ANTs DenoiseImage for fMRI EPIs?

fmri
ants
denoising
preprocessing

#1

I understand that DenoiseImage is a spatial filtering program that removes noise assuming an underlying noise distribution (Rician or Guassian). The original paper this program is based on only discusses applications on structural scans.

Would anyone recommend using DenoiseImage instead of spatial smoothing on EPI collected during functional MRI? I’ve ran it through a few examples and the images look incredibly clean - I’m just wondering if I’m removing information from the EPI that I shouldn’t be.

Wondering if anyone has any opinions before I actually put in the legwork to see how denoising with his method impacts downstream analysis.

Title of original ref:
Adaptive Non-Local Means Denoising of MR Images With Spatially Varying Noise Levels

Note this is a cross post on the ANTs Issues board


#2

I looked into this a while ago and:

  1. images (both EPI and T1w) look indeed much cleaner
  2. this is a computationally intensive algorithm (will take a long time running this on every single EPI volume)
  3. I did not see any analysis of how this influences tSNR or statistical power (or in context of T1w images how it interacts with VBM or surface reconstruction). If anyone knows of such analysis I would love to read it.

More context: https://github.com/poldracklab/fmriprep/issues/1020


#3

Just to add a note: adaptive NLM is typically used on diffusion MRI preprocessing so it seems it may make sense for EPIs. Please note it would also enhance drop-out regions, accentuating signal loss.


#4

Is there any literature on how this influences tractography or other DWI analyses?


#5

I don’t know any :frowning:


#6

Thank you both for your valuable feedback!

I can run my resting fmri data through DenoiseImage and compare it against smoothed data with various kernel sizes.

My original plan was to align, slice time correct, spatial smooth/denoise, register to template space, ICA AROMA denoise then derive networks using whole brain group ICA with split-half sampling for estimating stable model order.

Here is what I had in mind for comparing the effect of DenoiseImage against Gaussian Spatial Blurring:

  • Biases in reproducibility of stable model orders revealed with split-half sampling. I.e) does the method of smoothing influence the number of components that are reproducible across subjects?
  • Spatial cross-correlation between estimated ICA networks between different methods
  • DenoiseImage generates noise estimate maps. These can be compared against the ICA components for the smoothed data. Noise estimate maps can also be compared against the mean denoised EPI to identify drop-out regions (any ideas on quantifying degree of drop out?)
  • temporal signal to noise before / after

Any other ideas or things you’d like to see?

Oh and one question for the community - would you perform spatial denoising/smoothing before or after registering to a template space? I would think prior denoising/smoothing might improve the registration (using ANTs here with a study-specific template).