Error in mrregister while running dwiintensitynorm

Dear all,

  1. I am working on a DTI data wher the problem is with acquisition the earlier data was acquired with bvalue =800 and 30 DW directions. Subsequent data after an intervention was acquired with bvalues=1000 and 56 DW directions. I am trying to analyze the apparent fiber density changes in thalamic structure after intervention using the fixel based analysis.
    So while running the 5th step of intensity normalization, I face an error which is as follows:
    mrregister: /usr/include/eigen3/unsupported/Eigen/src/MatrixFunctions/MatrixSquareRoot.h:127: void Eigen::internal::matrix_sqrt_quasi_triangular_diagonal(const MatrixType &, ResultType &) [MatrixType = Eigen::Matrix<double, 4, 4, 0, 4, 4>, ResultType = Eigen::Matrix<double, 4, 4, 0, 4, 4>]: Assertion `T(i,i) >= 0’ failed.
    I am wondering is it because of those differences? or any other reason.
  2. Is it possible to perform the apparent fiber density comparison between hemispheres using the thalamus as a seed region for connectome analysis to avoid the acquisition differences between scans?

Please let me know your suggestions.

Best regards,

The mrregister assert failure should not happen. What version of mrregister are you using? It would be helpful if you could have a look in dwiintensitynorm’s temporary directory. There should be subfolders fa and population_template. If you could share those (and the masks folder if you used any) with me, I could have a look at what’s going on.

As to your study design: If your groups (pre- and post- intervention) were acquired with different protocols (b-values, TE, …), any group difference in AFD or connectivity could and likely will be confounded by the difference in protocols. Unless you have a placebo intervention, I don’t see how you would analyse this data across groups.

Hi Maxpesch,
Thanks for your response. I am sharing the folders with the processing, dwi images and their masks. Please let me know How I can share the data with you.

The intervention was related with only single hemisphere so the another hemisphere will work as a control to compare the AFD in thalamus. and later we can compare the ratio of AFD which is independent of measure between pre and post condition.

You can share a link to the data via email or direct message or you could create a shared private OSF project (

What is the output of mrregister -version?

The mrregister version: mrregister 3.0_RC3
I will try to share it with email.


Dear Maxpietsch,
I tried to upload the data but it is too large almost 6 GB.Is it ok if I can send only the temp folder created in processing which contains masks, fa and population template subfolders.
Best Regards,

Yes, the masks, fa and population_template folders are sufficient. In case you want to use OSF, they have unlimited storage but a file size limit of 5GB so you’d need to create 4GB chunks for instance via
tar -czvf - /path/to/temporary_directory | split -b 4000M - upload.tar.gz

Hi Max,
I have uploaded on OSF and you can access it on this link

Please have a look and let me know your suggestions.
One more query, Is it necessary to perform intensity normalization for fixel based analysis or we can skip it ?


The mask for subject pre_20 contains only a small portion of the brain parenchyma. If this is not intentional, please fix this mask and rerun dwiintensitynorm. Otherwise, I’d create the fa template using all other subjects and try to register the problematic subject to that template using a custom mask for the template that roughly mimics the content of that subject’s mask.

Yes, you need to normalise your images. For more information, see here.

(dwiintensitynorm is not necessary for multi-shell data, see here).

Thanks for your response Max. I will check if I can use the coregistered T1 weighted mask for the 20 pre patient. If it is not resolve than I will get back to you.

Best regards,