Contents of book

[tsalo]

Just a few initial thoughts:

1. Getting started
   1. Download data
      • Download fMRIPrepped multi-echo data using tedana datasets module. Probably just one run from the Cambridge dataset and something that's compatible with dual echo denoising (i.e., first echo <= 5ms).
      • Download dual-echo data as well.
   2. Software installation
   3. Recommended reading
      • Papers on multi-echo denoising
      • Paper on dual echo denoising
      • At least one paper on BOLD-based physiological noise
2. Theoretical Background
   1. A primer on MR physics
   2. Single- and multi-echo fMRI
      • With sequence diagrams
   3. Signal decay
      • The monoexponential model, T2* and S0
      • Information on other models of signal decay
   4. BOLD, non-BOLD, and TE-dependence with tedana
      • Building an intuitive understanding of T2* and S0 with gifs
3. Practical Resources
   1. Available multi-echo datasets
   2. Acquiring multi-echo data
   3. Processing multi-echo data
4. Analysis tutorials
   1. Optimal combination with t2smap
      • How does optimal combination improve coverage and increase SNR
   2. Multi-echo denoising with tedana
      • tedana's approach
      • Dependence metrics
      • Group-level analysis (e.g., applying transforms)
   3. Dual echo data
      • Just with some nilearn code. Should be easy enough to do first echo-regression.
   4. 3dMEPFM
   5. Cerebrovascular reactivity mapping
5. Other possible analyses
   1. Dynamic distortion correction (requires phase data)
6. Contributing tutorials
7. Glossary

[tsalo]

Now that I've added a bit to the repo, I'm tagging @ME-ICA/tedana-devs. Does anyone have any thoughts?

[jbteves]

My only modification I'd suggest is that we should probably implement the data fetchers instead of asking people to download, then it's an all-in-one solution.

[tsalo]

Are there any other major analyses/denoising pipelines we should include? Unfortunately, the links to posters we have in the docs appear to be dead, so I didn't find anything useable there. I hate how OHBM poster links never stay up for more than a year or two.

EDIT: Just thought of one. Volume-wise T2*/S0 (aka the FIT method).

[tsalo]

Could we maybe do something with CVR? We'd need breath-hold ME-EPI data, of course. Not sure if there's any public data with a breath hold task.

[eurunuela]

Could we maybe do something with CVR? We'd need breath-hold ME-EPI data, of course. Not sure if there's any public data with a breath hold task.

@smoia has an open breath-hold dataset. See here: https://openneuro.org/datasets/ds003192/versions/1.0.1

[tsalo]

🎉 That's awesome! We should definitely add EuskalIBUR to the tedana resources documentation.

[tsalo]

From today's call, another analysis/paper that may be worth including is Evans, Kundu, Horovitz, & Bandettini (2014), titled "Separating slow BOLD from non-BOLD baseline drifts using multi-echo fMRI".

[tsalo]

Maybe we could include dynamic distortion correction (or even just a blurb about the possibility), since @handwerkerd will be acquiring data that should work for it soon.

[tsalo]

What about a tutorial on manually correcting classifications? I'm sure people could use recommendations for identifying good or bad components.

[eurunuela]

What about a tutorial on manually correcting classifications? I'm sure people could use recommendations for identifying good or bad components.

Sounds like a great idea!

[tsalo]

Could someone summarize what we would do with Evans 2014? Like what that analysis tutorial would entail?

[dowdlelt]

My recollection of this would be an analysis in which high-pass filtering would not be done because it would remove task effects. From Evans, it was slowly changing contrast gratings, think, which were long enough that they would be buried in scanner low frequency drift. I believe @handwerkerd said there was more data like this.

I'd imagine a tutorial which shows that these effects can not be seen using a typical modeling approach, but with tedana, it is possible to separate out drift from slowing varying task effects. Maybe under the heading Feasibility of very slow task designs? Analyses would be simple, but its a strike against the "Can't have super long blocks" dogma in fMRI (assuming it works....)

[tsalo]

@dowdlelt That's super helpful, thanks! Not sure there will be any public data for that though. I mean, most folks stick with the rule against super long blocks in their datasets. If anyone knows of any datasets that would work for this, please let me know.

[jsheunis]

@tsalo Just want to drop in here and say how glad I am that I've found this repo! Been wanting to work on something like this for years, so thanks for doing it! I've got some content lying around that could be useful for the book, so I'll create issues for that.

[tsalo]

That would be great! Thanks @jsheunis!