pnet failed with the volume DataType

[albertleemon]

Dear Dr. Ma, Dr. Fan,

Hi. I would like to consult an issue regard with working with dataType='volume'.

1. python ~/work/images/pNet/fmripnet.py -c BEST_fmri_surf_subjs.toml --hpc
   failed.

data type is volume

dataType = 'Volume'

data format is NIFTI, which stores a 4D matrix

dataFormat = 'Volume (*.nii, *.nii.gz, *.mat)'

2. Then I went to the script folder to run the script,

(fmripnet) $ bash -x ~/BIDS/derivatives/pnet/Data_Input/Script/cluster_job_workflow.sh
++ date +%F-%H:%M:%S

• echo -e 'Start time : 2025-09-16-13:21:13\n'
  Start time : 2025-09-16-13:21:13

• /cluster/intel/oneapi-2023.1.0/intelpython/latest/bin/python ~/BIDS/derivatives/pnet/Data_Input/Script/cluster_job_workflow.py
  Intel(R) Extension for Scikit-learn* enabled (https://github.com/intel/scikit-learn-intelex)
  All setups are finished

Start to run

Start FN computation using PyTorch at 2025-09-16 13:22:43

Settings are loaded from folder Data_Input, FN_Computation and Cluster
Brain template is loaded from folder Data_Input
FN computation uses spatial-regularized non-negative matrix factorization method
Traceback (most recent call last):
File "~/BIDS/derivatives/pnet/Data_Input/Script/cluster_job_workflow.py", line 178, in
pnet.run_FN_computation_torch_cluster(dir_pnet_result)
File "/home/we46dav/.local/lib/python3.9/site-packages/pnet/Module/FN_Computation_torch.py", line 512, in run_FN_computation_torch_cluster
gFN = reshape_FN(gFN, dataType=dataType, Brain_Mask=Brain_Mask)
File "/home/we46dav/.local/lib/python3.9/site-packages/pnet/Module/Data_Input.py", line 1208, in reshape_FN
if FN.shape[0] != np.sum(Brain_Mask > 0):
TypeError: '>' not supported between instances of 'dict' and 'int'
++ date +%F-%H:%M:%S

• echo -e 'Finished time : 2025-09-16-13:23:49\n'
  Finished time : 2025-09-16-13:23:49

3. here is the test samples:
   $ mrinfo '~/BIDS/derivatives/xcp_d/sub-BEST001s01/func/sub-BEST001s01_task-rest_space-MNI152NLin2009cAsym_res-2_desc-denoised_bold.nii.gz'

---

Image name: "~/BIDS/derivatives/xcp_d/sub-BEST001s01/func/sub-BEST001s01_task-rest_space-MNI152NLin2009cAsym_res-2_desc-denoised_bold.nii.gz"

---

Dimensions: 97 x 115 x 97 x 450
Voxel size: 2 x 2 x 2 x 1.58
Data strides: [ 1 2 3 4 ]
Format: NIfTI-1.1 (GZip compressed)
Data type: 64 bit float (little endian)
Intensity scaling: offset = 0, multiplier = 1
Transform: 1 0 0 -96.5
0 1 0 -132.5
0 0 1 -78.5

what could be the reason for this error?
FN.shape[0] != np.sum(Brain_Mask > 0):

Do you have any suggestion how to fix this with my cohort?
Many thanks for your attention.

Best regards
Meng

[MLDataAnalytics]

The image mask and group FNs are in different spatial dimensions. Please verify that the image mask is registered to the same space as the preprocessed fMRI scans.

[albertleemon]

Thank you very much for your rapid feedback.
Following your suggestion, I checked the file at pNet/tree/main/src/pnet/Brain_Template/MNI_Volume/Brain_Mask.mat:

whos
Name Size Bytes Class Attributes

Brain_Mask 91x109x91 3610516 single

I noticed that the MNI152NLin2009cAsym space has the dimension 97 x 115 x 97. Could I simply replace the above Brain_Mask.mat with a version matching this resolution and then reinstall pNet? Or can I override the definition of Brain_Mask.mat in pNet/fmripnet.py?

As a feature request, would it be possible for pNet to provide two versions of Brain_Mask.mat—one with 91x109x91 and another with 97x115x97? Since more and more datasets are moving to the 97x115x97 resolution, I believe having both options readily available would make pNet more flexible and convenient for a broader range of users.

Thank you again for your kind support and for continuously improving this valuable tool.

[MLDataAnalytics]

Thanks for the update and your helpful suggestion. We've provided a script that can generate templates tailored to custom needs. You can find it at: src/pnet/cli/Generate_VolumetricTemplate.py

Please feel free to reach out if you have any further questions!