Test-retest reproducibility of quantitative susceptibility mapping in a multi-site study on Alzheimer disease: effect of reference region of interest choice