Presenting novel de-noising techniques for brain MRI

Standard acquisition of MRI presents Rician statistical noise that degrades the performance of other steps of the image analysis. In this paper we present new ways to reduce the noise of brain images in the preprocessing stage. We propose a novel wavelet domain method for noise restoration based on discrete wavelet packets transform (WPT). The developed techniques combine adaptive Wiener filter and soft threshold for the 2D wavelet packet coefficients of the best tree decomposition. The novel presented techniques are compared with the most traditional one considering qualitative and quantitative results. In the comparison the Mean Square Error and the normal cross correlations are considered for a complete set of structural (T1-w) brain MRI. Moreover we show by experiments that the common prior adaptive Wiener filtering often used by many authors is a dispensable procedure.