Locally low-rank denoising in transform domains

Area of Science:

• Medical Imaging
• Signal Processing
• Quantitative MRI

Background:

• Locally low rank (LLR) denoising is effective for MRI series but struggles with limited image counts, common in quantitative MRI.
• Existing LLR methods' performance degrades with fewer images, impacting quantitative analysis accuracy.

Approach:

• A novel transform-domain extension to LLR denoising is proposed to enhance signal representation fidelity.
• This method improves the locally low rank approximation, even with limited data.

Key Points:

• The transform-domain approach effectively denoises MR image series using as few as 4 images.
• It demonstrates efficacy across various data types including k-space, DICOM, and SENSE-1 images.
• Preserves local image variability, unlike image-domain LLR methods that tend to homogenize quantitative maps.

Conclusions:

• The developed transform-domain LLR denoising method yields high-quality MR images.
• It is compatible with both raw k-space and vendor-reconstructed data.
• Enables improved imaging and more accurate quantitative MRI analyses and parameter extraction.