De-noising of 3D multiple-coil MR images using modified LMMSE estimator

Area of Science:

• Medical Imaging
• Signal Processing
• Computational Science

Background:

• De-noising is vital in Magnetic Resonance Imaging (MRI) for preserving image details.
• Multiple-coil MRI systems accelerate imaging but introduce different noise models than single-coil systems.
• Single-coil MRI noise is Rician, while multiple-coil MRI noise follows a noncentral Chi (nc-χ) distribution.

Purpose of the Study:

• To propose a novel filtering method for de-noising multiple-coil MR images corrupted by nc-χ noise.
• To enhance image quality and detail preservation in accelerated MRI acquisition.
• To address the limitations of single-coil de-noising methods in multiple-coil scenarios.

Main Methods:

• A new filtering method based on the Linear Minimum Mean Square Error (LMMSE) estimator is proposed.
• The Bayesian Mean Square Error (BMSE) criterion is utilized and proven for the nc-χ noise model.
• A nonlocal voxel selection methodology is developed for the nc-χ distribution.

Main Results:

• The proposed method demonstrates robust and accurate performance in de-noising.
• Effective detail preservation is achieved during noise suppression in multiple-coil MR images.
• The method outperforms existing state-of-the-art de-noising techniques on both ideal and GRAPPA reconstructed images.

Conclusions:

• The novel LMMSE-based filtering method effectively de-noises multiple-coil MR images with nc-χ noise.
• The proposed approach offers superior performance in terms of accuracy and robustness compared to current methods.
• This technique contributes to improved image quality in accelerated multiple-coil MRI.