Spectral subtraction de-noising of MRI

Abstract

De-noising techniques can improve the signal to noise ratio (SNR), and quality of magnetic resonance (MR) images. In this work, we introduce a spectral subtraction de-noising (SSD) method that operates directly on the acquired raw MR signals and then we reconstruct images using the de-noised signals to improve the SNR. MR images acquired using coil arrays and reconstructed using parallel imaging techniques exhibit spatially varying noise distribution, which hampers the performance of image de-noising techniques applied in the image domain. The proposed SSD method is applied in the k-space (Fourier) domain of each of the individual coil array elements and is thus not affected by non-uniform noise distribution. Using numerical simulations and experimental data, we show that up to 45% improvements in SNR in both single and multi-channel coil data can be achieved.