@Martin_Bzr, good catch! I did not realize this before.

If you look closely at how NGTDM is defined, you can see there is a volume dependency in the calculation of `s_i`

(sum of absolute differences between centervoxel and average of neighborhood). As the image becomes larger, this sum also increases.

In some NGTDM formulas, this is corrected by some use of ` N_vp`

(total number of voxels in ROI), but not for Coarseness (and Busyness and Strength). A potential non-volume confound variation on Coarseness would be `N_vp / SUM(s_i)`

, which would reflect the (inverse of the) average of differences between center voxel and neighborhood.

As to GLN for GLRLM and GLSZM, that is very likely, and maybe partly the reason the GLNN formula is also there.