ds006072 dataset cifti parcelation issue

[Giannis Fotis]

Hi everyone,

I am working on this dataset for my master's thesis, and I am trying to apply a parcelation on the cifti files in the NON_BIDS/ciftis/ folder using the hcp library in python. 

For some reason, it seems that all the parcelation maps have 91282 grayordinates, while the ciftis in my dataset have 91206 . Here's the exact error I'm getting:
IndexError: boolean index did not match indexed array along dimension 1; dimension is 91206 but corresponding boolean dimension is 91282

I havent managed to figure this out for a couple of days now... Can anyone help? Thanks!

Giannis

[Tim Coalson]

You can run wb_command -file-information -no-map-info on your cifti files and see how many indices are being used in each structure.  The standard 91k grayordinate space is 91282, and looks like this:

$ wb_command -file-information -no-map-info global/templates/91282_Greyordinates/91282_Greyordinates.dscalar.nii

<SNIP>

Number of Rows:                 91282
Number of Columns:              1
Volume Dim[0]:                  91
Volume Dim[1]:                  109
Volume Dim[2]:                  91
Palette Type:                   Map (Unique for each map)
CIFTI Dim[0]:                   1
CIFTI Dim[1]:                   91282
ALONG_ROW map type:             SCALARS
ALONG_COLUMN map type:          BRAIN_MODELS
    Has Volume Data:            true
    Volume Dims:                91,109,91
    Volume Space:               -2,0,0,90;0,2,0,-126;0,0,2,-72
    CortexLeft:                 29696 out of 32492 vertices
    CortexRight:                29716 out of 32492 vertices
    AccumbensLeft:              135 voxels
    AccumbensRight:             140 voxels
    AmygdalaLeft:               315 voxels
    AmygdalaRight:              332 voxels
    BrainStem:                  3472 voxels
    CaudateLeft:                728 voxels
    CaudateRight:               755 voxels
    CerebellumLeft:             8709 voxels
    CerebellumRight:            9144 voxels
    DiencephalonVentralLeft:    706 voxels
    DiencephalonVentralRight:   712 voxels
    HippocampusLeft:            764 voxels
    HippocampusRight:           795 voxels
    PallidumLeft:               297 voxels
    PallidumRight:              260 voxels
    PutamenLeft:                1060 voxels
    PutamenRight:               1010 voxels
    ThalamusLeft:               1288 voxels
    ThalamusRight:              1248 voxels

I have never seen 91206 before, but maybe someone intersected the medial wall ROIs?  nibabel should expose the information of exactly what vertices/voxels are used in the embedded XML extension, but I'm not familiar with its API.  There is also wb_command -cifti-export-dense-mapping if you want to inspect it via text files.

Tim

--
You received this message because you are subscribed to the Google Groups "HCP-Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hcp-users+...@humanconnectome.org.
To view this discussion visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/de975df4-0ae5-4663-a136-6d37d884192cn%40humanconnectome.org.

[Giannis Fotis]

Dear Tim,

Thank you for the tip! Here's what I get when I run this command on my time series data:

Type:                           CIFTI - Dense Data Series
Structure:                      CortexLeft CortexRight
Metadata Start:                 -----------

[...]

Data Size:                      543.59 Megabytes
Maps to Surface:                true
Maps to Volume:                 true
Maps with LabelTable:           false
Maps with Palette:              true
All Map Palettes Equal:         true
Map Interval Units:             NIFTI_UNITS_SEC
Map Interval Start:             0.000
Map Interval Step:              1.761
Number of Rows:                 91206
Number of Columns:              1490

Volume Dim[0]:                  91
Volume Dim[1]:                  109
Volume Dim[2]:                  91

Palette Type:                   File (One for all maps)
CIFTI Dim[0]:                   1490
CIFTI Dim[1]:                   91206
ALONG_ROW map type:             SERIES
    Start:                      0.000
    Step:                       1.761
    Units:                      Seconds

ALONG_COLUMN map type:          BRAIN_MODELS
    Has Volume Data:            true
    Volume Dims:                91,109,91

    Volume Space:               2,0,0,-90;0,2,0,-126;0,0,2,-72

    CortexLeft:                 29696 out of 32492 vertices
    CortexRight:                29716 out of 32492 vertices

    AccumbensLeft:              84 voxels
    AccumbensRight:             111 voxels
    AmygdalaLeft:               256 voxels
    AmygdalaRight:              291 voxels
    BrainStem:                  3581 voxels
    CaudateLeft:                621 voxels
    CaudateRight:               618 voxels
    CerebellumLeft:             9316 voxels
    CerebellumRight:            9260 voxels
    DiencephalonVentralLeft:    734 voxels
    DiencephalonVentralRight:   724 voxels
    HippocampusLeft:            705 voxels
    HippocampusRight:           688 voxels
    PallidumLeft:               335 voxels
    PallidumRight:              332 voxels
    PutamenLeft:                772 voxels
    PutamenRight:               805 voxels
    ThalamusLeft:               1312 voxels
    ThalamusRight:              1249 voxels

Someone suggested to me that I "use wb_command -cifti-create-dense-from-template 1 to resample one to the other".  Does that sound right to you? And should I resample me time series data or the parcelation maps? Many thanks!

Best,
Giannis

[Glasser, Matthew]

Your subcortical voxels do not match ours.  How was this file generated?

    ThalamusRight:              1248 voxels

Matt.

