kjamison / fmriclean

fMRI denoising and parcellation

fmriclean

fMRI outlier detection, and parcellation, and denoising

To be be run AFTER preprocessing. The expected order of operations would be:

1. fmri_save_parcellated_timeseries.py: Save parcellated timeseries with NO denoising, filtering, etc
   • Inputs = full NIfTI time series, ROI labeled volume
   • Outputs = *_ts.mat file with [timepoints x ROIs]
2. fmri_outlier_detection.py: Save information about motion and global signal outliers (comparable to ART in CONN toolbox)
   • Inputs = full NIfTI time series and motion parameter file, and a brain mask file for computing global signal (recommended if data is very large)
   • Outputs = outliers.txt (Tx1 binary vector of 1s and 0s where 1 = outlier timepoint) and outlier_parameters.mat (timeseries used for outlier estimation including global signal, FD, dvars, and individual motion regressors)
3. fmri_save_confounds.py: Compute nuisance regressors from data
   • Inputs = full NIfTI time series, motion parameter file, outlier.txt file, masks for WM, CSF, and GM (for GSR and CompCor regressors)
   • Outputs = confounds.mat (TxM) time series with columns for GSR, CompCor, motion, and outlier nuisance regressors
4. fmri_clean_parcellated_timesries.py: Denoise and/or filter the parcellated time series from (1), and save FC matrices
   • Inputs = parcellated _ts.mat from (1), confounds.mat from (3)
   • Options = Perform bandpass filtering after denoising, compute FC matrix using correlation (Pearson), covariance, precision (inv(cov)), or partial correlation
   • Outputs = FC(type).mat (RxR connectivity matrix), and/or _tsclean.mat (TxR denoised time series)

fmriclean's Issues

QA Questions

Hi there,

Thank you so much for making this denoising pipeline publicly available. I am relatively new to fMRI research and this pipeline has been a tremendous help throughout my graduate studies - both for learning the topics surrounding denoising and for denoising execution.

I have a couple of novice fMRI connectivity questions:

1. Following preprocessing, should the fMRI data (rs-fMRI data in my case) be normalised prior to denoising? Specifically, upon inspection of the ts.mat file I noticed that the time series values are in the e+04 range. I know that when examining the time-series data on FSL it's possible to normalize the data from -1 to +1. If so, are there any efficient ways of normalising fMRI data following preprocessing?
2. Are there any QA checks that you would recommend implementing during/following the fmriclean denoising pipeline? Currently, I ensure that the GM, WM, and CSF masks are registered correctly to the fMRI data, as well as, examine the cofounds.mat file to observe the extracted ts for the WM and CSF confounds. Are there any visual checks similar to the CONN toolbox that you would recommend?

Once again, thank you so much.
Abhi