Multi-shell, multiband acquisition issue

Hi all,

I’m hoping for some more general thoughts (not MRtrix specific). I am helping to set up a study and we are intending to collect a multiband multishell acquisition (b=0,500,1000,2000) at 2x2x2mm. What we are finding is that the acquisition is ok at b=0 but increasingly noisy as shell values. I know that noise is expected but the amount of noise is hard for me to wrap my head around. Any thoughts on what might be causing this? Any MRtrix that might help me track down whether there is some aspect of the protocol that is causing this.

Thanks in advance for any thoughts.

Note: All acquisitions to date are on Siemens systems and the noise in the screenshots below is representative of acquisitions on ACR phantoms and people and multiple sites. Note also that the non-b0 images have had the intensity rescaled manually in ITKSnap just to make anything visible (particularly the b=2000)

b=0
uci_b0

b=500
uci_b500

b=1000
uci_b1000

b=2000
uci_b2000

The amount of noise should be the same for each b-value. If you use the same intensity scaling for each image they should all look equally noisy.

However, due to the nature of diffusion weighting, in areas where diffusion is taking place, the amount of signal reduces with increasing b-value.

As a result, the ratio of the amount of signal and the amount of noise, or signal-to-noise ratio (SNR) decreases with increasing b-value. This is what you are observing in your images and this is entirely normal.

So it’s not a matter of ‘more noise’, but a matter of ‘less signal’.

Thanks @bjeurissen and your points are well taken.

To an extent I guess my underlying question is, is there a way to determine whether the low signal and what feels like high noise expected and reasonable versus an error in the protocol and to what extent it’s going to adversely affect our data. I recognize that the second half is much more nuanced (processing pipeline choices, research questions/hypotheses) but trying to problem solve on the front end rather than ending up in a bind after the data is collected.

As a quick aside: I’m not familiar with the ACR phantom, but from what I can see, it’s filled with doped water. This means it’ll have diffusion properties close to free water, with diffusivity much higher than tissue. You’d expect it to dephase rapidly as soon as the b-value goes up, what you’re seeing is entirely consistent with that. If you really want to check the sequence properly, you’ll want to use a diffusion-specific phantom, e.g. filled with agar gel. There are a few properly calibrated phantoms for diffusion that I’ve seen proposed, but I’m not sure there’s a standard as such.

Personally, I would look only at the SNR in the b=0 images as a rough indicator of quality – and even then, with a pinch of salt as the phantom’s T2 might differ substantially from the tissue (though I’m guessing it will have been designed to match T1 & T2, so that’s unlikely to be an issue in practice, just something to double check). There’s little value in looking at the diffusion weighted images unless you know what the diffusivity of the phantom actually is at that temperature…

1 Like