Quantification of Cerebral Blood Flow by Arterial Spin Labelling at 3T in Multi-Site Setting: Segmentation and Quality Rankings of CBF images

Quantification of cerebral blood flow (CBF) by ASL is gaining acceptance in the clinical assessment of a number of conditions, including stroke, brain tumours and dementia [1,2]. As 3T MR scanners become widely available in clinics, the gain in signal-to-noise-ratio (SNR) in ASL scans relative to 1.5T systems has meant more consistent quantification of gray matter (GM) CBF[3]. Clinical interest in CBF also concerns white matter (WM). However, blood flow in WM is approximately 50% of GM, making ASL quantification more challenging given the scan time constraints in clinical settings. Clinical ASL pulse sequences have been modified to provide larger signal difference between control and labelled images (cf. pulsed vs pseudo-continuous ASL), and a wide selection of ASL methods are now available in clinical scanners. In the current study, ASL data was collected for a volunteer scanned in 6 UK sites using standard 3T clinical systems within clinically acceptable scan times. This data was processed to reduce noise and correct for motion, and the scans were then ranked by mean SNR before WM/GM segmentation through a triple Gaussian model.

This study demonstrates automatic quality ranking of ASL scans from six different MR sites within normal clinical constraints. Effective GM/WM segmentation was demonstrated in two sites (sites 3 and 6, ranked Good) but results showed significantly more noise in the remaining four sites (especially those ranked Poor). These scans are a good representation of the large variation in SNR seen in ASL imaging in the clinical setting. Large SNR variation was observed between sites even though constraints such as field strength (3T) and spatial resolution were identical, along with very similar scan times. To improve segmentation in CBF images with lower SNR more advanced denoising efforts may be successful.