Cerebral perfusion measurements using continuous arterial spin labeling: accuracy and limits of a quantitative approach