Structural - McIntosh-Lab/tvb-ukbb GitHub Wiki

Our pipeline largely retains the structural (T1w, T2 FLAIR) preprocessing steps from the UK Biobank pipeline (Alfaro-Almagro et al., 2018). These include brain extraction and nonlinear registration to the MNI152 standard-space T1 template, defacing, bias correction and tissue-class segmentation (Figure 2). Processing of T2* images (brain extraction, registration to MNI152 and T1w, bias correction) has been added. Other major modifications and additions to the structural sub-pipeline are outlined below.

_{Figure 2: Structural sub-pipeline workflow. Original components of the UK Biobank pipeline with few or no modifications are in green; pipeline components with major changes or additions are indicated in white; and new components are indicated in orange. Dotted lines indicate components that are included in the QC report. Black lines indicate components that are used downstream by other sub-pipelines or included in ‘TVB Inputs’. GM: grey matter, WM: white matter}

To support connectome-based modelling in TVB, our additions to the structural sub-pipeline allow users to create connectomes from T1w, dMRI and resting-state fMRI data by specifying a brain parcellation of their choice. Currently, our pipeline supports parcellations defined on the MNI152 1mm template. For ease, we include three different parcellations in our repository. Two are combinations of the Schaefer cortical (Schaefer et al., 2018) with either the Tian subcortical (Tian et al., 2020) or Harvard-Oxford subcortical (Frazier et al., 2005) parcellation and the third is the Regional Map parcellation (Bezgin et al., 2017). The Schaefer-Tian parcellation is offered at three different scales of granularity and, if the user wishes, other scales can be created from the parcellations shared on the respective GitHub repositories. A tab-separated look-up table for the parcellation that specifies image labels and label names is required. The parcellation is registered to the T1w image using the warps from the nonlinear registration of the template to T1w.

In both healthy older adult and neurodegenerative samples, accurate tissue classification using T1w images is hindered by decreasing image contrast with age (Bansal et al., 2013). Additional difficulties in T1w tissue classification arise from the presence of white matter pathology, where white matter lesions become misclassified as grey matter (Levy-Cooperman et al., 2008). Since tissue classification is a vital part to defining accurate ROIs for both structural and functional connectivity, we have implemented a number of modifications to the segmentation procedure to improve ROI assignments. We derive an initial image segmentation following the UK Biobank’s procedure using FSL’s FAST toolbox. We then refine the grey matter subcortical segmentation by adding the outputs of FSL’s FIRST toolbox (an object model-based segmentation and registration tool) to the grey matter mask.

To address inaccuracies in the grey matter mask due to the presence of WM pathology, we have implemented two alternative methods that may be used depending on available image modalities. The first method, if T2 FLAIR images are available, uses the outputs of the WM lesion classification (FSL’s BIANCA) to exclude any misclassified voxels from the grey matter mask and add them back to the white matter mask. The second method is an option for when T2 FLAIR images are not available. In these cases, we use age-specific image classes (Fillmore et al., 2015) as tissue priors. T1w images from adults aged 40 or over are registered to the template for their age decile (e.g., 40s, 50s, etc.) while subjects aged under 40 are registered to the FSL-distributed tissue priors. These template space-registered T1w images are then segmented using the set of matching age-specific priors. Segmented images are registered back to T1w space. Age-specific templates are provided up to the 80s decile. Subjects greater than 89 years are registered to the 80s decile template.

The user-provided parcellation is registered to the T1w image and the grey matter mask is labelled with ROI indices. The labelled grey matter volume serves as input to the functional MRI sub-pipeline. The white and grey matter segmentations are both used to create the grey matter–white matter interface for dMRI tractography. This interface consists of voxels of white matter adjacent grey matter and, when labelled, will serve as the seed and target masks for tractography in the diffusion MRI sub-pipeline.