Input parameter sensitivity analysis and comparison of quantification models for continuous arterial spin labeling

Abstract

The regional cerebral blood flow (rCBF) values determined using continuous arterial spin labeling (CASL) are subject to several sources of variability, including natural physiologic variations, sensitivity to the input parameters, and the use of different quantification models. To date, a thorough analysis of the impact of input parameters and the choice of quantification model has not been performed. These sources of variability were investigated through computer simulations using bootstrap techniques on actual CASL data. Coefficients of variation for representative single voxels were 6.7% for gray matter and 29% for white matter, and for eight-voxel regions of interest they were 4.5% for gray matter and 23% for white matter. Comparison of nine CASL quantification models showed differences in gray matter rCBF values of up to 42%. An analysis of the sensitivity of the rCBF to input parameters for each of the nine quantification models demonstrated that accurate quantification of the inversion efficiency, tissue and arterial blood longitudinal relaxation times, and transit times were critical in calculating precise rCBF values. The large potential variations in rCBF and the effect of the choice of quantification model suggest that interpreting absolute rCBF values in CASL studies can be challenging and requires great care.