HCP beta-release of the Functional Connectivity MegaTrawl

Group-ICA [Smith 2014a, FastICA/MELODIC] was applied at 5 dimensionalities (d=25,50,100,200,300) to preprocessed rfMRI data, with surfacebased alignment (“MSM-sulc”) utilising folding patterns to align different subjects’ surfaces with each other [Glasser 2013, Salimi-Khorshidi 2014, Robinson 2014]. These group-ICA “parcellations” (where each ICA component comprises a “node” or “parcel”) were used to estimate associated timeseries from each subject, with two methods: multiple-spatial-regression against the group-ICA spatial maps (“ts2”) or via eigenregression (“ts3”, [Smith 2014b]). From node timeseries, network matrices (netmats) were estimated using partial correlation with limited L2 regularisation (“NetmatMethod=3”, setting rho=0.01 in the Ridge Regression netmats option in FSLNets). This Parcels, Timeseries and Netmats (PTN) dataset (and associated documentation) is available at db.humanconnectome.org/data/projects/HCP_500 – as is the MegaTrawl.