| Abstract: | The use of cerebral blood volume (CBV) maps generated from dynamic MRI studies tracking the bolus passage of paramagnetic contrast agents strongly depends on the signal-to-noise ratio (SNR) of the maps. The authors present a semianalytic model for the noise in CBV maps and introduce analytic and Monte Carlo techniques for determining the effect of experimental parameters and processing strategies upon CBV-SNR. CBV-SNR increases as more points are used to estimate the baseline signal level. For typical injections, maps made with 10 baseline points have 34% more noise than those made with 50 baseline points. For a given peak percentage signal drop, an optimum TE can be chosen that, in general, is less than the baseline T2. However, because CBV-SNR is relatively insensitive to TE around this optimum value, choosing TE ≈ T2 does not sacrifice much SNR for typical doses of contrast agent. The TR that maximizes spin-echo CBV-SNR satisfies TR/T1 ≈ 1.26, whereas as short a TR as possible should be used to maximize gradient-echo CBV-SNR. In general, CBV-SNR is maximized for a given dose of contrast agent by selecting as short an input bolus duration as possible. For image SNR exceeding 20–30, the Γ-fitting procedure adds little extra noise compared with simple numeric integration. However, for noisier input images, as can be the case for high resolution echo-planar images, the covarying parameters of the Γ-variate fit broaden the distribution of the CBV estimate and thereby decrease CBV-SNR. The authors compared the analytic noise predicted by their model with that of actual patient data and found that the analytic model accounts for roughly 70% of the measured variability of CBV within white matter regions of interest. |
