Denoising Magnetic Resonance Images Using Collaborative Non-Local Means

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Image Processing

Background:

• Noise artifacts in magnetic resonance (MR) images complicate analysis and reduce reliability.
• Effective denoising is crucial for robust quantitative analysis of MR images.
• Non-local means (NLM) filters offer state-of-the-art denoising but rely on repeating patterns within a single image, which are often scarce in complex structures like the human brain.

Purpose of the Study:

• To introduce a novel collaborative non-local means (CNLM) approach for denoising MR images.
• To leverage repeating structures across multiple images to improve denoising performance.
• To enhance the preservation of structural details during the MR image denoising process.

Main Methods:

• Proposed a collaborative denoising strategy utilizing repeating structures from multiple MR images.
• Spatially aligned multiple images (potentially from different subjects) to a target image.
• Performed NLM-like block matching using aligned images as a reference to increase matching structure availability.

Main Results:

• The proposed CNLM method demonstrated superior denoising performance compared to the classic NLM filter.
• Experiments on synthetic and real data confirmed the effectiveness of CNLM.
• CNLM yielded results with significantly improved preservation of structural details.

Conclusions:

• Collaborative Non-Local Means (CNLM) effectively denoises MR images by exploiting cross-image structural similarities.
• CNLM offers a significant advancement over traditional NLM filtering, particularly for complex datasets.
• The method holds promise for improving the reliability and detail in quantitative MR image analysis.