ICA-based denoising for ASL perfusion imaging.

Area of Science:

• Neuroimaging
• Medical Physics
• Radiology

Background:

• Arterial Spin Labelling (ASL) generates brain perfusion images but suffers from intrinsically low signal-to-noise ratio (SNR).
• Current denoising strategies increase scan duration and are susceptible to motion artifacts, limiting clinical utility.
• Time constraints and patient motion in clinical settings necessitate more efficient denoising methods for ASL.

Purpose of the Study:

• To investigate the efficacy of Independent Component Analysis (ICA) for denoising ASL data.
• To assess the potential for automation of ICA-based denoising in ASL imaging.
• To compare ICA denoising with existing methods like aCompCor and ENABLE.

Main Methods:

• Analyzed 72 pseudo-continuous ASL datasets from 30 acute stroke patients across 5 post-labeling delays.
• Applied manual and automated ICA denoising, comparing results to aCompCor and ENABLE.
• Evaluated data quality using metrics including cerebral blood flow (CBF), arterial transit time (ATT), SNR, and repeatability.

Main Results:

• ICA denoising significantly increased mean CBF and ATT values (p<0.001) and reduced their variance (p<0.001).
• ICA demonstrated significantly improved SNR (p<0.001) and repeatability (p<0.05) compared to raw data.
• ICA-denoised data achieved higher SNR with 50% less acquisition time than raw data, outperforming aCompCor and ENABLE.

Conclusions:

• ICA effectively separates signal from noise in ASL data, enhancing overall data quality.
• Automated ICA denoising performs comparably to manual methods, offering potential for clinical workflow integration.
• ASL acquisition time could potentially be reduced by 50% without compromising data quality, warranting further investigation.