The mask used for optimal combining of the echos

Hi everyone,

I have the following two questions about fmriprep/tedana.

1- It seems to me that fmriprep when combining echos uses the mask generated from the first echo. While fmriprep calculates and outputs a combined mask from all echos I was wondering why it does not use the combined mask when using t2smap to optimally combine all echos.

2- If we use tedana to combine the preprocessed echos generated by fmriprep (using --me-output-echos), do we miss in processing of the fmriprep pipeline (assuming aroma denoising is not acticated)? In other words, the optimally combined output from tedana (not the denoised one) is identical to what fmriprep outputs in own space? The dataset is MBME.


Hi Ben

Sorry for a bit of a delayed response.

  1. tedana does a few interesting things with adaptive masking. The basic logic is that we can’t calculate an estimate across echoes if a later echo has too much dropout in a voxel. In the most recent version of tedana (not sure if fmriprep is locked to the most recent version) the optimally combined data will fit a model and use the weighted average of the echoes for each voxel with at least 3 good echoes. If there are fewer than 3 echoes, then (I think) it just takes the unweighted average of the good echoes. The denoised data are fit to a model only using voxels with >=3 good echoes, but the output is based on the mask used for the optimally combined data.

  2. I am not an fmriprep user, but my understanding is that fmriprep uses tedana to generate the optimally combined time series. That means you should get the same optimally combined volumes if you feed the preprocessed multi-echo data into tedana or run tedana through fmriprep. @tsalo is a regular tedana contribute who is also very active with fmriprep so he might correct me.

Hope this helps.