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.