I have some resting-state EPI data (340 volumes), 2.5mm voxels.
I have been attempting to replicate a previous analysis done by another research group and I am wondering if it is normal for my (unzipped files) to be so large or if I am doing something wrong. Here are the steps I am taking:
Rest EPIs start at 244mb
Realigned
Coregistered T2 to T1, and then the EPIs are coregistered to that step’s output . This is because we want to do our analysis in T1 space (1mm voxels)
5.21gigs
Smoothing
Denoising (confound regression + band pass filter)
10.41 gigs
Are these sizes normal? Is it good practice to zip output files?
I would not resample your BOLD data to T1w resolution. That is unnecessarily increasing your file size by a factor of ~15.6. Would be better to simply perform a rigid transformation to align the BOLD to T1 without any voxel upsampling.
Most images are compressed as .nii.gz files, that is standard.
Are these raw or after preprocessing? I notice there was no mention of common preprocessing steps, e.g., motion correction.
Right now I am using the SPM’s “Coregister: Estimate and Reslice” option. If I were instead to simply use “Estimate” my EPIs would be realigned but NOT resampled to 1mm space is that correct?
The eventual goal is to compare the localization of rest, ICA- derived components with the task-based GLM results of the same patient. The GLM results are in 1mm T1 space and so I want the rest results to be in the same comparable space l.
Is a better option to run my ICA in the 2.5mm EPI native EPI resolution and then upscale the final results to T1 space? That’s what I had been doing originally, but the results came out very “blocky”
A) Motion correction
B) Coregistered T2 to T1, and then the EPIs are coregistered to that step’s output
C) Smoothing
D) ART outlier detection
E) Regression of ART detected outliers and head motion (rr files of motion correction stage)
F) bandpass filtering .01 -.08
I am not a SPM user, so I cannot say, but that sounds possible.
I would not analyze fMRI data that has been upsampled that much. Was the modeling performed on native resolution or upsampled data?
I would not upsample at all.
It might be (sorry, not an SPM user), I guess it might depend on if this is how you are aligning the whole time series to T1w space (BOLD-to-T1w registration) vs aligning the BOLD internally to a BOLD reference (standard for motion correction and necessary to get accurate rotation and translation motion parameters).
I need to double check, but I am guessing the original group did all their GLM analysis in the EPIs original 2.5mm space…I need to figure out how they got from there to displaying the results on the patients 1mm T1 (since upsampling does not sound like the answer).
Scratch that. Here are some notes I found from the previous group:
“Each case’s language data was processed twice; once using a standard clinical pipeline (no normalization, for laterality Index calculation and normative reference maps) and once using normalization to allow processing in standard space (see below).
Processing. Batch steps: (1) Realign (est, write): all four runs’ EPI images were aligned with the mean (quality 0.9, 2mm FWHM smoothing; interpolation 4th-degree B-Spline); (2) Coregister (est, reslice): T2 brain to MPRage brain; (3) Coregister (est, reslice): All realigned EPI images [step 1] to the T2 brain in MPRage space [step 2]”