Diffusion protocol for neonatal imaging in limited acquisition time


For a longitudinal study investigating the effect of several therapies on the neurodevelopment of neonates we are looking for an optimal protocol to perform tractography and quantitative diffusion measurements on neonates. The available acquisition time would be around 6 minutes and multiband acquisition is possible.
I read on the dHCP protocol (20 b=0, 64 x b = 400, 88 x b=1000 and 128 x b=2600 s/mm2) yet with 20 minutes acquisition time (multiband 4) that would be too extensive for the allotted time.

We’re currently thinking to go for just the b=0 and b=1000 shells of the dHCP protocol. From @ThijsDhollander 's impressive presentation at ISMRM (see here) this seems to allow acceptable FODs in regions of crossing fibers. That would allow us to do connectivity and DTI analysis which would already be very valuable.

Would this approach be reasonable? Would an additional shell be useful for tractography? Besides (potential) benefits for tractography it would allow higher order models such as DKI or NODDI, yet will likely increase the scan time too much.

Looking forward to your insights.,
Kind regards,

Dear @tbilliet,

We are doing a multi-shell neonatal study and our approach is to acquire the two shells independently. I think the full acquisition time is around 7 minutes (2 shells + reverse phase encoding B0s). With this acquisition you can do connectivity + all multishell models. I hope this helps.

Best regards,


Hi Thibo,

Sorry to be rather late to reply. If you have a say in the acquisition protocol, I would advise against using a single shell protocol in the neonatal brain for anything beyond DTI. The tissue microstructure in the developing brain is in rapid transition, and personally I’m not convinced that it can be adequately modelled based on single-shell data. Certainly I would not pass on the opportunity to acquire multi-shell data if you can.

The dHCP protocol was designed by @jdtournier using the optimization procedure published here. I understand that the exact protocol may be too long for you, but there are two conclusions from that paper that can help. The first is that the cost function scales linearly with the number of directions, so you can rescale the protocol to fit within the available scan time. In your case, that would get you 128 * 6/20 = 38 directions on the outer shell, which I acknowledge might be a bit on the low side. The second is that the optimization can be run for any selected number of shells. The optimal 2-shell protocol consists of b=0, 400, and 1800 s/mm^2, with 12%, 29%, and 59% of the total no. directions respectively (Fig.6 and Table 3). Given your time constraints, my advice would be to use this approach. Assuming TR=4000ms you can acquire 90 directions in 6min, which would get you to 11 x b=0, 26 x b=400, and 53 x b=1800 s/mm^2. I would much prefer such protocol over a single-shell b=1000 scan.

Kind regards,



Hi Daan,

This is great information, thank you very much. I also feel that the opportunity of having multi-shell data should be taken, in order to investigate models beyond DTI.
I’ll try to have this implemented and tested. Thanks a lot.

Kind regards,

1 Like