Non-local mean denoising in diffusion tensor space.

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Image Processing

Background:

• Diffusion Tensor Imaging (DTI) is crucial for visualizing white matter architecture.
• DTI data is susceptible to noise, which can affect tractography and analysis.
• Existing denoising methods may not fully preserve the complex tensor properties.

Purpose of the Study:

• To introduce a novel non-local mean (NLM) denoising method specifically for DTI data within the tensor space.
• To evaluate the effectiveness of tensor similarity measures for voxel weighting in DTI denoising.
• To compare the proposed method against traditional NLM and unbiased NLM approaches.

Main Methods:

• Developed a novel non-local mean (NLM) method operating in the diffusion tensor space.
• Implemented tensor similarity measures including Euclidean distance (rotational invariance), Riemannian distance, and Log-Euclidean distance (affine invariance).
• Quantitatively and qualitatively assessed the denoising performance using these measures in DTI space and diffusion-weighted image space.

Main Results:

• The novel NLM method demonstrated improved denoising capabilities in the tensor space.
• Tensor similarity measures effectively weighed voxels, preserving geometric and orientation features of diffusion tensors.
• Comparisons showed advantages over the original NLM and unbiased NLM in terms of accuracy and efficacy.

Conclusions:

• The proposed tensor-space NLM method offers a robust approach for denoising DTI data.
• Utilizing tensor similarity measures enhances the preservation of complex diffusion tensor properties.
• This novel method holds potential for improving the reliability of DTI-based neuroimaging studies.