3D MR image denoising using a modified adaptive high order singular value decomposition method^.

Area of Science:

• Medical Imaging
• Signal Processing
• Image Denoising

Background:

• Magnetic Resonance (MR) images are susceptible to Rician noise, degrading image quality.
• Accurate noise reduction is crucial for reliable MR image analysis and diagnosis.

Purpose of the Study:

• To develop an advanced adaptive high order singular value decomposition (HOSVD) method for effective Rician noise removal in 3D MR images.
• To enhance the quality of noisy MR images by leveraging nonlocal self-similarity and weighted Schatten p-norm regularization.

Main Methods:

• A modified adaptive HOSVD approach was developed, incorporating nonlocal self-similarity and weighted Schatten p-norm.
• Similar 3D image patches (cubes) were identified using Euclidean distance to form a fourth-order tensor.
• Adaptive rank determination for unfolding matrices was achieved through weighted Schatten p-norm regularization.

Main Results:

• The proposed adaptive HOSVD method successfully reconstructed latent noise-free 3D MR images.
• Experimental results on synthetic and clinical data demonstrated superior performance compared to existing Rician noise removal techniques.
• The method effectively reduced Rician noise while preserving important image features.

Conclusions:

• The developed adaptive HOSVD method offers a significant improvement for Rician noise removal in 3D MR imaging.
• This technique holds promise for enhancing diagnostic accuracy and reliability in clinical MR applications.
• The study highlights the effectiveness of nonlocal self-similarity and weighted Schatten p-norm in MR image denoising.