University of Minnesota
University Relations
http://www.umn.edu/urelate
612-624-6868
myU OneStop


Go to unit's home.

Home | DTI | 2012–13 funded proposals | Pierre-Francois Van de Moortele, Zhi-Quan Luo, Thomas R. Henry

Initiatives in Digital Technology: 2012–13 Funded Proposals

Pierre-Francois Van de Moortele, Zhi-Quan Luo, Thomas R. Henry

Motion Artifact Correction in Very High Spatial Resolution MR Images based on Compressed-Sensing Iterative Reconstruction

In this project we propose a new MR image reconstruction approach to address physiological head motion induced artifacts in very high spatial resolution brain images acquired at Ultra High Magnetic Field (7 Tesla and higher).

Recently developed Ultra High Field human MR scanners (7 Tesla and higher) provide higher Signal to Noise Ratio (SNR) allowing for unprecedented spatial resolution in human brain imaging. Using very small, sub-millimetric voxels however greatly increases sensitivity to head motion occurring during the several minutes needed for the acquisition of one image. Resulting artifacts (ghosting, blurring), even when head movements are of small amplitude, can be sufficient to make these images unusable for detailed structural analysis. Existing methods to correct for these artifacts typically fail to properly correct the periphery of raw k-space data in very high spatial resolution because of a very low SNR in high spatial frequencies.

Here we exploit inherent sparse characteristics observed in MR images to develop an iterative compressed-sensing based reconstruction. A fundamental property of compressed-sensing reconstruction is to escape classical limits in conventional MRI reconstruction (Nyquist frequency sampling rule, SNR proportionality to the square root of acquisition sample size), allowing for using a small subset of acquisition samples to reconstruct final images with very limited SNR penalty. Our hypothesis is that using full k-space head motion corrupted acquisition it will be feasible to generate multiple compressed-sensing based reconstructed images using subsets of k-space samples. Because physiological motion does not affect kspace data uniformly, we anticipate that several of these compressed-sensing reconstructions will yield images unaltered by motion induced artifacts. These cleaner images will be retained and averaged while corrupted images will be discarded. We believe that a successfully implementation of this new approach can significantly expand the scope of high spatial resolution imaging at 7 Tesla.