An Optimized LMMSE Based Method for 3D MRI Denoising

Area of Science:

• Medical Imaging
• Signal Processing

Background:

• Post-acquisition denoising is crucial for accurate quantitative measurements in magnetic resonance (MR) imaging.
• Existing methods like Linear Minimum Mean Square Error (LMMSE) can be improved for better performance.

Purpose of the Study:

• To introduce a new, robust filtering method for MR image denoising.
• To leverage image self-similarity and Bayesian Mean Square Error (BMSE) for improved signal restoration.

Main Methods:

• A novel filtering technique based on Linear Minimum Mean Square Error (LMMSE) estimation is proposed.
• The method utilizes the self-similarity property of MR data and a patch-based L(2)-norm similarity measure.
• Bayesian Mean Square Error (BMSE) is employed for optimizing sample selection and minimizing estimation error.

Main Results:

• The proposed method demonstrates robust estimation performance compared to LMMSE and SNR-adapted LMMSE (SNLMMSE).
• Experimental results show competitive performance against state-of-the-art denoising techniques.
• Optimized sample selection and automatic parameter adjustment contribute to enhanced denoising.

Conclusions:

• The developed twofold data processing approach effectively denoises MR images.
• This method offers a significant improvement in MR image denoising for quantitative analysis.