Stochastic variational inference improves arterial transit time estimation from multi-delay arterial spin labelling data

Motivation: Arterial transit time (ATT) is a quantity of increasing clinical interest for understanding neurovascular health. ATT may be measured using multi-delay arterial spin labelling (ASL), but due to the low signal to noise ratio and non-linear kinetics of this imaging technique, obtaining accurate and robust measurements can be challenging. Goal(s): To better estimate arterial transit time from multiple-delay arterial spin labelling data. Approach: Model-fitting via stochastic variational Bayesian inference with joint spatial and non-spatial prior regularisation. Results: The new method is more robust than existing methods on noisy data, and produces maps with greater anatomical detail on acquisition data. Impact: The advantages of this new method will enable researchers to better exploit ATT measurement via multi-delay ASL, furthering understanding of neurovascular health.