Overview Of Mesoscopic fMRI Data Analysis
In recent years, interest in anatomical and functional MRI data with very high spatial resolution has increased, especially for data measured with ultra-high field (UHF) MRI scanners with magnetic field strengths of 7 Tesla or higher. These devices allow to obtain images with sub-millimeter spatial resolution that can be used for "mesoscopic" brain imaging targeting cortical layers and cortical columns. In BrainVoyager, a set of specific tools are provided that either extent standard tools to be applicable for high-resolution data or provide new functionality enabling new analyses that are unique to sub-millimeter data.
One prerequisite for laminar and columnar fMRI analysis is that the cortex is segmented with very high quality both at the inner (white/grey) and outer (grey/CSF, pial) boundary. Furthermore blood vessels within grey matter should be removed to reduce vascular artifacts. The currently most advanced pipeline in BrainVoyager to achieve these prerequisites largely automatic is the deep neural network based cortex segmentator.
Another issue with UHF MRI data is that intensity inhomogeneities in T1-weighted data are much stronger than on 3 Tesla scanners. To cope with these issues, a proton-density based estimation technique has been implemented that largely corrects these inhomogeneities.
The creation of volume time course (VTC) data from original functional FMR data sets has been adjusted to properly handle functional data with sub-millimeter resolution. A new approach has also been implemented that creates VTC files without changing the voxel time course data in FMR-STC data. In this approach, the FMR data is transformed in FMR-VTC space, which is essentially the original FMR-STC space with some 90-degree rotations to adjust to the sagittal orientation of VMR space. The advantage of this FMR-VTC space is that it allows to run VTC space analysis tools (e.g. MVPA, grid sampling) without the necessity to resample the original data (even in case of non-iso voxels). The disadvantage of this approach is that the data can not be analyzed at the group level using standard normalized space since no transformation to MNI (or Talairach) space is performed. Other forms of group-level analyses (e.g MVPA, ROI analyses) can, of course, be applied.
The high-spatial resolution of UHF fMRI data demands very precise alignment of the data from multiple successive runs within a session. Alignment is also more difficult than for standard whole-brain data since sub-millimeter fMRI scans cover usually only a small slab of the brain. For an optimal across-run alignment, a new (masked) grid search VTC-VTC alignment tool is available.
Because of its high spatial resolution, UHF MRI data allows to look inside the intrinsic organization of the cortex both in depth (laminar fMRI) and along the cortex (columnar fMRI). In order to enable sampling functional data at different cortical depth levels, two approaches are availabe. One works with high-resolution - but otherwise standard - cortex meshes that are reconstructed at different relative cortical depth levels. A more advanced but only locally operating approach creates regular two-dimensional grids at different relative cortical depth levels. The latter approach has been successfully used to map the columnar-like organization of functional features in specialized brain areas as well as to separate functional data from different cortical layers. In some cases only small slabs of the brain are available and it might be necessary to perform precise segmentations of white and grey matter boundaries in a small regions; it might be, for example, useful to perform manual segmentations in a small brain region directly in functional data in order to avoid any misalignments with anatomical datasets. An increasingly powerful set of tools is currently developed that are described in topic "Manual Segmentation Tools" of the chapter "Brain and Cortex Segmentation". Most of the time BrainVoyager's deep neural network (DNN) or advanced segmentation tools should provide good cortex segmentations without the need of extensive manual editing.
Copyright © 2021 Rainer Goebel. All rights reserved.