The role of noise in the measurement of cerebral blood flow and partition coefficient using xenon-enhanced computed tomography

Monte Carlo simulations have been used to study the accuracy which can be expected in the quantification of blood flow and the partition coefficient using xenon-enhanced computed tomography in the presence of noise. We have demonstrated that the markedly asymmetric frequency distribution of estimates increases in size rapidly with an increase in the standard error of the input CT data. On the basis of our results, we recommend that controllable sources of noise (eg. CT number drift) be corrected and that estimates be obtained by averaging CT data and then fitting, rather than averaging blood flow and partition coefficients derived from individual pixels, as the latter procedure results in the introduction of considerable bias.