To view this discussion visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/c4ef53f5-d3f6-4ddd-803f-77297af12been%40humanconnectome.org.

 

---

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.

[Giannis Fotis]

Dear Matt,

These files are the already preprocessed fMRI recordings in this openneuro.org dataset, found in the NON_BIDS/ciftis/ folder. They are were published together with the paper: 
Siegel, J.S., Subramanian, S., Perry, D. et al. Psilocybin desynchronizes the human brain. Nature 632, 131–138 (2024). https://doi.org/10.1038/s41586-024-07624-5 

This is pretty much the first time I work on fMRI data, so I probably can't say much more that would be accurate, and thats why I'm refering you to the source material... I hope this is useful. If there is anything specific I can search for you I'd be more than glad to. Thank you for your time.

Best,

Giannis

[Tim Coalson]

Given that you have more brainstem and cerebellum voxels, and fewer voxels in many other cortical structures, I wonder whether the subcortical data in your file is actually in MNI space or not.  If it isn't, you will need to use a transform in -cifti-resample in order to preserve as much of the data as possible (-cifti-resample doesn't allow data to cross between subcortical structures, so if they aren't aligned, it will clip off data).

-cifti-create-dense-from-template would be reasonable when the subcortical definitions agree.  Here they don't, so -cifti-resample is the better choice (particularly because of its -volume-predilate option).  I also notice that your voxel data uses LPI order (the X spacing is positive), while the standard 91k grayordinates use RPI order (negative X spacing), so -cifti-create-dense-from-template would reject the file anyway.

Tim

[Giannis Fotis]

Dear Tim,

I think the data is in MNI space, but I'm not sure. Here's what the article that these data were published with say:

"Resampling in MNI152 2 mm³ atlas space was accomplished for all echoes in one step combining (i) motion correction; (ii) distortion correction; (iii) gain field correction; (iv) linear registration of average volumes across visits; and (v) non-linear MNI152 atlas registration via the FSL FNIRT^52,53,54,55. The multi-echo data in MNI152 space then were combined using the weighted summation approach (as described in Posse et al. equations 6 and 7⁵⁶).

After cross-modal registration, the data underwent several pre-processing steps. These included: FreeSurfer segmentation for tissue-based regression, elimination of signals with false variance, temporal filtering to include 0.009–0.08 Hz bands, and frame censoring⁵⁷. False signals included: six parameters obtained by rigid body correction of head motion, extra-axial noise, white matter, and ventricles. Frame censoring was performed on all resting state fMRI data at FD > 0.30 mm⁵⁵."

Also, on a more general note, when I first open the cifti data files I get a [#time_steps]x[91206 grayordinates matrix]. Are these grayordinates ordered spatially? I.e. would just "cropping" the parcellation maps as below maintain the spatial continuity, or would it completely mess up the parcellation? I'm sure this is at least suboptimal, but was wondering if I could make-do with this for now until I figure out the parcelation to avoid this bottleneck... 

parcellation_indices_91282 = hcp.yeo17.map_all     # load map

parcellation_indices_91206 = parcellation_indices_91282[:X.shape[1]]     #trim to match data grayordinates

roi_ids = np.unique(parcellation_indices_91206)
roi_ids = roi_ids[roi_ids != 0]  # remove unassigned (ID 0)
num_rois = len(roi_ids)

print(f"Number of ROIs (parcels) found: {num_rois}") #check if all parcels still there

Many thanks!
Giannis

[Glasser, Matthew]

For better or for worse, this group does not use the HCP pipelines or recommended HCP-Style processing.  Instead, they have some in-house code they use.  It is good that they process the data on the surface; however, it seems they are making incompatible CIFTI files.  I reached out to one of the authors to ask  about that, but did not hear back.

Matt.

 

From: Giannis Fotis <gfw...@gmail.com>
Reply-To: "hcp-...@humanconnectome.org" <hcp-...@humanconnectome.org>
Date: Friday, November 7, 2025 at 2:59 AM
To: HCP-Users <hcp-...@humanconnectome.org>
Cc: "tim.c...@gmail.com" <tim.c...@gmail.com>, HCP-Users <hcp-...@humanconnectome.org>, Giannis Fotis <gfw...@gmail.com>
Subject: Re: [hcp-users] ds006072 dataset cifti parcelation issue

 

Dear Tim,

To view this discussion visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/dba2ab81-25d9-4992-aaf9-920109109af1n%40humanconnectome.org.

[Tim Coalson]

Since you have different numbers of voxels used in each structure (some more, some less), not only would truncation not get the spatial information to agree, it would move some data from some subcortical structures to others.  Cifti stores a list of voxels/vertices that are members of each structure, in the order of their data index.

If you display the 91206 file in wb_view on top of an MNI average (and the dlabel file) and toggle it on and off, you can check whether the subcortical data aligns with the expected structures.  If you don't need the parcellation of the subcortical data, you could use a version of the dlabel file without subcortical information.

If the dlabel you are using is our HCP MMP 1.0, the subcortical labels are just the structure labels.  You could pull the information of their subcortical definitions from their data file with wb_command -cifti-separate with the -volume-all -label suboption, and then combine that with the cortical information from the dlabel file (-cifti-separate to get the surface label information, or start from the version of the HCP MMP 1.0 that doesn't include subcortical, -cifti-create-label to either make the final dlabel from gifti+volume, or make a subcortical-only dlabel and use -cifti-merge-dense to put it together with the surface-only HCP MMP 1.0 dlabel).

Tim