Multi Echo-Time ASL MRI Denoising via Joint Complex Low-Rank Regularization.

Arterial spin labeling (ASL) magnetic resonance imaging (MRI) is the only non-invasive technique for measuring regional cerebral blood flow (CBF) and related neurovascular properties. ASL data can be acquired at multiple echo times (TE) to estimate transit time from blood to tissue or blood to CSF. However, this introduces a significant challenge in ASL data processing due to the increased signal decay during long TEs, further exacerbating the low signal-to-noise ratio (SNR) of ASL signal. Existing ASL denoising methods predominantly focus on magnitude, neglecting the more informative complex data containing both magnitude and phase. To address this gap, we propose Complex Rank minimization with Adaptive data Selection (CRAS) for multi-TE ASL MRI denoising. CRAS capitalizes on the holistic processing of complex data and leverages the correlation between label and control images, forming genuinely low-rank matrices. By filtering patch groups collaboratively, CRAS exhibits robust noise reduction and preserves intricate local structures within multi-TE perfusion maps. It autonomously learns from the input data, devoid of any reliance on ground-truth training data. This trait is particularly advantageous, given the unavailability of high-quality noise-free ASL MRI in practical scenarios. Evaluation on a real-world ASL MRI dataset demonstrates that CRAS significantly enhances multi-TE ASL perfusion maps compared to the standard pipeline and the current state-of-the-art MRI denoising method.