Yuchou Chang1, Dong Liang1, Leslie Ying1
We improve the convolution model in GRAPPA using a kernel approach. We map the acquired k-space data through a nonlinear transform to a high-dimensional space and then linearly combine the transformed data to estimate the missing k-space data. Both polynomial and Gaussian kernels are investigated for the nonlinear transform. The proposed kernel model characterizes the system noise in reconstruction more accurately. Experimental results using in vivo data demonstrate that the proposed kernel GRAPPA method can significantly suppress the noise in conventional GRAPPA without introducing artifacts.