Hi, A couple extensions to Tim’s recipe. PALM has a “transposedata” option, so you can always transpose at the PALM stage if you prefer to not explicitly create a transposed CIFTI file.
PALM can indeed accept CIFTI files “as is”, *if* you want to do permutation on the max statistic across grayordinates. If you want to do TFCE, you currently do indeed need to separate the CIFTI first. We have an example of that particular approach in the Task fMRI practical of the HCP Course. (Jenn can hopefully provide an update on when we anticipate being able to get the materials from the latest course online). Similar to Tim, I’m not exactly sure what you’re intending to permute. Are you going to compute a dense connectome *for each subject*? Then you could take the difference between the “A” and “B” maps for each subject, and use those as input to PALM to test whether that difference is consistently different from 0 across subjects (using sign-flippings). As Tim suggests, it would likely be much easier to compute a parcellated connectome for each subject instead. 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: mha...@wustl.edu From: <hcp-users-boun...@humanconnectome.org<mailto:hcp-users-boun...@humanconnectome.org>> on behalf of Timothy Coalson <tsc...@mst.edu<mailto:tsc...@mst.edu>> Date: Monday, July 17, 2017 at 7:52 PM To: "Regner, Michael" <michael.reg...@ucdenver.edu<mailto:michael.reg...@ucdenver.edu>> Cc: "hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>" <hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>> Subject: Re: [HCP-Users] Statistical comparison of whole brain (surface "voxels" + subcortical / cerebellar voxels) connectivity between two explicitly defined voxels On Mon, Jul 17, 2017 at 6:21 PM, Regner, Michael <michael.reg...@ucdenver.edu<mailto:michael.reg...@ucdenver.edu>> wrote: Hello Matt and HCP Community, Thank you for the helpful e-mail and words of encouragement. We are very encouraged by the data / results we can view in the Connectome Workbench; however, statistical analysis has proved challenging. Just to reiterate what we are attempting to do: we are hoping to compare whole-brain (all ~91k brainordinates) connectivity between two predefined brainordinates to determine the areas of the brain in which there is a statistically significant difference in connectivity. We intend to eventually extend this analysis to compare two different ROIs (sets or masks of brainordinates); however, for simplicity’s sake (and to prove to our PI that we can do this) we would like to perform this comparison first between two voxels / vertices (henceforth brainordinates “A” and “B”). The general pipeline (using the 33 GB resting state dense connectome CIFTI file as our starting point) is as follows step-by-step: 1. Reduce the resting state dense connectivity CIFT file (91k x 91k) into two separate CIFTI files, which are NOT dense. They would in fact still be dense, just on one dimension of the matrix, rather than both. Read this if you haven't yet (and let me know if it doesn't explain it well enough): http://www.humanconnectome.org/software/workbench-command/-cifti-help These CIFTI files (“A” and “B”) would contain the whole-brain connectivity data from two a priori specified brainordinates (“A” and “B”). By size, these intuitively should be around 91k in number of brainordinates. We are not exactly sure how to best achieve this in wb_command… would “CIFTI-parcellate” do the trick? Honestly, you might as well skip straight to using ROIs, especially if they are from an existing parcellation, as it may actually be easier (and more closely relates to the commands you would need to use). When you want to compare connectivity of two ROIs, it is highly advisable to get the ROI's average timeseries first, and then do a fresh correlation to the timeseries - this removes a large amount of noise-based variance from the denominator (among other things). So, if you are in fact using an existing parcellation in cifti dlabel format, you would want to do -cifti-parcellate on a dtseries file, and then you can do -cifti-cross-correlation with the dtseries file and this new parcellated file (ptseries). If you want to view the per-parcel dense maps in workbench, you should have the ptseries file as the first input, and name the output ending in ".pdconn.nii". To arrange them similarly to a dscalar file, which is probably what PALM expects, you can use -cifti-transpose to turn it into a .dpconn.nii file. If you want to use arbitrary ROI files (which can overlap), then you instead need to use -cifti-average-roi-correlation on the dtseries file. Note that this also performs a fisher small-z transform on the correlation (as it was written to also average across results from multiple files). Or, can we read the CIFTI data into MATLAB and simply select an individual row corresponding to the brainordinate of interest? If you have any thoughts as to the most direct way to do this, it would be appreciated. 