Functional - McIntosh-Lab/tvb-ukbb GitHub Wiki

Functional MRI sub-pipeline

The fMRI sub-pipeline processes both resting-state- and task-fMRI data (Figure 3). The processing of both data types by the UK Biobank pipeline relies on FSL’s FEAT toolbox. As best practices for preprocessing of fMRI data are both dataset-dependent and constantly evolving (Uddin, 2017), the pipeline allows users flexibility on selecting the right preprocessing methods for their needs. Users may specify their preferences, which can include brain extraction, motion correction via realignment of fMRI images (MCFLIRT), slice timing correction, spatial smoothing, intensity normalization, and temporal filtering. Registration to the T1w image and MNI152 template is performed. For resting-state fMRI data, automated classification and removal of noise artefacts is performed using FMRIB’s ICA-based Xnoiseifier (FIX) (Griffanti et al., 2014). We have modified the UK Biobank pipeline to now accept an arbitrary number of fMRI sessions. Other major additions and modifications are described below.

Figure 3: fMRI sub-pipeline workflow. Original components of the UKBB pipeline with few or no modifications shown in red; pipeline components with major changes or additions shown in white; and new components shown in orange. Dotted lines indicate components that are included in the QC report. Black lines indicate components that are included in the TVB Inputs.

_{Figure 3: fMRI sub-pipeline workflow. Original components of the UKBB pipeline with few or no modifications shown in red; pipeline components with major changes or additions shown in white; and new components shown in orange. Dotted lines indicate components that are included in the QC report. Black lines indicate components that are included in the TVB Inputs.}

Field map correction

The UK Biobank pipeline performs geometric distortion correction for the unwarping of EPI (e.g., fMRI and dMRI) images. This correction requires a reverse phase-encoded B0 dMRI image for estimating the field map, which is not always available. To support more “traditional” field map acquisitions for EPI distortion correction, such as those in the Cam-CAN dataset, we have implemented the option for dual echo-time gradient distortion correction using FSL’s FUGUE toolbox.

Resting-state fMRI

We have updated the pipeline’s FIX version from 1.063 to 1.06.15. Although FMRIB provides a default trained-weights file, and we provide trained-weights files for both the ADNI3 and Cam-CAN datasets, the classifier performs best when trained with the user’s specific dataset. The most notable addition to resting-state fMRI processing is the replacement of group-ICA-based detection of resting-state networks with the parcellation of the resting-state fMRI data to accommodate connectome-based modelling. Following denoising, the parcellation output from the structural sub-pipeline (Figure 2) is registered to a reference resting-state fMRI volume and the average BOLD response across voxels is computed for all ROIs (i.e., ROI time series). The Pearson correlation coefficient between all ROI time series is also computed to obtain a measure of functional connectivity.

Task-based fMRI

In our implementation of the fMRI sub-pipeline, task-based fMRI data are minimally preprocessed but not further analyzed. Users may choose to re-implement a GLM-based analysis using FEAT or, alternatively, they may take the preprocessed task-fMRI data and apply other analytic methods (e.g., Partial Least Squares; McIntosh and Lobaugh, 2004).