Challenge: how can one guarantee that any given standard space voxel contains data from the same part of the same white matter (WM) tract from each and every subject?
We project individual subjects' FA data into a common space in a way that is not dependent on perfect nonlinear registration. This is achieved through:
- an initial approximate nonlinear registration;
- projection onto an alignment-invariant tract representation (a.k.a. mean FA skeleton)
VBM is most commonly carried out using the SPM software package. It allows one, subject to interpretation caveats, to find changes anywhere in the brain - it is not necessary to prespecify regions or features of interest.
Alignment issues: one must be very careful not to misinterpret residual misalignments. If one cannot guarantee that alignment is "correct", then it must be assumed that sensitivity to true differences is suboptimal.
Smoothing issues: it is not generally known in advance if the extent of smoothing is matched to the spatial extent of the structure of interest. Smoothing increases the partial voluming problem.
Alternative: specify an ROI. More sophisticated approaches use tractography; the relevant tracts are usually identified by initialising/constraining tractography using hand-drawn ROIs.
What is a MNI152 space? - Kindly supplied by Andrew Janke; derived from 152 structural images, averaged together after high-dimensional nonlinear registration into the common MNI152 co-ordinate system. It corresponds to the "152 nonlinear 6th generation" atlas. (http://www.ncbi.nlm.nih.gov/pubmed/17354756 )
In TBSS, we attempt to bring together the strengths of each approach (VBM-style analysis and Tractography-based approach). We aim to solve the alignment and smoothing issues, while being fully automated, investigating the "whole" brain - not requiring prespecification of tracts of interest.
This is achieved by estimating a "group mean FA skeleton", which represents the centres of all fibre bundles that are generally common to the subjects involved in a study. Each subject's FA data is then projected onto the mean FA skeleton in such a way that each skeleton voxel takes the FA value from the local centre of the nearest relevant tract, thus hopefully resolving issues of alignment and correspondence.
TBSS Steps:
- Preprocessing
- Nonlinear alignment
- Identifying the target for alignment
- Creating the mean FA image and its skeleton
- Statistics and thresholding
沒有留言:
張貼留言