What sequence parameters do you recommend for multishell-CSD?

Dear all,

I want to use the multishell CSD approach described in Jeurissen et al. 2014 to reconstruct structural connectomes in patients with epilepsy. What specific sequence parameters do you recommend for this technique, and how long does data acquisition take? Do you advise to acquire an anatomical scan additionally, or can I calculate partial tissue volumes directly?

Many thanks and kind regards

Eugenio Abela

Clinical Senior Lecturer
King’s College London
Institute of Psychiatry, Psychology and Neuroscience
Division of Neuroscience / Epilepsy

Hi @Eugenio_Abela,

What I (personally) typically advise to people in this case, is if they want to go for multi-shell data, to definitely not compromise on the highest b-value shell’s number of directions. The first lower b-value to add in as an extra, would best be something around b=1000. So a good yet simple protocol may be:

  • 6x b=0
  • 20x b=1000
  • 60x b=3000

That allows you to do the MSMT-CSD with 3 tissue types out of the box. In due time, we’ll add the single-shell 3-tissue CSD (SS3T-CSD) algorithm as well; which will then allow you to achieve 3-tissue CSD results just from single-shell data (so that would be the above, but not requiring the b=1000 shell!).

An anatomical scan can always come in handy if you want to do e.g. connectomics; but it’s not a strict requirement for any of the CSD steps per se. With the most recent addition of the dhollander algorithm for dwi2response response function selection, you do not need a T1 image any more to get very reliable WM, GM and CSF responses. That is if you’re working with mature human data though; so humans of 3 or more years old let’s say.

Don’t hesitate to ask for further information… we’re sorry it took such a long time to reply (everyone’s busy and may have overlooked your question :slight_smile:). I welcome other answers and advice on this question as well though… there’s not clear cut guidelines out there; and we have no simple anwers to what is “the most optimal” acquisition strategy.


Dear experts,

I would also like to get best possible protocol for structural connectivity and fixel-based analysis and would very appreciate your opinions/discussion on this. Namely:

We have Siemens Prisma 3T, 64-channel head coil.
We have multi-band DWI sequence at our disposal which could allow to acquire more directions per time interval. Do you have any experience with this sequence? What multiband factor do you recommend? Are there any pitfalls? What is the sensitivity of this sequence to movement? We have quite often facing problems with movement artifacts in patients in our studies.

For the unwarping, do you recommend to run reverse phase-encoding pair of all images to get reliable estimation of voxel value even in distorted areas? Or it is better to spend time on more directions?

What is the recommended trade-off between SNR, shells, number of directions per shell and voxel size?
We have, let’s say, 15-20 minutes for DWI protocol. What concrete recommendation for the protocol (number of directions per particular shells, b values of shell, full forward/reverse phase-encoding pairs on whole dataset or pairs only for b=0 images, voxel size) you would recommend?



1 Like