Using the robust principal component analysis algorithm to remove RF spike artifacts from MR images

Abstract

Purpose Brief bursts of RF noise during MR data acquisition ("k-space spikes") cause disruptive image artifacts, manifesting as stripes overlaid on the image. RF noise is often related to hardware problems, including vibrations during gradient-heavy sequences, such as diffusion-weighted imaging. In this study, we present an application of the Robust Principal Component Analysis (RPCA) algorithm to remove spike noise from k-space. Methods: Corrupted k-space matrices were decomposed into their low-rank and sparse components using the RPCA algorithm, such that spikes were contained within the sparse component and artifact-free k-space data remained in the low-rank component. Automated center refilling was applied to keep the peaked central cluster of k-space from misclassification in the sparse component. Results: This algorithm was demonstrated to effectively remove k-space spikes from four data types under conditions generating spikes: (i) mouse heart T1 mapping, (ii) mouse heart cine imaging, (iii) human kidney diffusion tensor imaging (DTI) data, and (iv) human brain DTI data. Myocardial T1 values changed by 86.1 ± 171 ms following despiking, and fractional anisotropy values were recovered following despiking of DTI data. Conclusion: The RPCA despiking algorithm will be a valuable postprocessing method for retrospectively removing stripe artifacts without affecting the underlying signal of interest.