Multi-b-value DWI (diffusion weighted image) noise reduction method based on mutual information

Abstract

The invention discloses a multi-b-value DWI (diffusion weighted image) noise reduction method based on mutual information. The method comprises steps as follows: acquiring to-be-tested multi-b-value DWIs in multiple diffusion sensitive gradient directions by a magnetic resonance scanner; performing preprocessing such as head movement correction, eddy current correction and the like on the images; estimating different b-value image signal-to-noise ratios; performing noise reduction processing on the images through setting parameters of an optimal noise reduction algorithm so as to acquire DWIs with higher signal-to-noise ratios. The method fully considers features of different images, and three-dimensional structural features of the processed images are reserved as many as possible while noises are eliminated. During establishment of a high-order complex model, different b images have higher image quality, so that acquired high-order parameter images can have smaller errors and fewer singular values. The method can be better applied to preprocessing of a high-order model imaging technology and plays an active role in further research on white matter fiber structures.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a mutual information-based noise reduction method for multi-b value diffusion weighted images. Background technique [0002] Diffusion Weighted Image (DWI) is a new magnetic resonance (MR) imaging technique developed on the basis of nuclear magnetic resonance. It uses the diffusion motion characteristics of water molecules in living tissue for imaging, and can form diffusion-weighted images by adding diffusion-sensitive gradients to conventional MR sequences. Due to the complex electromagnetic environment in the imaging process, various noises and artifacts will appear in DWI, including artifacts related to the main magnetic field, artifacts related to the radio frequency magnetic field, eddy current noise, etc., and movement will occur due to factors such as head movement Artifacts. The existence of these noise factors will directly affect the identification of tissue structure...

AI Technical Summary

Problems solved by technology

However, studies have shown that the noise distribution of DWI images obeys the Rice distribution

Recently, some people have proposed weighted mean filter, non-local mean filter, linear minimum mean square error, etc., these methods can effectively remove the Rician distribution [1] noise, and retain the important boundary information of the image, but it is difficult for them to improve the data quality on a large data set with multiple b values ​​and multiple directions

Images