Robust motion correction in CEST imaging exploiting low-rank approximation of the z-spectrum.

PURPOSE: To introduce and evaluate an image registration technique for robust quantification of CEST acquisitions corrupted by motion.
METHODS: The proposed iterative algorithm exploits a low-rank approximation of the z-spectrum (LRAZ), to gradually separate the contrast variation due to saturation at different off-resonance frequencies and accompanying motion. This registration method was first tested in a creatine CEST analysis of a phantom with simulated rigid motion. Subsequently, creatine CEST acquisitions in the human thigh during exercise were exemplarily corrected.
RESULTS: The z-spectrum obtained by applying LRAZ to the corrupted phantom series exhibited a normalized RMS error with respect to the noncorrupted gold standard series of less than 4%. The corresponding creatine map resulting from an asymmetry analysis of the registered data showed only little difference with regard to the noncorrupted determination, too. A comparable performance was observed exploiting LRAZ for the correction of nonrigid motion within the dynamic CEST acquisitions in skeletal muscles. While for the phantom simulations, high-quality registration was also possible by using a single reference image for the whole series and mutual information as similarity metric, this conventional approach resulted in inappropriate correction of the more complicated motion of the human thigh.
CONCLUSION: The newly introduced method allows for a robust registration of CEST image series, which are corrupted by rigid and nonrigid motion of the investigated organ. The technique therefore improves the diagnostic value in various applications of CEST.