Anja Brau1, Peng Lai1, Srihari
Narasimhan2, Babu Narayanan3, Vijaya Saradhi2
1Global Applied Science Laboratory,
GE Healthcare, Menlo Park, CA, United States; 2Computing &
Decision Sciences Lab, GE Global Research, Bangalore, India; 3Medical
Image Analysis Lab, GE Global Research, Bangalore, India
Computation of kernel weights as part of calibration in Compressed Sensing and Parallel Imaging algorithms like ESPIRiT and L1-SPIRiT is computationally expensive, especially for high channel count reconstructions. The weights are computed by obtaining a least squares fit for predicting target points in the calibration region using a set of source points in their neighborhood. The number of points in the neighborhood and their distance from the target point define quality of fit and computational complexity of the solver. We show that an optimally shaped neighborhood can give a significant improvement in the computational performance without sacrificing the image quality.