Multi Shell Multi Tissue CSD in neonates

Hi all,
I am working with pre-term and term born infant diffusion MRI data with HARDI acquisition ( 6 b=0, 30 b=700, 60 b=2800) . I am currently applying different microstructural models to those data (such as DKI, NODDI, FORECAST, MAPL and Multi-Shell Multi-Tissue CSD). As for all the models, default parameters seem to work well, with exception for MSMT-CSD. Indeed 55tgen algortihm seems to take too much time for some subjects without never terminating. Moreover, I have read it is not suitable for neonates due to poor WM/GM contrast.
Here you find attached an example outcome for 5ttgen algorithm applied to one subject. In order there are: csf, wm and gm tissue contribution map.


Instead, this is the 5tt.mif image

and finally, this is the WM fODFs displayed together with the tissue signals contribution map
image
Are they satisfactory?
I have at disposal 3dT1 and HARDI dMRI data for those infants, not the T2 image. Is there an alternative way with these data to compute MSMT-CSD?
thank you
Rosella

Hi Rosella,

As I’m sure you know, the issue is not with MSMT-CSD as such, but with the response function estimation that precedes it. 5ttgen relies on the expected T1w contrast in adult brain for WM/CGM/CSF segmentation, as well as an adult brain atlas to segment DGM, and is therefore suboptimal in neonatal brain imaging. You could use a different tissue segmentation approach adjusted to the T1w/T2w contrast in the developing brain, or use a manual segmentation, as input to RF estimation instead.

More fundamentally, the issue with the “standard” WM/GM/CSF MSMT-CSD is that WM in the developing brain is far from homogeneous. @maxpietsch has shown this very nicely in this paper, which I highly recommend, and proposed a response function estimation process for MSMT-CSD in neonates that can model the changing WM microstructure in the developing brain.

Regards,
Daan

3 Likes

Thank you very much. I will inspect the paper. Has it an mrtrix3 corresponding implementation?
thanks
Rosella

1 Like