Fmriprep does not combine multi-echo timeseries

Hi all,

I am running fmriprep 1.2.6-1 on dual-echo data but the timeseries do not get combined.

Here is the command I use:

singularity run /home/xyz/Documents/fmriprep-1.2.6-1.simg
–participant-label $p
–fs-license-file /home/xyz/Documents/license.txt
–output-space {T1w,template}
–n_cpus 48

Is there anything I have to change about my command or data structure so that multi-echo timeseries are not being processed separately anymore?

Thank you very much for your help!

Dual-echo time series are not currently combined as, to the best of my understanding, the current literature does not recommend a combination method for dual-echo acquisitions.

Instead, dual-echo acquisitions generally follow the Bright and Murphy denoising strategy, in which the first echo is used as a regressor time series.

It would not be possible to combine dual-echo time series using the optimal combination method advocated by Kundu and colleagues, as we require at least three echos to map the T2* decay.

Is there other relevant literature you can recommend in which dual-echo time series are used to generate an optimal combination ? Happy to be pointed to other combination strategies !


For our last study, we weighted the average of the two images (for the 2 TEs) by the tSNR in each voxel calculated from 30 additional TRs measured at the beginning of each scan. See (in that paper, the method is called “CNR weighted”) and (“CoCNR” or “dual_weighted”).

Thanks for sharing these ! I’m less familiar with the dual-echo literature, so it’s great to see this has been addressed. Do you have a python implementation of this combination method ?


No sorry, just Matlab. If you’re interested in the code, I can ask if the person that provided it is willing to share.

By the way, the Halai et al. paper (the latter one above), shows that plain averaging (linear summation) of the two images basically works just as well the CNR method for dual-echo data, and this is also what several studies I know have done, e.g. So the CNR method might not be worth the trouble, especially if you want to use independent data of your task data to calculate tSNR (you should, of course…), which requires additional measurement time.

I’m running into the same situation here. Should I just add the raw files of the two echos and then feed it into fmriprep as if it is single echo? Or should it be done in some later stages?

Thanks in advance