QSDR registration: how it works

[Alessandro De Maio]

Hi Frank, 

Amazing toolbox, here in our lab we're using it more and more for our projects.

Looking forward to possibly expand on some atlases too.

I have some more basic questions though on how QSDR is implemented in DSIstudio.

1) I read your papers on QSDR formulations, but what parameters does DSIstudio implement during linear and non linear registrations to the template? Is SPM used under the hood (like in your papers), and if yes, which parameters?

2) When I export the QA map after QSDR, its voxel size is the same as the original DWI volume. Is this ok or should the voxel size match the one of the template?  

3) If I wanted to test a different registration algorhythm: in order to obtain the same procedure as QSDR, is the following equivalent?

     -Estimate Linear+ NonLinear registrations between B0 and Template

     -Apply estimated warp to each DWI volume of the series

     -perform GQI signal reconstruction

Thank you so much in advance,

Best, 

Alex

[Frank Yeh]

> 1) I read your papers on QSDR formulations, but what parameters does DSIstudio implement during linear and non linear registrations to the template? Is SPM used under the hood (like in your papers), and if yes, which parameters?

The QSDR in DSI Studio now used its own implementation.
The linear registration uses mutual information similar to SPM.
The nonlinear registration is a diffeomorphic mapping method that uses
dual-modality (anisotropy and isotropy). The approach is different
from SPM.

>
> 2) When I export the QA map after QSDR, its voxel size is the same as the original DWI volume. Is this ok or should the voxel size match the one of the template?

It should be okay.

>
> 3) If I wanted to test a different registration algorhythm: in order to obtain the same procedure as QSDR, is the following equivalent?
> -Estimate Linear+ NonLinear registrations between B0 and Template

The linear registration uses the sum of all DWI and aligns it with the
isotropic template.
The nonlinear part needs to use both anisotropy and isotropy.

> -Apply estimated warp to each DWI volume of the series

Yes

> -perform GQI signal reconstruction

Yes, and make sure to include the Jacobian matrix as what is done in
QSDR paper (NTU90, Yeh, Neuroimage).

Best regards,
Frank

>
> Thank you so much in advance,
> Best,
> Alex
>

> --
> You received this message because you are subscribed to the Google Groups "DSI Studio" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to dsi-studio+...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/75f9bf2e-6930-406d-bb62-76383b615898n%40googlegroups.com.

[Alessandro De Maio]

Perfect,  Thank you!

So, just to be clearer:

If I applied the warp to each DWI volume of the series before the GQI signal reconstruction, should I also add the jacobian matrix once again (supposing the DWI images and their resulting reconstructed signal are both already aligned to the template)? 

I'm sorry if I didn't understand this aspect, but I thought the GQI reconstruction at this point would already be registered to the template.

And, why different sigma values for human vs animal (what kind, rodents?): 1.25 vs 0.6. Can you provide a reference or is it especially experimentally/ experience driven as you show in your documentation?

[Frank Yeh]

> So, just to be clearer:
> If I applied the warp to each DWI volume of the series before the GQI signal reconstruction, should I also add the jacobian matrix once again (supposing the DWI images and their resulting reconstructed signal are both already aligned to the template)?
> I'm sorry if I didn't understand this aspect, but I thought the GQI reconstruction at this point would already be registered to the template.

For registering, yes.
The jacobian matrix is only needed in QSDR part to rotate the b-table
so that the final orientation of ODF is correct.

>
> And, why different sigma values for human vs animal (what kind, rodents?): 1.25 vs 0.6. Can you provide a reference or is it especially experimentally/ experience driven as you show in your documentation?
>

0.6 is my experience for ex-vivo scans.
Sorry I don't have a reference, and the best setting also need to
tests different values (e.g. 0.5 0.6 0.7...etc).

Best,
Frank

> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/eb62d4bb-147b-4d0b-8762-0b605c75bac5n%40googlegroups.com.

[Alessandro De Maio]

Thank you so much for your support.

Regarding the sigma values: I understand now the possible variability.

Best, 

Alex