Also, as came up yesterday in a separate thread, for working with the data efficiently, ft_read_cifti should be modified to respect the native data type of the CIFTI (i.e., single precision for floats) rather than loading as double precision. This is a steep penalty in terms of necessary RAM when attempting to load something like a dense connectome (or even just a dense timeseries).
cheers, -MH -- Michael Harms, Ph.D. ----------------------------------------------------------- Conte Center for the Neuroscience of Mental Disorders Washington University School of Medicine Department of Psychiatry, Box 8134 660 South Euclid Ave. Tel: 314-747-6173 St. Louis, MO 63110 Email: [email protected] From: Timothy Coalson <[email protected]<mailto:[email protected]>> Date: Tuesday, October 25, 2016 at 6:42 PM To: Benjamin Risk <[email protected]<mailto:[email protected]>> Cc: "Glasser, Matthew" <[email protected]<mailto:[email protected]>>, Michael Harms <[email protected]<mailto:[email protected]>>, "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>> Subject: Re: [HCP-Users] Problem with handling dense FC matrix The main issues with this are: 1) ciftiopen and ft_read_cifti return matrices of different dimension, with ft_read_cifti's being larger than what is actually in the file 2) ft_write_cifti puts those added NaNs *into the output file*, such that attempting to use them in wb_command with cifti that "should" match results in an error Tim On Tue, Oct 25, 2016 at 6:13 PM, Benjamin Risk <[email protected]<mailto:[email protected]>> wrote: Right, the NaNs are all in the location of the medial wall, so you can go back and forth between your representation in ciftiopen and ft_read_cifti using the code in the previous email. To see this, sum(nomedialwall==0) tabulate(subjLR1.brainstructure(nomedialwall==0)) which returns ans = 5572 Value Count Percent 1 2796 50.18% 2 2776 49.82% So there are 2796 vertices in the left cortex in the freesurfer 32k space corresponding to the medial wall and 2776 in the right. Does this seem correct to you? On Tue, Oct 25, 2016 at 7:04 PM, Glasser, Matthew <[email protected]<mailto:[email protected]>> wrote: The issue we have found is that it doesn’t properly represent the grayordinates space (it adds NANs). Peace, Matt. From: Benjamin Risk <[email protected]<mailto:[email protected]>> Date: Tuesday, October 25, 2016 at 5:59 PM To: Matt Glasser <[email protected]<mailto:[email protected]>> Cc: Timothy Coalson <[email protected]<mailto:[email protected]>>, "Harms, Michael" <[email protected]<mailto:[email protected]>>, "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>> Subject: Re: [HCP-Users] Problem with handling dense FC matrix In my opinion, there are some advantages to ft_read_cifti. I personally have not run into bugs with ft_read_cifti in my applications and am able to go back and forth between wb_view and matlab with vertices and voxels in the correct ordering. The main benefit of ft_read_cifti is that it reads in additional information, for example, reading the rs-fMRI data results in subject=num2str(100206); subjLR1 = ft_read_cifti(['~/HCP_rs/Subjects/' subject '/rfMRI_REST1_LR_Atlas_MSMAll_hp2000_clean.dtseries.nii']); subjLR1 subjLR1 = dimord: 'pos_time' time: [1x1200 double] hdr: [1x1 struct] unit: 'mm' brainstructure: [96854x1 double] brainstructurelabel: {1x21 cell} dim: [91 109 91] pos: [96854x3 double] transform: [4x4 double] dtseries: [96854x1200 double] and in particular .brainstructure and .brainstructurelabel can be used to extract subcortical regions easily. In contrast, ciftiopen returns a gifti object containing a single array >> subjLR1_ciftiopen = ciftiopen(['~/HCP_rs/Subjects/' subject >> '/rfMRI_REST1_LR_Atlas_MSMAll_hp2000_clean.dtseries.nii'],... '/home/samsi/Applications2/workbench/bin_linux64/wb_command'); subjLR1_ciftiopen subjLR1_ciftiopen = cdata: [91282x1200 single] I just made this example of how to get the data to match up: subject=num2str(100206); subjLR1 = ft_read_cifti(['~/HCP_rs/Subjects/' subject '/rfMRI_REST1_LR_Atlas_MSMAll_hp2000_clean.dtseries.nii']); subjLR1 % optional: subjLR1.dtseries = single(subjLR1.dtseries); subjLR1_ciftiopen = ciftiopen(['~/HCP_rs/Subjects/' subject '/rfMRI_REST1_LR_Atlas_MSMAll_hp2000_clean.dtseries.nii'],... '/home/samsi/Applications2/workbench/bin_linux64/wb_command'); subjLR1_ciftiopen nomedialwall = ~isnan(subjLR1.dtseries(:,1)); test = subjLR1.dtseries(nomedialwall,:); size(test) all(subjLR1_ciftiopen.cdata(:)==test(:)) norm(subjLR1_ciftiopen.cdata - test,'fro') If people have other thoughts on ft_read_cifti or concerns with anything above, maybe we can get a new thread going. Best, Ben On Tue, Oct 25, 2016 at 6:20 PM, Glasser, Matthew <[email protected]<mailto:[email protected]>> wrote: ciftiopen uses single precision. I remain puzzled where people are getting the idea to use ft_read_cifti from for MRI data. Peace, Matt. From: <[email protected]<mailto:[email protected]>> on behalf of Timothy Coalson <[email protected]<mailto:[email protected]>> Date: Tuesday, October 25, 2016 at 3:55 PM To: "Harms, Michael" <[email protected]<mailto:[email protected]>> Cc: "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>> Subject: Re: [HCP-Users] Problem with handling dense FC matrix The ciftiopen option of matlab cifti support should allow you to load the file, though I think it loads as double, and you'd need to convert it to single to reclaim some memory (which comes close to your memory+swap cap, as it needs the double and single versions allocated simultaneously in order to do the conversion). One possibility, if you don't really need a 90k