Dear HCP Users, A frequently discussed topic on the HCP-Users mailing list is how to clean HCP fMRI data above and beyond the recommended spatial ICA + FIX cleanup that has already been carried out. Several papers have noted that there is residual structured noise in HCP data and have suggested that global signal regression is needed to remove this noise. At the same time, it is known that there are downsides to global signal regression in that it will also remove global (or semi-global) neural signal in addition to global noise. For this reason I and others have recommended using partial correlation, or if full correlation is required, that analyses be run with and without global signal regression to assess whether results are consistent across the two approaches and are not due to either global noise or the removal of global neural signal. If the desired output is full correlation; however, this is not an ideal solution.
I have just posted on BioRxiv https://www.biorxiv.org/content/early/2017/09/27/193862 a manuscript that describes a method to get around this problem. We use temporal ICA to decompose the fMRI data into temporally independent components, allowing us to separate global signal and global noise components. Temporal ICA works for global components because it aims to maximize temporal independence, rather than spatial independence (it is hard to make global spatial maps uncorrelated with all other components as is required by spatial ICA). We then regress the temporal ICA noise components out of the fMRI data and show that this removes residual global noise in the HCP data. Should the manuscript successfully navigate peer review, we will use the method to clean the HCP fMRI data and make the code available in the HCP Pipelines. The method works at the group level, rather than the individual subject level, and would run after the HCP’s spatial minimal preprocessing pipelines, ICA+FIX, and MSMAll. We are sharing this manuscript as a preprint ahead of peer review so that the community is alerted to this new approach to the global noise dilemma and to solicit feedback about the work. The topic of global signal regression has engendered strong opinions on both sides of the debate, including within the HCP consortium. The manuscript acknowledges that the positions on both sides have merits and limitations, but proposes a data driven best of both worlds approach. Matt. _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users