2. Apply “CIFTI-separate” to both CIFTI files A and B. This should result in two 2D surface maps (for the left and right hemispheres) in NII format for both brainordinates “A” and “B”, along with their corresponding GIFTI files to register them in space. I think PALM can accept cifti files as-is, so you may not want to do this. You may need to either extract or concatenate the cifti maps of interest to satisfy PALM's input syntax, which you can do with -cifti-merge. The single-hemisphere-only surface format is GIFTI, ending in .gii (.func.gii for data values). The surface geometry is not inside the CIFTI files (it is in .surf.gii files). See http://www.humanconnectome.org/software/workbench-command/-gifti-help . A single 3D volume map of the subcortical / cerebellar brainordinates should also result from this for both brainordinates “A” and “B”. So, we end up with 6 NIFTI maps and 4 GIFTI files. For each ROI, two .func.gii files and one volume .nii.gz file. The surface files exist elsewhere in the subject directory (and could be useful for the ROI averaging or parcellating step, to account for differences in vertex sizes). 3. Apply FSL PALM with TFCE correction (syntax similar to randomize) to each of the three pairs of maps produced in #2 (left cortex, right cortex, and subcortical) to compare “A” versus “B.” We will need to include the midthickness file as an argument for the 2D surface data, in order for PALM to appropriately correct for volume. This should result in three “A > B” contrast maps (left cortex, right cortex, subcortical). In order to do permutation-based statistics (which using TFCE requires), you need to have many things to permute, otherwise you can't build a distribution to compare to the real maps. I don't know what you would be permuting here, with only 2 maps. If the difference in correlation values is due only to chance, how would you tell? 4. Use wb_command to reconstruct whole brain CIFTI files from the three contrast maps produced in #3, which can then be viewed in Connectome Workbench for inspection. We hope that this will provide areas of the brain in which connectivity between brainordinate “A” and “B” are significantly different, with the appropriate correction for multiple comparisons in #3. Any comments / suggestions / heckling would be sincerely appreciated. Thank you in advance! Mike Michael F. Regner, M.D. Departments of Radiology and Bioengineering University of Colorado – Denver E-mail: michael.reg...@ucdenver.edu<mailto:michael.reg...@ucdenver.edu> From: Glasser, Matthew [mailto:glass...@wustl.edu<mailto:glass...@wustl.edu>] Sent: Monday, July 10, 2017 8:15 PM To: Regner, Michael <michael.reg...@ucdenver.edu<mailto:michael.reg...@ucdenver.edu>>; hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org> Subject: Re: [HCP-Users] Statistical comparison of whole brain (surface "voxels" + subcortical / cerebellar voxels) connectivity between two explicitly defined voxels Hi Michael, I’m happy to hear you are making good progress with CIFTI and Connectome Workbench. You can indeed use the PALM software to do statistical inference on CIFTI data, and that is the tool we recommend. Peace, Matt. From: <hcp-users-boun...@humanconnectome.org<mailto:hcp-users-boun...@humanconnectome.org>> on behalf of "Regner, Michael" <michael.reg...@ucdenver.edu<mailto:michael.reg...@ucdenver.edu>> Date: Monday, July 10, 2017 at 2:40 AM To: "hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>" <hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>> Subject: [HCP-Users] Statistical comparison of whole brain (surface "voxels" + subcortical / cerebellar voxels) connectivity between two explicitly defined voxels Dear HCP Community, I am relatively new to the HCP data and Connectome Workbench. Our neuroradiology laboratory at the University of Colorado is beginning to use it. It has already proved to be extremely helpful in aiding the interpretation of existing resting state results. My question is: does the HCP Connectome Workbench provide an internal mechanism for the generation of statistical comparisons / contrasts within or between CIFTI files? I am exploring the dense resting state connectivity data. My goal is to construct a contrast map for statistical inference. I would like to compare whole brain (surface "voxels" + subcortical / cerebellar voxels) connectivity between two explicitly defined voxels within the dense resting state connectivity CIFTI file (or construct new whole-brain connectivity map CIFTI files for two explicitly defined voxels, and statistically compare these two maps). So, the desired resultant map / CIFTI file would illustrate the statistically significant difference in connectivity between these two points across the whole brain. Is there a way to do this with “-cifti-math” or the correlation command? As a corollary, can this be performed in FSL / PALM? This is probably a very basic question, but coming from years of SPM8 / SPM12 experience it is not very intuitive… Any comments, suggestions, or pointers would be much appreciated! Thank you, Michael F. Regner, M.D. Departments of Radiology and Bioengineering University of Colorado – Denver E-mail: michael.reg...@ucdenver.edu<mailto:michael.reg...@ucdenver.edu> _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org<mailto:HCP-Users@humanconnectome.org> http://lists.humanconnectome.org/mailman/listinfo/hcp-users _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org<mailto:HCP-Users@humanconnectome.org> http://lists.humanconnectome.org/mailman/listinfo/hcp-users _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org<mailto:HCP-Users@humanconnectome.org> 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 HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users
