A dual-polarity grappa kernel for the robust reconstruction of accelerated EPI data

The quality of high-resolution Echo Planar images of the human brain has improved greatly in recent years, enabled by novel multi-channel receiver coil arrays and parallel imaging. However, in regions with local field inhomogeneity, EPI artifacts limit which parts of the brain can be imaged successfully. In this work, we present evidence that certain image artifacts can be attributed to nonlinear phase errors that are present in regions of local susceptibility gradients and certain coil array elements. Because these phase errors cannot be corrected with conventional Nyquist ghost correction, we propose a new method that integrates ghost correction with parallel imaging reconstruction. The proposed Dual-Polarity GRAPPA method operates directly on raw EPI data to estimate k-space data from the under-sampled acquisition while simultaneously correcting inherent EPI phase errors. We present examples of this method successfully removing strong phase-error artifacts in high-resolution 7T EPI data.