I wrote an algorithm which uses the number of pixels per region in the aseg+aparc.mgz file. As this file has 1x1x1mm3 I was expecting the number of pixels to match very closely the value of the volume; as each pixel represents a 1 mm3 cube.(I did the same for cortical volume and got a similar result, with an even larger discrepancy).
The aseg.stats file lists the nVoxel value, and it is basically the volume rounded up.
here are some sample values for subcortical regions:
Left-Lateral-Ventricle Volume (mm3): 7829.8
Left-Lateral-Ventricle (number of Pixels in aseg+aparc.mgz): 7396
Left-Inf-Lat-Vent Volume (mm3):594
Left-Inf-Lat-Vent(number of Pixels in aseg+aparc.mgz):569
Left-Cerebellum-White-Matter Volume (mm3):12668
Left-Cerebellum-White-Matter(number of Pixels in aseg+aparc.mgz):12203
Left-Cerebellum-Cortex Volume (mm3):44113.2
Left-Cerebellum-Cortex(number of Pixels in aseg+aparc.mgz): 43804
Left-Thalamus-Proper Volume (mm3):5752.8
Left-Thalamus-Proper(number of Pixels in aseg+aparc.mgz): 5899
Pretty much the values are close to the volume, but different enough to be significant. I wonder why this discrepancy.