Imputation models and error analysis for phase contrast MR cerebral blood flow measurements

Abstract

Cerebral blood flow (CBF) supports brain metabolism. Diseases impair CBF, and pharmacological agents modulate CBF. Many techniques measure CBF, but phase contrast (PC) MR imaging through the four arteries supplying the brain is rapid and robust. However, technician error, patient motion, or tortuous vessels degrade quality of the measurements of the internal carotid (ICA) or vertebral (VA) arteries. We hypothesized that total CBF could be imputed from measurements in subsets of these 4 feeding vessels without excessive penalties in accuracy. We analyzed PC MR imaging from 129 patients, artificially excluded 1 or more vessels to simulate degraded imaging quality, and developed models of imputation for the missing data. Our models performed well when at least one ICA was measured, and resulted in R2 values of 0.998–0.990, normalized root mean squared error values of 0.044–0.105, and intra-class correlation coefficient of 0.982–0.935. Thus, these models were comparable or superior to the test-retest variability in CBF measured by PC MR imaging. Our imputation models allow retrospective correction for corrupted blood vessel measurements when measuring CBF and guide prospective CBF acquisitions.