ADAPTIVE JOINT DATA SELECTION FOR SPARSITY BASED ARTERIAL SPIN LABELING MRI DENOISING.

Area of Science:

• Medical Imaging
• Biophysics
• Signal Processing

Background:

• Arterial Spin Labeling (ASL) perfusion MRI is a non-invasive, radiation-free method for quantifying tissue perfusion.
• ASL MRI signals suffer from low signal-to-noise ratio due to T1 decay, complicating accurate perfusion quantification and texture preservation.
• Existing denoising methods often struggle with preserving local image structures.

Purpose of the Study:

• To develop an unsupervised denoising strategy for ASL perfusion MRI that enhances signal-to-noise ratio and preserves local image structures.
• To introduce a joint data selection strategy that leverages correlations between label and control images for improved sparsity regularization.
• To demonstrate the effectiveness of the proposed method without relying on ground-truth training data.

Main Methods:

• A joint data selection strategy was developed, capitalizing on correlations between paired label and control (L/C) ASL images.
• Highly correlated content was assembled into potentially sparse matrices for adaptive sparsity regularization.
• The method employs local sparse models, outperforming global models in signal-noise separation and structure preservation.

Main Results:

• The proposed method significantly enhances the quality of ASL perfusion maps.
• It effectively preserves local image structures during the denoising process.
• The approach outperforms conventional pipelines requiring multiple L/C pairs, using only a single L/C pair.

Conclusions:

• The novel unsupervised denoising method improves ASL perfusion MRI quality and texture preservation.
• The joint data selection and sparsity regularization strategy is adaptive and effective.
• This approach offers a significant advancement for ASL perfusion imaging, particularly in noise reduction without ground truth data.