Inclusion Criteria For Voxels on Segment Statistics

Hi all,

I’m using segment statistics to use some MRI-defined segments to export some quantitative PET data. This usually works pretty well, but does not work if my segmentation is small (less than half the PET voxel volume). I suspect this might be because of the inclusion criteria - binary mode as some softwares support (example on pg 188 I think it would make sense in my scenario to use fraction mode (example also on pg 188 of the previous link). Is there a way I can use fraction mode in Slicer? If not - is this something that might get implemented?


There is fractional labelmap support in Slicer but it would require some work to make it work with Segment statistics. If you could contribute to this then let us know.

Alternatively, you can choose the segmentation geometry to have 2-3x oversampled resolution relative to the master volume (and if the input volume is highly anisotropic then it may worth forcing isotropic spacing):


1 Like

Hi Andras,

Thank you for the prompt reply. It’s good to know that labelmap can use this. I ought to contribute to that, but right now I have to go with a quick solution. I did not attempt the suggested alternative because the segmentation geometry was already oversampled (high resolution MRI) relative to the PET, and I suspect that’s what got me in this pickle. Instead, I used the module Crop Volume to resample the PET image at 4 times the 1D resolution using nearest neighbour interpolation (to preserve the numbers).

Many thanks,