Jonathan R. Polimeni1, Kawin Setsompop1, Christina Triantafyllou1, Lawrence L. Wald1, 2
1Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Harvard Medical School, Massachusetts General Hospital, Charlestown, MA, United States; 2Harvard-MIT Division of Health Sciences and Technology , Massachusetts Institute of Technology, Cambridge, MA, United States
Here we introduce a coil combination method that accounts for the spatially-varying noise covariance matrixpresent in both accelerated imaging and in fMRI time-seriesto boost SNR. Because this matrix must be inverted and its rank varies spatially, a per-voxel regularization is required to yield SNR gains. We demonstrate an improvement in image SNR for GRAPPA-reconstructed phantom data and a tissue-specific improvement in time-series SNR for resting-state fMRI data. These results can also provide insight into the non-stationarity of physiological noise covariance across coil channels.