How important is an individual MRI?

[jyeatman]

Compare the results going from sensor -> Individual MRI -> FSaverage
vs
sensor -> FSaverage

• simulation
• real data

[mdclarke]

@jyeatman left side is prek_1715 co-registered and warped to fsaverage. Right side is prek_1715 co-registered to own anatomy then morphed to fsaverage. Doesn't look like much of a difference.

Next step is to simulate some sources .

[jyeatman]

@mdclarke this is a really really important finding - if it turns out that accuracy is quite similar with and without an individual MRI, that would be incredibly useful to quantify and report. I think it would be useful to do this a couple ways:

1. Using actual data like this and showing how different the group summary looks
2. Simulating sources coming from an individuals MRI and then doing the same analysis on simulated data

If you are up for it, I think it would be worth making some summary figures of:

1. What you have above but averaged for all the subjects in the pre session
2. A simulation
   And then looping in Eric and Pat and considering pushing through a methods paper

[larsoner]

@jyeatman I got access to this repo to check out some processing scripts so I'm implicitly looped in. (If this is problematic for some/any reason let me know!) Just wanted to pop in to say that I'm not up on the latest literature, but there is at least some existing work on surrogate MRI accuracy. One paper that is worth looking at would be this 2006 one about thin-plate spline (TPS) warping vs affine warping benchmarked against using individual MRIs:

https://www.ncbi.nlm.nih.gov/pubmed/16037984

Sources were simulated at 972 locations evenly distributed over the inner cortical surface of the atlas. The mean error over all 10 subjects was 8.1 mm [for TPS] ..., using an affine transformation of the electrodes into atlas space and an FEM model generated from the atlas produced mean errors of 22.3 mm in subject space ...

From looking at that paper's citations it's probably possible to find a lot of the relevant more recent literature, too,

FWIW the TPS warp they talk about is implemented in MNE, but so far I haven't used it because I wasn't confident that the infant digitization was reliable enough to give enough usable degrees of freedom for the higher-order TPS fitting.

[jyeatman]

Glad to have you on the repo!

How would this compare to the current approach of using fsaverage?

[larsoner]

How would this compare to the current approach of using fsaverage?

Our mne coreg offers 3-axis scaling (plus standard rotation/translation), so in theory it should be pretty comparable to the affine transformation case they talk about above (affine is more general than what mne coreg will do, but I doubt the additional degrees of freedom would matter here).