A Non-Local Rician Noise Reduction Approach for3-D Magnitude Magnetic Resonance Images

The visual quality of Magnetic Resonance Images(MRI) plays an important role in accuracy of clinical diagnosiswhich can be seriously degraded by existing noise duringacquisition process. Therefore, denoising is of great interest fordiagnostic aims and also the ability of automatic computerizedanalysis. Noise in Magnitude MRI is usually modeled by Riciandistribution which introduces a signal-dependent bias andreduces the image contrast. In this article an efficient approachfor enhancement of the noisy magnitude MRI based on therecently proposed linear minimum mean square error (LMMSE)estimator is introduced. The natural redundancy of the acquiredMR data is employed to improve the performance of unknownsignal estimation. Since in practice, the MR data is in a largemajority 3-D, the proposed method is developed to deal with 3-DMR volumes. The quantitative and qualitative metrics have beenused to demonstrate and compare the performance of theintroduced approach with several state-of-arts denoisingschemes. Experimental results show that the proposed methodrestores delicate structural details conveniently while thecomputational cost remains low.