Research On MR Complex Image Denoising With Wavelets

Magnetic resonance imaging (MRI) is a powerful diagnostic technique in Medical Imageology in recent years. However, fast imaging or high resolution achievement often lead to great noise artifacts in reconstructed magnetic resonance (MR) images, which interfere with subsequent clinical diagnosis and cure. The noise in MR magnitude images obeys a Rician distribution. Rician noise, unlike additive Gaussian noise, is signal-dependent and consequently noise removal is a difficult task. As we know, the noise contributions, arising from the real and imaginary parts of the k-space complex data, are additive, assumed to be uncorrelated, and characterized by a zero-mean Gaussian noise. So, it is available to denoise the real and imaginary parts of the complex images.The paper studies wavelet-domain filtering method for the complex Gaussian noise removal, and introduces Translation-Invariant (TI) property into the Wiener-like filtering, combining one SURE harding-threshloding with two Wiener filters and produces one novel wavelet-domain filtering algorithm, which is applied to denoise respectively the real and imaginary parts of the MR complex images. This algorithm has a great improvement to Wiestain's method in noise removal. However, because of the independent filtering of the real and imaginary parts, the phase distortions occur in the denoised outputs and it is time-consuming. In order to reduce computation complex and handle the phase distortion, this paper uses directional extension for wavelets, divides three wavelet subbands obtained by undecimated wavelet transform (UWT) at each scale into six finer directional subbands to improve the directionality of wavelet transform, adopting only SURE harding-threshloding, and presents the other novel wavelet-domain filtering algorithm, which is applied to denoise the complex data as a whole in MR images. The results of the simulated experiments suggest that the two proposed algorithms can effectively reduce the noise artifacts, retrieve more marginal information and outperform previous MR images denoising methods.