I'm trying to study a different filtering process (not as Gaussian FWHM usually used for fMRI), in a BOLD signal in a healthy control just to see if this new filter could provide a better activation segmented area. But I need some tips for making the more appropriate quality measurements for my research. What would be the best measurements for this purpose? Is it the root mean square error (RMSE) between the Gaussian filtering and my filtering approach in the activation segmented area a reasonable quality measurement?
I don't have much expertise in fMRI analysis, for this reason any suggestion from a specialist in the fMRI area will be a great help for me. I'm using FSL MELODIC tools to create the ICA for activation maps.
Hi!
Did you create the filter yourself? Do you already have some experience with simulated data and different types of noise and how your filter reacts? Starting with real data is nice only if you have true-positive and true-negative control conditions to compare. If your filter is dynamic you will need that. Another thing you may want to do is to filter the data twice, first with your filter and then with the conventional one. If your filter is better, you should get better results after filtering data twice or after filtering it with your filter only. Better than only looking at the rmse would be to run some bootstrapping and to have a look at the posterior distribution of activated voxels for your true-positives and true-negatives.
Cheers,
G
- Badges
- Science topic