Hello-
I have been using SPM8's "New Segmentation" feature with a new tissue probability map (essentially, we want to include a lot more of the neck for better ERP source localization). The new TPM was generated from a population of 25 healthy individuals where we had 0.85mm isotropic T1 (MPRAGE) and T2 (SPACE) images. I started with the default SPM8 TPM and reiteratively generated a TPM for our population. In general, the results are really very good. This web page shows the data from a single indvidual:
http://www.cabiatl.com/CABI/resources/newseg/
While the new segmentation does in general produce really good results, it also generates a few really spurious results - for example "bone" well outside the head (perhaps due to the bias correction being applied extrapolated to my very large bounding box).
Here is my question: is there any way to include a "tissue cleanup" phase? For example, the standard (not new) SPM8 segmentation includes a tissue cleanup step. An alternative would be an option to ignore classifications for voxel locations that have a 0% chance of being that tissue type in the TPM. This would be terrible during TPM generation, but once the TPM is established for a population it works out pretty good (I have implemented this filter for images saved to normalized space, but would require changing the new segmentation code to efficiently do this for the images saved in standard space). In other words, is there some alternative form of tissue cleanup for new segmentation, or some way to enforce a stricter compliance to the TPM values?
-chris
|