A study on the comparison of median filter regularization methods in diffusion tensor MRI.

Area of Science:

• Neuroimaging
• Medical Physics
• Computational Neuroscience

Background:

• Diffusion Tensor Magnetic Resonance Imaging (DT-MRI) non-invasively visualizes central nervous system axonal fiber architecture.
• Noise in DT-MRI eigenvectors can lead to inaccurate fiber orientation estimation.
• Accumulated noise errors in tractography compromise the integrity of long fiber tract reconstruction.

Purpose of the Study:

• To evaluate matrix-valued median filters for structure-preserving denoising in DT-MRI.
• To compare the efficacy and efficiency of three matrix median computation algorithms: Simple Median, Gradient Descent, and Successive Fermat Methods.

Main Methods:

• Implementation of Simple Median Method for matrix median computation.
• Application of Gradient Descent Method for matrix median computation.
• Utilizing the Successive Fermat Method for matrix median computation.
• Assessing the performance of these methods in structure-preserving denoising of DT-MRI data.

Main Results:

• The Successive Fermat Method demonstrated superior denoising performance compared to the Simple Median Method.
• The Successive Fermat Method achieved comparable performance to the Gradient Descent Method.
• The Successive Fermat Method exhibited faster computation than the Simple Median Method.

Conclusions:

• Matrix median filters are effective for denoising DT-MRI data.
• The Successive Fermat Method presents a promising algorithm for DT-MRI tractography, balancing performance and efficiency.
• This method can enhance the accuracy of central nervous system fiber architecture analysis.