Imputation models and error analysis for phase contrast MR cerebral blood flow measurements in heterogeneous pediatric and adult populations.

Cerebral blood flow (CBF) supports brain function and health. Cerebral blood flow is affected by normal brain development, disease, medications use, and other interventions. One method to measure CBF is phase contrast magnetic resonance (PC MR) imaging, a particularly fast and reliable method to measure blood flow through major arteries such as the internal carotid (ICA) or vertebral arteries (VA). Unfortunately, sometimes PC MR can be compromised due to errors by the technologist during image acquisition, patient movement, or complex vessel structures. Our goal was to develop mathematical models to estimate CBF for a wide age range of patients whenever 1 or more vessels are not correctly measured. To investigate this, we studied a set of 258 PC MR acquisitions from a group of 196 patients with one or three acquisitions per subject (165 single images, 31 acquisitions of 3 images) ranging in age from 0.4 to 61.3 years (mean [μ] = 13.1, standard deviation [σ] = 12.3). We deliberately excluded measurements from one or more arteries in each volunteer to mimic situations with low image quality. Subsequently, we developed mathematical models to predict the missing data. Our predictive models performed well; across the human lifespan when at least one ICA measurement was available, our normalized root mean squared error values were low (<0.137), our R-squared values were high (>0.91), and we observed high intra-class correlation coefficients (>0.951). In summary, these imputation models are effective in estimating CBF in children and adults.