Multi-stage automated local arterial input function selection in perfusion MRI

Cerebral blood flow (CBF) quantification using dynamic-susceptibility contrast MRI can be achieved via model-independent deconvolution, with local arterial input function (AIF) deconvolution methods identifying multiple arterial regions with unique corresponding arterial input functions. The clinical application of local AIF methods necessitates an efficient and fully automated solution. To date, such local AIF methods have relied on the computation of a singular surrogate measure of bolus arrival time or custom arterial scoring functions to infer vascular supply origins. This paper aims to introduce a new local AIF method that alternatively utilises a multi-stage approach to perform AIF selection.

A fully automated, multi-stage local AIF method is proposed, leveraging both signal-based cluster analysis and priority flooding to define arterial regions and their corresponding vascular supply origins. The introduced method was applied to data from four patients with cerebrovascular disease who showed significant artefacts when using a prevailing automated local AIF method.

The immediately apparent image artefacts found using the pre-existing method due to poor AIF selection were found to be absent when using the proposed method.

The results suggest the proposed solution provides a more robust approach to perfusion quantification than currently available fully automated local AIF methods.