[ENH] Implement finite BOLD response- separate (FS) modeling

[tsalo]

Summary

Fixes None. References #197.

Here is an example FS design matrix.

List of changes proposed in this PR (pull-request)

• Add the --fir-delays argument to the CLI and all associated functions.
• Compute a separate beta series for each delay when using the FIR HRF model and the LSS estimator. This method is not applicable with the LSA estimator.

[tsalo]

@jdkent I still need to add tests and update documentation (hence the WIP label), but I wanted to get your thoughts on the method and code before doing all of that.

[codecov[bot]]

Codecov Report

Merging #204 into master will decrease coverage by 12.54%. The diff coverage is 46.66%.

@@             Coverage Diff             @@
##           master     #204       +/-   ##
===========================================
- Coverage   85.88%   73.33%   -12.55%     
===========================================
  Files          10       10               
  Lines         425      435       +10     
  Branches       47       50        +3     
===========================================
- Hits          365      319       -46     
- Misses         50      101       +51     
- Partials       10       15        +5

[codecov[bot]]

Codecov Report

Merging #204 into master will decrease coverage by 0.07%. The diff coverage is 88.23%.

@@            Coverage Diff             @@
##           master     #204      +/-   ##
==========================================
- Coverage   86.04%   85.97%   -0.08%     
==========================================
  Files          10       10              
  Lines         430      442      +12     
  Branches       48       52       +4     
==========================================
+ Hits          370      380      +10     
- Misses         50       52       +2     
  Partials       10       10

[jdkent]

Hi @tsalo, I'm currently working on reducing a backlog of duties, but I will get to this by next Tuesday at the latest.

[jdkent]

Hi @tsalo, sorry I haven't gotten to this yet, I've been working on simulations (using the LSA code you provided, thank you!) and I am presenting on that data next Tuesday. The process has been a little more taxing than I anticipated.

[tsalo]

I don't think FIR + LSA would work, because any overlapping timepoints across trials would have duplicate FIR regressors. I haven't tried it though, so maybe it's possible if the delay times are chosen carefully? Still, it's not something I've come across in the literature.

[tsalo]

I'm not sure why CI is currently breaking. My latest commit only removed the commented-out line of code and the commit before that was passing.

[jdkent]

This may be related to bids-standard/pybids#496, and it was fixed in niworkflows with poldracklab/niworkflows#405, but I'll see if there is something else going on (either instead or in addition to)

[jdkent]

Once #212 passes and I merge with master, this branch should be fixed once it is merged with master.

[jdkent]

P.S. I'm adding you as a collaborator, so when you have an approving review, you can merge in the pull request yourself. link

[jdkent]

Great, work, I think this is pretty much ready. For documentation I would like an entry in workflows.rst showing how this method is different from normal lss.

[tsalo]

Should the description go in the paragraph below the Participant Workflow figure? What about in the Beta Series page?

[jdkent]

Good question. I'm imagining that under Beta Series Workflow section header, we will have several subsections; one for plain lss, one for lss with fir, and in a separate pull request add one for lsa. I can do LSA so you can have a model of what I'm thinking.

As for the Beta Series page, we can add a sentence or two under conceptual background.

WDYT?

[tsalo]

Sounds good to me. Please let me know what you think of the changes.