wb_command -file-information has a column with a count of inf/NaN values. Since -fix-zeros has an effect, I suspect that some kind of thresholding is going on in fsl_view, rather than the smoothing being the issue.
The volume parcel smoothing will output zero values anywhere that you don't have a parcel, so of course it will have that kind of "hole" regardless of the options used. The .nii.gz file is compressed, so being 4.2G doesn't have that much relation to its uncompressed size in memory, and there is also the memory required for coloring information. You can use -volume-merge to pull a single frame from the file so that you can load it in wb_view. Tim On Thu, Feb 23, 2017 at 4:24 PM, Jared P Zimmerman < [email protected]> wrote: > Thanks for the reply Tim, > > I’m having trouble getting them to open in wb_view. Running out of memory > even when I request 60G, though the files themselves are 4.2G. There’s > only 1 overlay in the red-yellow palette there and it’s overplayed over the > MNI152_T1_2mm template. I think it looks like multiple overlays because of > the parcel smoothing with the FSL structseg restricting the smoothing to > cortex, CSF, sub-cortex etc. > > I don’t think there are NaNs in my data because I used -fixnan 0 in my > call to -volume-math when demeaning and normalizing (see below). Is there > an easy way to check for NaNs with workbench, or do I have to read it into > Matlab? > > When I do -volume-smoothing with -fix-zeros I don’t get holes but I do get > dilation. Guessing I can fix this by including a brain mask with -roi > though. That may be sufficient for my needs right now, though I’m confused > why that doesn’t fix it with the parcel smoothing. > > I did use -exclude-outliers 3 3 in calling -volume-reduce to make my mean > and stdev images, which I thought would be important for removing > background that appear to be 0, but maybe I’m mistaken and this is the > source of my problem? > > > demeaning and variance normalization: > > wb_command -volume-math '(x-mean)/stdev’ \ > ${subID}_demean_${run}_hp2000_clean.nii.gz \ > -fixnan 0 \ > -var x ${run}_hp2000_clean.nii.gz \ > -var mean mean_${run}_hp2000_clean.nii.gz -repeat \ > -var stdev stdev_${run}_hp2000_clean.nii.gz -repeat > > Best, > Jared > > On Feb 23, 2017, at 4:17 PM, Timothy Coalson <[email protected]> wrote: > > The -fix-zeros options are basically designed to to dilation into regions > of zero values, so you probably shouldn't be using it. > > Maybe your input data has NaN values in it? That would lead to holes that > expand with the smoothing kernel (but should have cube shapes, so it > doesn't quite seem to fit your images). > > I also don't know what causes fslview to draw a "hole", so there might be > other possible causes - see what it looks like in wb_view. I'm also not > sure exactly what you are overlaying in those images, it almost looks like > the first one has more than one image overlaid in the red/yellow palette. > > Tim > > > On Thu, Feb 23, 2017 at 12:21 PM, Jared P Zimmerman < > [email protected]> wrote: > >> Hi HCP users, >> >> I’m having a problem smoothing volume data with workbench, where both >> -volume-smoothing and -volume-parcel-smoothing are creating holes in my >> data. See the attached screenshot for an example. I’ve tried both with >> and without -fix-zeros and this fixes the holes for -volume-smoothing, >> however it also seems to dilate my image so that it becomes ~120% the size >> it should be (image also attached). Parcel smoothing seems to fix the >> dilation problem, but it still introduces holes in the data regardless of >> whether the -fix-zero flag is used. For my parcels I’m using FSL’s >> MNI152_T1_2mm_structseg.nii.gz as a first pass. Any idea as to why these >> holes are popping up? >> >> >> <Screen Shot 2017-02-23 at 12.53.24 PM.png> >> >> >> <Screen Shot 2017-02-23 at 1.06.49 PM.png> >> >> >> thanks, >> Jared >> >> _______________________________________________ >> 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
