MR KLEAN: a Generalized Acquisition-agnostic LLR k-Space Denoising Method for High-dimensional Imaging.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Medical Imaging
• Signal Processing

Background:

• Accelerated MRI acquisitions are often limited by thermal noise, impacting image quality.
• Existing image-domain denoising methods struggle with non-Cartesian sampling and advanced reconstructions due to noise distribution assumptions.
• There is a need for denoising techniques that are independent of acquisition trajectory and reconstruction strategy.

Purpose of the Study:

• Introduce MR KLEAN, a k-space low-rank denoising framework.
• Develop a method agnostic to MRI acquisition trajectory and reconstruction strategy.
• Address limitations of current denoising methods in high-dimensional and dynamic MRI.

Main Methods:

• MR KLEAN exploits locally low-rank structure in multichannel, high-dimensional k-space data.
• Data are prewhitened, and Casorati matrices from local k-space patches are denoised using singular-value thresholding.
• Thresholds are determined via Monte-Carlo simulations based on known noise statistics.

Main Results:

• MR KLEAN significantly increased SNR and CNR in phantom studies.
• In vivo ASL imaging showed reduced noise, improved relative SNR, and enhanced functional connectivity analysis.
• Accelerated cardiac cine imaging demonstrated noise reduction and improved delineation of anatomical features with preserved temporal fidelity.

Conclusions:

• MR KLEAN offers robust, acquisition- and reconstruction-agnostic k-space denoising.
• The method improves MRI image quality and enables flexible spatial-temporal trade-offs.
• High-dimensional k-space data possess intrinsic low-rank structure, even with temporal signal variations.