Kernel ridge regression-based TV regularization for motion correction of dynamic MRI

Abstract

Motion occurring during dynamic magnetic resonance imaging (dMRI) acquisition is a major factor of image quality degradation. A study adopted the temporal-based total variation (TV) constraint to correct the motion corrupted image has aroused wide concern. However, such motion correction scheme depends on the reference frame. If the quality of the reference frame is bad or is corrupted by noise in its acquisition, it will adversely affect subsequent reconstruction results. To address above issue, the kernel ridge regression (KRR)-based TV reconstruction framework is proposed, where the corrupted reference image can be corrected by exploiting the KRR technique. In addition, the gradient-domain-guided filtering is incorporated in the proposed model to preserve the edge features of MR image. Experimental results show that the proposed KRR-TV method can produce MR images with better objective and visual qualities.