Image Denoising Exploiting Sparsity & Low Rank Approximation (DSLR) in Slide Encoding for Metal Artifact Correction

Sangcheon Choi¹, Hahnsung Kim², Jaeseok Park¹

Metal-induced field inhomogeneity is one of the major concerns in magnetic resonance imaging near metallic implants. Slice encoding for metal artifact correction (SEMAC) is an effective way to correct severe metal artifacts by employing additional z-phase encoding steps for each excited slice against metal-induced field inhomogeneity and view angle tilting (VAT). Despite the advantages of metal artifact correction, since noisy resolved pixels are included in image reconstruction, SEMAC suffers from noise amplification. SEMAC with noise reduction , which employs a two-step approach (rank-1 approximation along the coil dimension followed by soft thresholding in the slice direction), does not consider noise correlation of coils and results in a direct tradeoff between image accuracy and de-noising. Thus, to further expedite noise reduction in SEMAC, in this work we develop a novel image de-noising algorithm that exploits 1) low-rank approximation using strong correlation of pixels (x-z) in the slice direction (t), 2) Best Linear Unbiased Estimator (BLUE) image combination in the coil direction with noise correlation, and 3) recovery of distorted slice profile using the sparsity of signals in the slice direction with orthogonal matching pursuit (OMP).