Hi,

Both the gifti toolbox and ciftiopen run into memory problems when loading
such large files, even with 128GB of RAM available, which is why I have
previously used fread to load "external binary gifti" files created with
-cifti-convert (which seems to work correctly).
I will try using -cifti-restrict-dense-map to create the 29k matrices and
then manually insert the medial wall mask in matlab. I want to visualize
maps using SurfStat, so the extra zeros are necessary.

Thanks again,

Estrid

On Tue, Mar 14, 2017 at 11:23 PM, Timothy Coalson <[email protected]> wrote:

> 1) No, a simple fread cannot correctly read any of our usual data file
> formats (the -to-gifti-ext conversion is not a "usual" data format, see #4)
> - they have headers on them, some of them are gzip compressed, and gifti is
> not a binary format anyway (it is XML, which in unusual circumstances can
> make use of a secondary binary file).  There is matlab code available
> specifically for the task of correctly reading in cifti and gifti data:
>
> https://wiki.humanconnectome.org/display/PublicData/HCP+
> Users+FAQ#HCPUsersFAQ-2.HowdoyougetCIFTIfilesintoMATLAB?
>
> The gifti toolbox it mentions is the recommended way to read gifti files
> into matlab.  Behind the scenes, ciftiopen and ciftisave use
> -cifti-convert, and then the gifti toolbox, to prevent any guessing and
> data scrambling on your part.
>
> 2) No, -cifti-restrict-dense-map will not expand the file to add the
> medial wall indices with zeroes.  What are you wanting to do with it that
> you need these extra zeroes in the matrix?  This can be done with
> -cifti-create-dense-from-template, but first you'll need to make a
> template 32k cifti file (only needs 1 map, not an entire square) with
> -cifti-create-dense-scalar or similar.
>
> 3) No, .surf.gii files cannot be used as ROIs, they are surface geometry
> (coordinate) files.  You can use the standard grayordinate space ROIs (in
> the pipelines repository, global/templates/91282_Greyordinates/*.shape.gii)
> or just make some all-1s metric files (.func.gii) with -metric-math, the
> result of -cifti-restrict-dense-map will be 29k with either (as long as you
> don't cross your hemispheres).
>
> 4) -cifti-convert does not create metric files (metric files are single
> surface, all vertices).  It abuses whichever output format is chosen in
> order to stuff the entire cifti matrix into it, to allow for tricks for
> exchanging cifti data with other tools that don't have cifti support.
>
> Tim
>
>
> On Tue, Mar 14, 2017 at 11:31 AM, Estrid Jakobsen <[email protected]>
> wrote:
>
>> Hi Tim,
>>
>> Thanks for your reply. I see my mistake was in assuming the output from
>> cifti-separate would be 32k x 32k and loading it in to matlab as such,
>> using *fread(fid,[32492 32492], 'float32')*. Since the file is actually
>> 32k x 91k it's no wonder the vertices aren't correct.
>>
>> Is it possible to get to the 32k x 32k matrices using
>> -cifti-restrict-dense-map (instead of 29k x 29k) and if so, can you advise
>> how? Can I simply use a surf.gii file as -left-roi and -right-roi?
>>
>> Presumably I can follow this with -cifti-convert -to-gifti-ext to convert
>> from dconn to metric and then load the matrix into matlab using fread as
>> before?
>>
>> Thanks again!
>>
>> Estrid
>>
>> On Mon, Mar 13, 2017 at 7:51 PM, Timothy Coalson <[email protected]> wrote:
>>
>>> That command should do what you intended, though it will give you two
>>> 32k x 91k files, not 32k x 32k.  Another possibility for doing this (which
>>> can also get you to the single-hemisphere 29k x 29k cifti file by running
>>> it twice) is -cifti-restrict-dense-map, which is also far less painful to
>>> load the results into wb_view, as it remains a dconn file (data loaded on
>>> demand, not in advance, knows not to show it as 91k maps in a combobox).
>>>
>>> I ran the exact command you quoted on the same file downloaded from
>>> ConnectomeDB, using workbench v1.1.1 linux64 (debian/ubuntu) as previously
>>> released on our website, and the resulting files I got look nothing like
>>> your capture (I am trying it again with v1.2.3 now, but it takes a while to
>>> run things on files this large - very little has changed in this command,
>>> so I expect the same result).  Did you do anything after the
>>> -cifti-separate that you didn't mention?  Which map(s) in the output file
>>> did you check?  What version of workbench and OS did you run it on, and how
>>> much memory did the machine have?
>>>
>>> Tim
>>>
>>>
>>> On Mon, Mar 13, 2017 at 10:13 AM, Estrid Jakobsen <[email protected]>
>>> wrote:
>>>
>>>> Hi folks,
>>>>
>>>> I have a question regarding extracting the left and right hemispheres
>>>> from .dconn cifti files. I've tried using wb_command -cifti-separate on the
>>>> 820-subject group average dense connectome (see command below), but when I
>>>> then visualize a connectivity vector from the resulting 32k x 32k matrix,
>>>> the vertices appear to be scrambled (see attached image). I'm guessing this
>>>> might have something to do with the command not being intended for
>>>> separating symmetrical matrices and perhaps only extracting correctly along
>>>> one dimension...?
>>>>
>>>> Is there a way to extract the left and right 32k matrices from the
>>>> dense connectome, or would I have to go back and create them myself from
>>>> the PCA Eigenmaps?
>>>>
>>>> Thanks in advance!
>>>>
>>>> Estrid
>>>>
>>>> wb_command -cifti-separate 
>>>> HCP_S900_820_rfMRI_MSMAll_groupPCA_d4500ROW_zcorr.dconn.nii
>>>> ROW -metric CORTEX_RIGHT HCP_S900_820_rfMRI_MSMAll_grou
>>>> pPCA_d4500ROW_zcorr_RIGHT.func.gii -metric CORTEX_LEFT
>>>> HCP_S900_820_rfMRI_MSMAll_groupPCA_d4500ROW_zcorr_LEFT.func.gii
>>>>
>>>>
>>>> ****************
>>>> Estrid Jakobsen
>>>>
>>>> Postdoctoral Fellow
>>>> Cognitive Neurosciences
>>>> Montreal Neurological Institute and Hospital, McGill University
>>>>
>>>> Max Planck Research Group Neuroanatomy and Connectivity
>>>> Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig
>>>>
>>>> _______________________________________________
>>>> HCP-Users mailing list
>>>> [email protected]
>>>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>>>>
>>>
>>>
>>
>>
>> --
>> ****************
>> Estrid Jakobsen
>>
>> Postdoctoral Fellow
>> Cognitive Neurosciences
>> Montreal Neurological Institute and Hospital, McGill University
>>
>> Max Planck Research Group Neuroanatomy and Connectivity
>> Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig
>>
>
>


-- 
****************
Estrid Jakobsen

Postdoctoral Fellow
Cognitive Neurosciences
Montreal Neurological Institute and Hospital, McGill University

Max Planck Research Group Neuroanatomy and Connectivity
Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig

_______________________________________________
HCP-Users mailing list
[email protected]
http://lists.humanconnectome.org/mailman/listinfo/hcp-users

Reply via email to