It seems that the voxel values in the output of 5tt2gmwmi are 0-1, but not a binary mask, what does the probability value mean?
Hi @LiuYuchen,
The values generated by the 5tt2gmwmi are a bit of a kludge.
In the ideal scenario, if you wanted to perform streamline seeding in such a way that those seeds are distributed approximately homogeneously across the cortical surface, then you would generate them based on a native representation of that surface. However we do not yet have that capability implemented.
In its initial implementation, the 5tt2gmwmi did produce a binary mask image. Candidate seeds would be drawn at random from within that mask, and those would be revised to the GM-WM isocontour in the interpolated 5TT image. However depending on the position & shape of the cortical surface relative to the image voxel grid, this could result in some regions of the brain having many more voxels in that mask than others, which would lead to some areas of the cortical surface possessing many more streamline seeds, and this bias would contribute to a inhomogeneity of streamline terminations across the cortical surface.
The non-binary output of 5tt2gmwmi is an attempt on my part to try to get behaviour closer to that of the ideal scenario, but still utilising image-based processing for which the code implementation is already in place. What is actually quantified in that command (as of 2014) is related to the magnitude of the gradient of the GM / WM partial volume images. These data are then used in a rejection sampling framework, where candidate streamline seeds are more likely to be placed in voxels with larger values.
The net effect of this is to reduce the extent to which different grey matter areas have different densities of streamline seeds per unit surface area. It’s a hack, but it helps a little.
Rob