SPM Multi-echo Preprocessing

, , ,
1

I am running tedana v25.1.0 on multi-echo fMRI data (3 echoes, TE=13.80/32.06/50.32ms) as part of an SPM preprocessing pipeline. I am comparing the optcom and denoised outputs.

I expected optcom to preserve more voxels in dropout-prone regions (e.g. OFC/PFC) due to its more liberal masking. However, quantitative comparison of voxel counts shows the two outputs have identical spatial coverage.

In the final preprocessed output overlaid on MNI152, I consistently see a sliver of signal dropout and missing coverage in the PFC/OFC region across both approaches and all subjects.

My questions:

  1. In tedana v25.1.0, does the denoised output apply the conservative ICA mask to the final timeseries, or is the denoising projected back onto the full optcom voxel space?

  2. Is PFC/OFC signal dropout expected with multi-echo fMRI and is there a recommended approach to improve coverage in these regions?

  3. Would upgrading to v26.0.3 change this behavior?

Thanks in advance,
Mary

2

Can you post the tedana call you used?

3

Thanks for the follow up.

tedana -d \
    dcrasub-XX_task-cdm_run-1_echo-1_bold.nii \
    dcrasub-XX_task-cdm_run-1_echo-2_bold.nii \
    dcrasub-XX_task-cdm_run-1_echo-3_bold.nii \
    -e 13.80 32.06 50.32 \
    --out-dir dcrasub-XX_task-cdm_run-1_echo-2_bold \
    --mask sub-XX_task-cdm_run-1_sbref_mean_bet_mask.nii \
    --fittype curvefit --maxrestart 10 --maxit 500 --tedpca kic
4

If you created the initial mask from data averaged across echoes, that may explain why you’re getting the same coverage as tedana. The main reason multi-echo data recovers signal in dropout-affected regions is that the early echoes are not as affected by dropout, so signal can be retained from them. Averaging across echoes should accomplish that, even outside of tedana, though tedana’s weighted average may improve signal recovery. @handwerkerd might know more about this.

5

More generally, A voxel that is not in your inputted mask will not be recovered by tedana. To maximize retained voxels, I suggest using a mask that was created on just your shortest echo, with the least amount of dropout. If you want to play around with bet you can possibly set it to more liberally retain voxels. –Dan

6

Thank you both for the explanation. This is very helpful.

Maryam