x 90k network, is downsampling or parcellating the dense connectome, see http://www.humanconnectome.org/software/workbench-command.php?function=-cifti-parcellate . If you do need the 90k x 90k resolution, and can work in c++ instead of matlab, you could use CiftiLib for reading the file (can read single rows from disk, rather than the entire matrix into memory), or code up a new command for wb_command (same support for cifti, but also support for other file types, access to other algorithms). Tim On Tue, Oct 25, 2016 at 3:21 PM, Harms, Michael <[email protected]<mailto:[email protected]>> wrote: If you’re talking about some sort of community/modularity analysis, then, no, there aren’t any tools for doing that directly in workbench. Maybe Tim C. will have some suggestions. I believe that the dconn is stored as a single precision float. Have you tried using ‘ciftiopen’? I’m not sure if that preserves the single precision format of the data, or if it balloons by a factor of 2 due to conversion to matlab’s default double precision format. If you can keep it as single precision in matlab, you might be able to work with it provided Matlab has access to a lot of RAM. cheers, -MH -- Michael Harms, Ph.D. ----------------------------------------------------------- Conte Center for the Neuroscience of Mental Disorders Washington University School of Medicine Department of Psychiatry, Box 8134 660 South Euclid Ave.Tel: 314-747-6173<tel:314-747-6173> St. Louis, MO 63110Email: [email protected]<mailto:[email protected]> From: "GLOMB, KATHARINA" <[email protected]<mailto:[email protected]>> Date: Tuesday, October 25, 2016 at 10:35 AM To: Michael Harms <[email protected]<mailto:[email protected]>> Subject: Re: [HCP-Users] Problem with handling dense FC matrix Hi Michael, wow, thank you for this lightning speed reply! I want to run some network analysis on this matrix.I don't need the whole matrix, I would just need the unique entries (off-diagonal upper or lower triangular, squareform() in Matlab) - maybe that would help? Can workbench help me in any way to do this? Or do you have any other ideas? Thank you! Katharina On Tue, Oct 25, 2016 at 5:21 PM, Harms, Michael <[email protected]<mailto:[email protected]>> wrote: What are you trying to do with the file in workbench? Have you reviewed the tutorial, because it includes examples (and a scene to help you along) of using that file. Trying to load/work with that file in matlab is probably going to cause loads of pain. It is just too large to load the entire thing into matlab. cheers, -MH -- Michael Harms, Ph.D. ----------------------------------------------------------- Conte Center for the Neuroscience of Mental Disorders Washington University School of Medicine Department of Psychiatry, Box 8134 660 South Euclid Ave.Tel: 314-747-6173<tel:314-747-6173> St. Louis, MO 63110Email: [email protected]<mailto:[email protected]> From: <[email protected]<mailto:[email protected]>> on behalf of "GLOMB, KATHARINA" <[email protected]<mailto:[email protected]>> Date: Tuesday, October 25, 2016 at 10:15 AM To: "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>> Subject: [HCP-Users] Problem with handling dense FC matrix Hello HCP users, I am new to HCP so please excuse me if I'm asking something that should be obvious. I've tried to find a solution but haven't been successful so far. I am trying to use the file HCP_S900_820_rfMRI_MSMAll_groupPCA_d4500ROW_zcorr.dconn.nii which to my understanding should contain a full FC matrix that is an average of 820 subjects. I want to feed this matrix into a Matlab function. Of course, this is problematic because of the memory requirements. I have tried to read the file using ft_read_cifti from fieldtrip, but Matlab crashes at 112 GB of memory. So clearly, this is not the way. I know I am supposed to be able to use workbench for that, but can't make it work so far. Can someone point me in the right direction? Is what I'm trying to do impossible? I am using a cluster that can handle up to 132 GB of memory per cpu (RAM+swap) and runs Debian. Thank you!! Katharina -- PhD candidate<http://www.cns.upf.edu/katharina> Computational Neuroscience Group<http://www.cns.upf.edu/> Center for Brain and Cognition<http://cbc.upf.edu/> Universitat Pompeu Fabra<http://www.upf.edu/> Barcelona, Spain I am funded by the INDIREA<http://www.indirea.eu/> Marie Curie Initial Training Network. _______________________________________________ HCP-Users mailing list [email protected]<mailto:[email protected]> http://lists.humanconnectome.org/mailman/listinfo/hcp-users ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. -- PhD candidate<http://www.cns.upf.edu/katharina> Computational Neuroscience Group<http://www.cns.upf.edu/> Center for Brain and Cognition<http://cbc.upf.edu/> Universitat Pompeu Fabra<http://www.upf.edu/> Barcelona, Spain I am funded by the INDIREA<http://www.indirea.eu/> Marie Curie Initial Training Network. ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. _______________________________________________ HCP-Users mailing list [email protected]<mailto:[email protected]> http://lists.humanconnectome.org/mailman/listinfo/hcp-users _______________________________________________ HCP-Users mailing list [email protected]<mailto:[email protected]> http://lists.humanconnectome.org/mailman/listinfo/hcp-users ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. _______________________________________________ HCP-Users mailing list [email protected]<mailto:[email protected]> http://lists.humanconnectome.org/mailman/listinfo/hcp-users ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
