Functional alignment vs. cross-correlation

[tsalo]

Would it make sense to try functional alignment as an alternative for cross-correlation?

@mattcieslak used functional alignment on time series in grad school, and he proposed using a tool like scikit-fda to estimate the sLFO signal (e.g., https://fda.readthedocs.io/en/stable/auto_examples/plot_shift_registration.html#sphx-glr-auto-examples-plot-shift-registration-py) and the dynamic lead/lag relationship over time for each voxel. @bbfrederick have you used tools like this in the past?

[bbfrederick]

I haven't - this is the first I've heard of it! I'm intrigued, and will look into it.

I do have a new time delay estimation method built into rapidtide based on the ratio of regression coefficients of the main regressor and its time derivative, which is used for the new delay refinement step. I'm still writing it up as a paper.

[tsalo]

@mattcieslak will be OOO until after ISMRM I believe, but maybe we can chat about this once he gets back?

Do you have any tools to simulate fMRI data affected by sLFOs? It might be useful to try out scikit-fda on.

[mattcieslak]

I was reminded of the srvf approach after the nipreps hackathon and would be really interested to see how karcher means and maybe some fisher rao distances/clustering work on BOLD!

[bbfrederick]

I do have a standalone simulator, and retrogress can make a limited simulation file. Let me see what I can come up with.