is provided.

A linear regression model is then used to remove these confounding factors.

We then calculate the Pearson correlation coefficients between pairs of brain regions. This gives a correlation coefficient matrix of 150 x 150. The upper (or lower) off-diagonal entries are functional connectivity between two regions. This is done by the Matlab script.

Here is the Matlab data file with a variable fconn for the calculated functional connectivity for subjects.