Magnetic resonance image denoising using spectral data substitution

A novel method for denoising MR images is proposed that is based on the principle of k-space data substitution. The method consists of classifying the k-space data of the image to be denoised into two subsets: the preserved set containing original higher SNR samples that are kept unchanged during denoising process, and the substitution set where the original samples having lower SNR are substituted by those reconstructed from a so-called two-dimensional (2-D) SFA model. That allows higher spatial frequencies, often ignored or altered by conventional denoising techniques, to be recovered, thus leading to better denoising performance. The denoising mechanism was mathematically formulated, and eventual denoising errors were theoretically analyzed. The denoising method produced significantly greater noise reduction while preserving more accurately image edges and details.