Purpose: To investigate a wavelet-based filtering scheme for denoising of arterial spin labeling (ASL) data, potentially enabling reduction of the required number of averages and the acquisition time.
Methods: ASL magnetic resonance imaging (MRI) provides quantitative perfusion maps by using arterial water as an endogenous tracer. The signal difference between a labeled image, where inflowing arterial spins are inverted, and a control image is proportional to blood perfusion. ASL perfusion maps suffer from low SNR, and the experiment must be repeated a number of times (typically more than 40) to achieve adequate image quality. In this study, systematic errors introduced by the proposed wavelet-domain filtering approach were investigated in simulated and experimental image datasets and compared with conventional Gaussian smoothing.
Results: Application of the proposed method enabled a reduction of the number of averages and the acquisition time by at least 50% with retained standard deviation, but with effects on absolute CBF values close to borders and edges.
Conclusions: When the ASL perfusion maps showed moderate-to-high SNRs, wavelet-domain filtering was superior to Gaussian smoothing in the vicinity of borders between gray and white matter, while Gaussian smoothing was a better choice for larger homogeneous areas, irrespective of SNR.