Abstract
OBJECTIVES: To numerically and experimentally investigate the robustness of intravoxel incoherent motion (IVIM) magnetic resonance imaging in measuring perfusion indexes in the human brain. METHODS: Eighteen healthy volunteers were imaged on a 3 T clinical system. Data of IVIM imaging (12 b-values ranging from 0 to 1000 s/mm(2), 12 repetitions) were fitted with a bi-exponential model to extract blood volume fraction (f) and pseudo-diffusion coefficient (D*). The robustness of measurement was assessed by bootstrapping. Dynamic susceptibility contrast (DSC) imaging and arterial spin-labelling (ASL) imaging were performed for cross-modal comparison. Numerical simulations were performed to assess the accuracy and precision of f and D* estimates at varied signal-to-noise ratio (SNRb1000). RESULTS: Based on our experimental setting (SNRb1000 ~ 30), the average error/variability is ~5 %/25 % for f and ~100 %/30 % for D* in gray matter, and ~10 %/50 % for f and ~300 %/60 % for D* in white matter. Correlation was found between f and DSC-derived cerebral blood volume in gray matter (r = 0.29 - 0.48 across subjects, p < 10(-5)), but not in white matter. No correlation was found between f-D* product and ASL-derived cerebral blood flow. CONCLUSIONS: f may provide noninvasive measurement of cerebral blood volume, particularly in gray matter. D* has limited robustness and should be interpreted with caution. KEY POINTS: • A minimum SNR b1000 of 30 is recommended for reliable IVIM imaging. • f may provide noninvasive measurement of cerebral blood volume. • f correlates with CBV DSC in gray matter. • There is no correlation between fD* and CBF ASL . • D* has limited robustness and should be interpreted with caution.