Jun Miao1, Wilbur C. K. Wong1,2, Donglai Huo3, David L. Wilson1,4
1Biomedical Engineering, Case Western
Reserve University, Cleveland, OH, USA; 2Computer Science and
Engineering, The Hong Kong University of Science and Technology, Kowloon, Hong
Kong; 3Keller Center for Imaging Innovation, St. Joseph Hospital and
Medical Center, Phoenix, AZ, USA; 4Radiology, University Hospitals
of Cleveland, Cleveland, OH, USA
In Parallel MR k-space reconstruction, most algorithms like GRAPPA make a common assumption that unsampled signals can be explained linearly and globally by sampled one. However, this global assumption is invalid theoretically. To better model the relationship between sampled and unsampled signals, we locally fit the linear function within clusters on a mathematically sound framework, Geographically Weighted Regression. Simulated and acquired MR data with different image contents and acquisition schemes including MR tagging data were tested. Results showed that comprehensive GRAPPA can significantly and robustly improve the image quality in parallel MR imaging.