An Unsupervised Method to Enhance both SNR and Edges for PPI

Introduction: Partially parallel imaging (PPI) techniques [1-2] reduce acquisition time at the cost of signal to noise ratio (SNR). Regularization techniques [3-4] can be used to improve image quality by balancing SNR and residual aliasing artifact/spatial resolution through a regularization parameter set. In this work, an unsupervised adaptive method is proposed to reduce noise and artifact level, as well as to sharpen edges. This method is based on Non-local Means (NL-Means) introduced by Buades et al. [5]. Results of the application to GRAPPA [2], with both phantom and in vivo data, demonstrate that the proposed method is able to increase SNR, to preserve the fine structures, and to sharpen the edges at the same time. Methods: Let } | ) ( { I i i v v ∈ = be the noisy input, NL-Means [4] uses weighted average of image values in non-local neighbors, instead of conventional spatial neighbors, to smooth the image. A pixel j is an non-local neighbor of a pixel i if they have statistically similar neighborhood,