Disabling eddy and topup?

I am trying to do preprocessing and reconstruction on minimally preprocessed ABCD dwi data, which I was told, already had eddy and TOPUP done on them.

I want to configure qsiprep so that those two are not applied, but cannot find the right command to do so:


https://qsiprep.readthedocs.io/en/latest/preprocessing.html#merge-denoise
In the image above in qsiprep docs, I thought that the first method could be used, but found that the first method still performs eddy. Is there a way to disable FSL based dwi preprocessing completely?

Hi,

I’m not sure QSIprep will be the best tool for you. Other steps such as denoising are supposed to be done before Eddy/TOPUP, so trying to run qsiprep on data with these steps already done is going to be problematic. I would either 1) preprocess the raw data, or 2) wait for the next ABCD-BIDS Collection 3165 Preprocessed Data Release, which will include QSIPrep outputs. Collection 3165 - ABCD-BIDS Community Collection (ABCC)

Best,
Steven

1 Like

Thank you for the reply! I looked at the link you sent me, but it seems that their latest release was two years ago, so I think that I should preprocess the raw data. Thank you for the insight :slight_smile:

@Steven Could it be possible to exclude denoising completely by doing --denoise-method none and run qsiprep? (and assume that the deonisiong is already done on my data), since, as this picture shows, some of the denoting already seems to have been performed?
image

If not running Eddy, TOPUP, or Denoising, then what additional preprocessing are you expecting QSIPrep to do? And like I said earlier, the order of steps matters, so I don’t think passing in minimally preprocessed data into QSIPrep is a good move. If you just want to run reconstruction, you can try to trick QSIPrep into thinking your data came from QSIPrep by applying BIDS naming to them, just as QSIPrep does for its outputs. I can’t guarantee everything will work as expected though.

1 Like

Hmm i see! Thank you for your insight!

Quick question: do you have access to raw, unprocessed ABCD diffusion data?

Raw ABCD data can be accessed after signing a Data Use Agreement according to instructions here: https://nda.nih.gov/tutorials/abcd/query_and_download.html?chapter=querying-abcd. For followup questions please start a new thread.

Hi Steven! If you dont mind me asking : When you mean by bids data, are you referring to the fast-track data?

Ah, my apologies. I was going to suggest that, if they do have access to the ABCD raw-imaging data (which looks like it’s an “all-package deal”) and want to do the full pre- + post-processing in QSIprep, I was going to suggest they either convert the dicoms themselves (some post-bids processing may be needed) or try using the abcd-dicom2bids pipeline from DCAN Labs

1 Like

My bad, @ajschadler, I thought you were asking how to get the raw data, I see what you were going for now. There are some very specific weird errors of running ABCD through QSIPrep that are being addressed by the Collection 3165 - ABCD-BIDS Community Collection (ABCC) team. I would just wait until they do their next release, which will include QSIPrep outputs.

1 Like

Since our lab has a small dataset acquired using ABCD diffusion sequences, this is good to hear!

1 Like

Thank you for your response! Are the errors wide-spread (i.e. do most of the ABCD subjects have the issue?)? The last update from ABCC seems to be like two years ago, and since we want the data fast, we were wondering if we could just run the raw data directly. Will this be bad because of the “weird errors” you mentioned?

Thank you for your suggestion! have you tried your suggestion before? (i.e. BIDS => DICOM => QSIPREP). If so, how was it? was it successful?

We processed some of our own diffusion data (again, using ABCD pepolar diffusion sequences on GE scanners) through QSIPREP; I was fairly happy with the few we tested.

Converting ABCD data to BIDS requires some extra processing steps after the initial BIDS conversion. You can try using the ABCD - dcm2bids pipeline I posted previously; that should do most of the “post-bids” processing for you.

N.B. I have not used used the DCANLab abcd-dcm2bids pipeline myself…

I am not sure how wide spread it is, but the reports are that SDC ends up creating an extra set of vesicles. It should be fixed in QSIPrep 0.16.1+

The BIDS data are here: https://nda.nih.gov/edit_collection.html?id=3165, at least for baseline data. They have not BIDSified longitudinal yet.

1 Like

Hi @Steven and @Kore_ana, The problem in ABCD was present in ~10% of the subjects and had to do with topup struggling when the b=0’s used for distortion correction look very different from the b=0’s in the processed dwi. If you use version 0.16.1 or higher we’ve verified that this issue is fixed.

If you look at the html report and see more than 1 corpus callosum in the topup report, then you’ve run into the issue.

2 Likes

Thank you everyone for your great insights!! :slight_smile: