Freesurfer uses an atlas scaling factor (ASF) to calculate the eTIV, so even though the ROI encompasses all others, it is not calculated in the same way as other ROIs (i.e. by counting the voxels inside the skull). Rather, it estimates a linear transformation to an atlas space (Talairach) and calculates the determinant of the transformation matrix, which correlates to ICV \cite{buckner2004unified}. So actually the mechanisms behind the biases in eTIV are different than those behind other regional volumes (i.e. regional variation in CNR and gradient distortions affect the boundary). Because of this, test retest reliability of eTIV was included, but was excluded from the scaling factor calculations and validations.