Related Experiment Videos
Adaptive anisotropic noise filtering for magnitude MR data
J Sijbers1, A J den Dekker, A Van der Linden
1Department of Physics, University of Antwerp, Antwerpen, Belgium. sijbers@ruca.ua.ac.be
Magnetic Resonance Imaging
|December 28, 1999
Summary
Magnetic resonance (MR) images often have noise that is Rice distributed, not Gaussian. This study introduces a new noise filter that accounts for Rice distribution, reducing bias in MR image analysis, especially in low signal areas.
Area of Science:
- Medical Imaging
- Signal Processing
- Biophysics
Background:
- Conventional noise filtering in magnetic resonance (MR) imaging assumes Gaussian noise.
- Magnitude MR data are actually Rice distributed, leading to biased filtering results.
- Bias is particularly pronounced in low signal-to-noise ratio regions.
Purpose of the Study:
- To develop a noise filter for magnitude MR images that accounts for the true Rice distribution of the data.
- To reduce bias introduced by conventional, Gaussian-based filtering methods.
Main Methods:
- Developed a novel noise filtering algorithm incorporating the Rice probability distribution.
- Evaluated the filter's performance on simulated and real MR data, focusing on low SNR regions.
Main Results:
- The proposed Rice-based filter significantly reduces bias compared to conventional filters.
- Improved noise reduction and image quality were observed, especially in areas with low signal intensity.
Conclusions:
- Accounting for the Rice distribution in MR noise filtering leads to less biased and more accurate results.
- The developed filter offers a significant improvement for MR image analysis, particularly in challenging low SNR conditions.