Brain perfusion imaging by multi-delay arterial spin labeling: Impact of modeling dispersion and interaction with denoising strategies and pathology.

Abstract

Arterial spin labeling (ASL) acquisitions at multiple post-labeling delays may provide more accurate quantification of cerebral blood flow (CBF), by fitting appropriate kinetic models and simultaneously estimating relevant parameters such as the arterial transit time (ATT) and arterial cerebral blood volume (aCBV). We evaluate the effects of denoising strategies on model fitting and parameter estimation when accounting for the dispersion of the label bolus through the vasculature in cerebrovascular disease. We analyzed multi-delay ASL data from 17 cerebral small vessel disease patients (50 ± 9 y) and 13 healthy controls (52 ± 8 y), by fitting an extended kinetic model with or without bolus dispersion. We considered two denoising strategies: removal of structured noise sources by independent component analysis (ICA) of the control-label image timeseries; and averaging the repetitions of the control-label images prior to model fitting. Modeling bolus dispersion improved estimation precision and impacted parameter values, but these effects strongly depended on whether repetitions were averaged before model fitting. In general, repetition averaging improved model fitting but adversely affected parameter values, particularly CBF and aCBV near arterial locations in patients. This suggests that using all repetitions allows better noise estimation at the earlier delays. In contrast, ICA denoising improved model fitting and estimation precision while leaving parameter values unaffected. Our results support the use of ICA denoising to improve model fitting to multi-delay ASL and suggest that using all control-label repetitions improves the estimation of macrovascular signal contributions and hence perfusion quantification near arterial locations. This is important when modeling flow dispersion in cerebrovascular pathology.