On the precision of diffusion/perfusion imaging by gradient sensitization.

Computer simulation is used to assess the precision and accuracy of diffusion and perfusion parameters derived from a set of gradient-sensitized images. Under ideal experimental conditions, a moderate signal-to-noise level (ca. 40) suffices to estimate diffusion coefficients to within 20% relative precision. However, estimation of a typical cerebral perfusion fraction of 5% to within 20% relative precision requires signal-to-noise levels of ca. 400. Simulations also show that systematic errors in perfusion fraction estimation, as well as underestimation of the uncertainties in perfusion parameters (by chi-squared analysis), will be found at moderate signal-to-noise levels.