An MRI Image Denoising Algorithm Using Neural Network And Wavelet Transform

Magnetic Resonance Imaging (MRI) plays an inreplaceable role in clinic diagnosis.However, the quality of MRI images is affected by the noise that comes inevitably during data acquisition. Because the distribution of noise in MRI images is not Gaussian, in other words, the noise is signal-correlated, the methods used to denoise Gaussian- distributed noise cannot be applied to MRI image denoising well.Wavelet transform has an important position in imaging processing, especially in image denoising. In area of image denoising, wavelet-domain methods have achieved results much better than those achieved by old image domain schemes. Artificial neural network is a simple simulation of human neural network. It has been used successfully in the area of pattern recognition. Through the process of training, neural network can find the hidden relationship between the input and expected output. During simulation, the neural network can give appropriate output according to the input. In this thesis, achievements in wavelet domain denoising and artificial neural network are combined to bring forth a new neural network based wavelet domain denoising scheme. The new scheme uses neural network to find the relationship between wavelet efficients of noisy image and wavelet efficients of clean image, instead of the traditional thresholding and shrinking. Because of the insensitiveness of neural network to the distribution of the noise, the proposed method can deal with different distributions of noise. According to the results of our experiments, our method can be successfully used in MRI image denoising.