To expand on the smoothing issue a bit, even 4mm FWHM spatial smoothing in the volume causes substantial signal mixing between areas on opposite sides of sulci, which sounds like a particularly bad idea for ICA. As I understand it, group ICA shouldn't care much about spatial noise, especially when you have a lot of subjects.
However, group ICA does care about alignment, that is, the same element (voxel or vertex) having the same functional meaning across subjects - volume alignment is not good at this for most of the human cortex (largely due to folding variability across subjects), which is why we strongly recommend using surface registration and analysis for cortex. Tim On Wed, Jan 3, 2018 at 1:10 PM, Glasser, Matthew <[email protected]> wrote: > There are few circumstances that one should do unconstrained smoothing in > the volume. > > Peace, > > Matt. > > On 1/3/18, 12:53 PM, "[email protected] on behalf of > Tobias Bachmann" <[email protected] on behalf of > [email protected]> wrote: > > >Hi Stephen, > > > >Am Mittwoch, 3. Januar 2018, 16:02:06 CET schrieb Stephen Smith: > >> > On 3 Jan 2018, at 14:44, Tobias Bachmann > >> > <[email protected]> wrote: > >> > when using the HCP's ICA+FIX data (NIfTI volume files) for a rather > >>simple > >> > group ICA, would you recommend further conventional preprocessing, > >>i.e. > >> > > >> > 1. smoothing (about which conflicting info is to be found) > >> > >> spatial or temporal? > > > >Spatial. > > > >> The simple answer is in general no, although for the purposes of > >>group-ICA > >> only (and not any later subject-specific analyses like dual > >>regression), it > >> might be the case that some lowpass temporal filtering could boost > >> effective CNR. > > > >I'm indeed planning on using dual regression later on. > > > >> > 2. discarding the first few volumes (I found hardly anything regarding > >> > this > >> > issue)? > >> > >> For most purposes it's not necessary, though there are some slight > >>residual > >> starting effects in the data, so you could delete a few. > > > >Okay, thanks a lot! > > > >Tobias > > > > > >_______________________________________________ > >HCP-Users mailing list > >[email protected] > >http://lists.humanconnectome.org/mailman/listinfo/hcp-users > > > _______________________________________________ > HCP-Users mailing list > [email protected] > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
