Dwinormalise group in neonates with low FA values

Dear MRtrix Experts

To be able to do group comparison based on CSD data, I learned that one must do global intensity normalisation (on bias field corrected data). I tried to use the command dwinormalise group to achieve this. As far as I understand this command is based on using FA threshold values. Unfortunately, this command did not work for me (I got some nan matrices) and I think it’s because I’m working with neonates who tend to have very low FA values.

Therefore, I was thinking of creating our own script following this normalisation concept. As far as I understand it it’s about creating a group white matter map to calculate the correction factor. I already possess a white matter mask of all subjects. My plan would be to calculate the median intensity of the (bias corrected) B0 images of every subject over the white matter mask of the individual, take group mean of these values, and normalize the b0 images so that the B0 WM median value is the same. Do you think that this process would achieve something similar like the dwinormalise group?

I appreciate any kind of inputs!

Thank you very much!

Best regards,

Anna

1 Like

Hi Anna,

I assume you have single-shell data right?

I think the best you can do with this data is perform MSMT-CSD but with only two shells (There are several posts on the forum about this). Then you can use mtnormalise to perform intensity scaling normalization.

This approach works as it is for neonatal data, the only thing you have to consider is to modify -fa the parameter in the dwi2response dhollander algorithm. I hope this helps.

Best regards,

Manuel

Hi Manuel

Thanks so much for your quick reply! Exactly, I’m working with single-shell data (b=700).

This is a great idea to perfom MSMT-CSD using b=0 and b=700 and perform then mtnormalise. Thanks a lot. However, at the moment, I wanted to compare the SSST-CSD with the SSMT-CSD beta-version. But if you think there is no clever way to perform this normalisation on a single-shell dataset then I will definitely use your suggested approach - especially if you already know that it works for neonatal data.

Thanks again so much!
Best regards,
Anna

Hi,

I don’t know if you are aware of this work, but maybe it is interesting for you.

Best regards,

Manuel

1 Like

Dear Manuel

Thanks a lot. Yes, I’m exactly using this SS3T algorithm to compare it with the SSST-CSD. :slightly_smiling_face:

Best,
Anna

Just to clarify, you can run the multi-tissue CSD pipeline including mtnormalise on single-shell data if you model more than one tissue component.

For neonatal single-shell data, I’d recommend a (b0 and outer shell) CSF + WM decomposition. If you have a range of subject ages, I’d use the oldest group to derive the WM response. For the CSF response, I use masks in dwi2response that exclude areas with ADC > 3e-3 to not include areas affected by flow artefacts which would slightly alter the meaning of this free-fluid response.

1 Like

Dear Max

Thank you very much.

Yes, this is now clear to me that I can run the multi-tissue CSD pipeline with the mtnormalise on my single-shell dataset when modeling 2 tissue components (WM & CSF). But since I wanted to compare the SSST-based Tournier algorithm (modeling only WM) with the SS3T-based Dhollander algorithm I was trying to find a way how to normalise my data from the SSST-based algorithm for group comparison.

After this discussion, it seems to me that I should rather not do the compairson between these two algorithms but between the SS3T & SSMT algorithm (using CSF + WM decomposition).

Thanks again!
Anna