Could threshold free cluster enhancement be adapted to segment WMH in FLAIR data? I know TFCE is designed to be used on statistical datasets but clusters of activation seem similar to the hyperintense voxels in anatomical FLAIR.
I tried experimenting with this and had some success before hitting a wall. I masked the FLAIR sequence by a freesurfer white matter mask. Then I normalized the dataset by converting to Z scores. The WMH are typically about 2-3 standard deviations above the mean. Then I ran:
fslmaths -tfce 0 0.5 6
The output covers the hyperintensities very well but is almost completely insensitive to their actual values. It assigns all WMH and surrounding tissue the same value (ceiling effect?). When I increase the H parameter an empty dataset is returned.
Why would increasing H result in an empty dataset? Is it even possible to use TFCE in this way or am I headed for a dead end with this?
Thanks in advance for any help and advice.
|