search

layerfMRI / LAYNII

Stand alone fMRI software suite for layer-fMRI analyses.
BSD 3-Clause "New" or "Revised" License
85 stars 26 forks source link

LN_NOISE_KERNEL: How to estimate anisotropic smoothness from its output? #97

Open samtorrisi opened 3 months ago

samtorrisi commented 3 months ago

Hi Renzo and Faruk-

I have data that I know is smoother in one dimension than the others. It seems LN_NOISE_KERNEL can theoretically provide anisotropic smoothing estimates but according to Figure 10 here https://layerfmri.com/2020/04/06/qa/ I'll still have to do steps 6 and 7 on my own, right?

So I'm curious if you have some ready-made code to do that so I don't have to build from scratch.

I simply need the FWHM in units of voxels or mm for the x, y and z dimensions separately. this would be for both low res and high res functional data (although i don't think that matters for this tool?)

let me know what you think and thanks!

-Sam

layerfMRI commented 3 months ago

Hi @samtorrisi, Yes, you are right. The extraction of the projection from the nifti file to a ascii text file is not done in LayNii. Neither is the Gauss fitting. I think the former is very easy to include into LN_NOISE_KERNEL. Thanks @ofgulban to add this to the mile stone for V2.8.0. I am less sure about the Gauss fitting. I am hesitant to include dependencies in LayNii to libraries that would do fitting. And fitting is annoying enough to implement from scratch in C++.

samtorrisi commented 3 months ago

hi Renzo-

that definitely sounds like a lot of work and from a developer stand-point i totally get why you'd not want to add extra dependencies. i'll give rolling-my-own a shot with either matlab or python, no prob and thanks for considering!

-Sam

ofgulban commented 3 months ago

I will look into this in summer after OHBM. I think I might be able to implement the Gaussian fitting from scratch. Thanks @samtorrisi for bringing this this program and the blog post section back to our attention.

samtorrisi commented 3 months ago

oh wow, well thank you @ofgulban, if you think it'll have wider interest at least give it a shot. note that i wouldn't be using Gaussian estimates for cluster correction (which Eklund et al 2016 and Cox et al 2017 demonstrated were problematic) but rather need relative but quantitative differences between different pulse sequences, for example. yes it can wait until after OHBM and have fun there!