fMRI analysis

1.     Analysis starts with Human Connectome Project (HCP) resting eyes-open fMRI data, which has already been pre-processed by the HCP consortium (WU-Minn HCP Consortium et al., 2013):

a.     Note, the pre-processed HCP fMRI data has been mapped to a standard cortical surface using a multi-modal registration algorithm, MSMAll (Robinson et al., 2014; Glasser et al., 2016), and for which structured artefacts (with origins including motion, heartbeat and cerebro-spinal fluid) have been removed by a combination of ICA and FIX (Salimi-Khorshidi et al., 2014), FSL’s automated noise component classifier.

2.    Parcel time courses are computed using the first stage only of dual regression (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/DualRegression)

3.    Partial correlation matrices are computed using Tikhonov regularisation (λ=0.01); using nets_netmats(ts,1,'ridgep',0.01) from FSLnets (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLNets).

a.     Note that the call to nets_netmats sets the flag do_rtoz=1 so that the partial correlations are converted to Z-values using Fisher’s transform. 

                                               i.     To do this conversion (from r2z) we need to know the scaling factor which is the effective temporal DOF. The code in nets_netmats  estimates the scaling by generating null data with the same AR(1) as the real data and seeing what scaling is necessary to get Z to have stddev 1.

4.    Group-level networks are estimated by:       

a.     Separately Z-transforming each correlation matrix for each session.

b.     Averaging over all sessions.

Category