Test-retest reproducibility assessments for longitudinal studies: quantifying MRI system upgrade effects