Hi,
I’m writing a review on spinal fMRI data processing (addressing reviewers comments), and I’m looking into the application of slice timing correction for spinal fMRI data. I revisited the following publications:
The Eippert/Kong articles thoroughly describe methods for incorporating slice-specific nuisance regressors (based on external pulseOx and respiratory signals), but I don’t find any mention of slice-timing correction for the underlying BOLD signal, for all the above articles.
The Barry et al. article looks at within-slice correlations, so I guess slice timing correction was not needed there, but in general, what is the recommended approach for slice-timing correction in spinal fMRI time series?
Any thoughts/references would be appreciated 
Julien
1 Like
Hi @jcohenadad - I did not use slice timing correction in the preprocessing of my 7T data because they were acquired with a 3D slab excitation and thus all slices were essentially acquired at the same time (or, perhaps more accurately, within the same temporal interval). I currently have a manuscript under review that acquired data using single-shot EPI and we did perform slice timing correction on those data (though the impact was presumably minor since the TR was quite reasonable at 2 sec). Hope this helps! -Rob
1 Like
Hi @jcohenadad
We did not use slice timing correction for our 3T spinal cord fMRI data as well. For task fMRI, it is not recommended to use slice timing in FSL, rather people would use temporal derivatives instead. That’s what we used in the GLM, to account for both slice timing and different HRF. For resting state data, we also used FEAT and GLM for pre-processing & physiological noise correction, again temporal derivatives was included.
Recently for our new data with ZOOMit or multi-band, we are exploring how different pre-processing strategies change results, we found that actually slice timing could improve results in terms of TSNR, activation and connectivity. We might recommend to use it in the future.
Best,
Joe
1 Like
@Yazhuo_Kong i’m not sure i understand why slice timing is not recommended for task fMRI, and how temporal derivatives can address this issue. E.g., in your approach, the nuisance regressors (cardiac, resp), include slice timing information, and therefore are slice-specific. But these are only the nuisance regressors, not the actual task-related BOLD signal, which would still have a different timing across slices, right? Same argument for resting-state: we could assume that neuronal connectivity is fast (and independent of) compared to slice timing acquisition, therefore correlations made across slices should account for this temporal offset 
Hi @jcohenadad ,
I was exploring whether or not slice-timing correction improves task fMRI results, and was wondering what your current thoughts were on this matter.
If I understood correctly, FSL can create a GLM with slice-specific regressors (motion, physiological noise) that have different values in each voxel, depending on their acquisition time. If this is truly the case, adding slice-timing information in the FSL pre-stats should help, especially with other regressors ?
1 Like
@raphaschl thank you for reaching out. I think @Yazhuo_Kong and @barryrl (and potentially other readers of this forum?) are more experienced than me on that matter. Hoping they will chip in 
2 Likes
Hi @raphaschl – thanks for your question. What’s the TR of your data? The conventional fMRI wisdom in the brain, which, in this case, likely holds true for the cord, is that slice timing correction doesn’t make much of a difference if your TR is ~1 sec or less, and makes a moderate difference if it is ~1-2 sec. My original 7T spinal cord fMRI data had an effective TR of ~3.3 sec, but I used a 3D volume excitation rather than contiguous 2D slices so I couldn’t apply slice timing correction anyhow.
In your GLM model, you can add both the motion and physiological noise regressors. Adding their first derivatives will effectively shift the original regressors in time. So, you can consider a couple things between preprocessing and GLM modeling, which likely interact in non-trivial ways (…such is life! 
).
-Rob
Hi, thanks for your quick answers ! My TR is of exactly 2s, it hence seems that there is no obvious course of action in this case. Adding first and second derivatives sounds good, and might avoid the not-so-useful temporal interpolation that comes with slice-timing correction. Thank you for your insight
1 Like