Tedana unable to find BOLD components

Hi tedana folks!

I’m working on putting together a pre-processing pipeline for some multi-echo data my lab has collected. So far, they are run through fmriprep, I grab the images that are head motion, slice time, and susceptibility distortion corrected using collect_fmriprep.py from the fmriprep_wf, and run those files through tedana. Approximately 2/3 of them are giving me a warning of “No BOLD components detected! Please check data and results!” I am using tedana .10.0 and it does change seeds and try again each time it can’t find any BOLD components, and the logs report a change of seed. However, this never solves the issue. I am using all the defauts, and only giving Tedana the data and echo times.

My scans range in length, some being a 4 minute rest scan and some being 9-12 minute task scans, with a TR of 1.49s and three echoes. In general, the 4 minute rest scans are more likely to have BOLD components, but there is variability. There does not seem to be a pattern across sessions or participants with which scans have BOLD components. I’m afraid I’m not familiar enough with component analysis to look at the reports and find anything meaningful.

Are there any additional troubleshooting steps for me to try? Default parameters to change?


Hi McKenzie,

There are a few things you can check - you mention the reports - one thing to look for is any sort of task structure, if you would expect it to be visible, or any task frequency. If it were a visual, block design task, you should expect to see a component that maps onto the visual cortex, with the timing of your blocks. Of course if the task is a little more complex, then this may not be possible. At the very least you should see some components that look like they correspond to gray matter boundaries and regions.

A few questions - you mention the TR, but other details may help - for example: what is the Field strength, brain coverage, TEs used and voxel size? And for a few runs where it didn’t find any components, how many dimensions was the data reduced to? As in, how many components were evaluated?

As for what you could try, you could try the following --tedpca kundu option, documented here which may have a big impact on component selection.

This should be a start to sorting out what is going on.


1 Like

That’s a good point re: component selection.

How many components are you getting @mphagen ? And are you using the default tedpca options?

1 Like

I agree with @e.urunuela that something going wrong in the PCA step resulting in too few components is the most likely cause. That said, if you’re running the identical command for both, it’s surprising to me that the long runs are having this problem more often, so there there might be something else going on here.
If you can share or screenshot the tedana_report.html for a couple of good an bad runs of the same run type, that might help identify some other issues.