In Magnetic Resonance Imaging(MRI), the acquired complex valued data obtained as the inverse Fourier transform of the raw k-space data are corrupted by Gaussian distributed noise. In this letter, we present an adaptive neighborhood selection algorithm by accounting for the spatial context information, calculated in the undecimated wavelet domain. A modified Wiener filter is approached by a novel bivariate thresholding, considering local signal variance by valid information from the selected neighborhood windows. In contrast to the previously proposed methods in the literature, our scheme can preserve edges more effectively while suppressing noise and artifacts; moreover, both the SNR and RAISE are improved obviously.