Improve multi-echo acquisition recommendations

[tsalo]

Summary

This stems from @dowdlelt's comments on NeuroStars.

Given that Gowland & Powell (2007) says you need a last echo time ~1.5x the highest T2* value you want to correctly estimate and Peters et al. (2007) found gray matter T2* values at 3T around 66 ms (mean = 66 ms, SD = 1.4), it seems like we can provide more specific recommendations. Namely, if folks want to correctly estimate T2* for higher T2* values (e.g., +2 SDs above the mean), then it seems like they'll need a last echo around 103 ms.

I don't know how this fits with Dipasquale et al. (2017)- especially point 3 in their recommendations:

3. Acquiring the latest TE image such that most (∼75%) of the brain volume has not fully dephases, i.e. most voxels have signal above the noise floor;

[dowdlelt]

I'd be hesitant to push it out that far as a solid recommendation - thats heaps and heaps of time - I just wanted to push back against the idea of "I should get 3 echos as fast as possible." In addition, I would argue that optimal for estimate T2* is not equivilent to optimal for task/rest fMRI, or BOLD contrast. And I'd want to do more reading. But - to the point that maybe this information should be there, or there should be some citations for how far you should go out - I agree.

[tsalo]

Shouldn't we also be able to predict when a given voxel will fully dephase given S0 and T2*? Or at least based on some feature(s) of the protocol?

[dowdlelt]

fully might be a challenge - but, should be able to at least say there is no point in going past XXXms because there isn't usable signal. I don't think most people would want to push out to 200ms or what have you, but putting it in writing couldn't hurt.

[tsalo]

I was also thinking that low-echo-count sequences (e.g., three echoes) wouldn't be useable for large swathes of the brain if there aren't enough echoes at low enough echo times. For example, if we hypothetically had a three-echo protocol that covered that upper limit of 103 ms (e.g., TEs = 20, 60, 100 ms), then many voxels would only have 1 - 2 good echoes, right?

[handwerkerd]

I think this is a conflict between empirical T2* mapping and using variation in T2* weighting for calculating stuff, like we do in tedana. As noted, the longer echoes will have more dropout so we want to strike a balance between having echoes that are far enough apart to create measurable variation vs still having SNR. This is a great research question, but I don't have clear new guidance on this... expect maybe add something like I just wrote as an explanation.