What modifications can be done to optimally combine ASL signal instead of BOLD ?

[sboylan]

Hello collaborators,

I have been looking around and trying to figure out how do it, but didn't find a simple way.
My Issue would be the continuation of issue #777 "Add minimally-preprocessed Cohen dataset to datasets module"

Summary

In Cohen's sequence, we use ASL to extract CMRO2 from BOLD data.
ASL is a method to measure the Cerebral Blood FLow (ASL), where we tag the blood in the neck on label volumes, to contrast it with control volumes to get the perfusion dataset cf :https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5794066/
ASL time-series has low tSNR, and currently, we only use one echo.
I am sure we can increase the tSNR of ASL data from such sequences if we can optimally combine the ASL data with all our echos, but I don't know which part of the code would be easier to change (decision trees, ICA selection... )

Additional Detail

There are several ways to compute the perfusion datasets : BOLD subtraction (label - control) or filtering BOLD data (high pass filter). I don't know if the BOLD ICA would be usable in such data.

Next Steps

[CesarCaballeroGaudes]

Hello @sboylan, why do you want to apply multi-echo with ASL data? AFAIK, in ASL one wants to use the shortest TE (minimize) to get the perfusion signal. Can you explain your case better?

[tsalo]

I am not familiar with multi-echo ASL generally. The only paper I'm aware of that discusses it is Mahroo et al. (2021). ExploreASL does support multi-echo ASL for blood-brain barrier perfusion, but I don't know much about it.

If you could share more details about how the ASL (control, label, M0) signal curves would look and what, if any, differential effect you'd expect to see between them, that would be very helpful for understanding whether tedana could help.

As an aside, I have recently started maintaining ASLPrep, so multi-echo ASL is particularly interesting to me.

[CesarCaballeroGaudes]

Certainly multi-echo ASL for BBB perfusion is very relevant!!

[sboylan]

Hello guys, thanks a lot for your inputs.
We are using ASL to get the CBF that we can input in Davis equation ΔBOLD / BOLD0 = M * (1−(CMRO2 / CMRO20)^β * (CBV / CBV0) * (CBF / CBF0)^−β), to extract the CMRO2. We use it to explore O2 consumption in "real time".
We use multi-echos for the BOLD, so the multi-echo ASL is bonus, the data of the three other TEs are discarding at the moment.
@CesarCaballeroGaudes The ASL signal is very noisy, with a mean close to 0. So I thought collecting all the information held by the other TEs could de-noise the signal.
@tsalo I will have a look at ExploreASL, it might help.
We don't have the M0, as we calculate a ratio, We only calculate M with a breath-hold paradigm.
I will have a look at the signal curve (on one TR right ?). I must say, I don't know where to start ^^.

We expect the CBF to change during activation, with a delay obviously. But I don't know any more.
I'll check what BBB is to see if it is relevant in our analysis

[CesarCaballeroGaudes]

Yes, I am aware of that sequence. @smoia and I have already collaborated with the developers of the multi-echo ASL-BOLD sequence that you mentioned, and a joint paper is under review.
Similar to you, we are also working on developing new ASL + multi-echo BOLD for calibrated fMRI purposes. However, I don't see how combining multiple-echoes can help with the ASL signal. Then, what would the difference be with respect with the multi-echo BOLD? To the best of my understanding, the first echo is used for perfusion, whereas the later echoes are combined for BOLD contrast. The multi-TE ASL sequence proposed in Mahroo et al. (2021) is different. To start with, the sequence used in Cohen et al. uses a 2D multi-echo EPI readout with SMS acceleration, whereas the sequence used in Mahroo et al. uses 3D GRASE.

[sboylan]

Thanks for the input, there are probably some aspects that are not clear to me.
Wouldn't the 2-4 echos have some information on the perfusion ?
If the signal is noisy, wouldn't it be interesting to use different echos to get a better idea of the value of the decay slope of the signal to get a better measurement of the ASL signal which would give us a better measurement of CBF ?
I am probably wrong, but I don't fully understand why I guess.