2D adaptive coil sensitivity estimation for dynamic parallel MRI reconstruction

Incorporation of parallel imaging technique in dynamic magnetic resonance imaging process accelerates significantly the imaging speed. The acceleration is achieved by acquiring the k-space data sparsely in each frame. To unwrap the aliased images caused by the k-space under-sampling, the reconstruction algorithms need the local sensitivity variation, the sensitivity map, of the receiving coils. Current estimation methods are unable to provide the true sensitivity map, and the calibration errors in the estimated sensitivity map can dramatically increase the image artifacts. This paper introduces a novel two-dimensional k-t modeling approach to improve the estimation of sensitivity map. A 2D model is fitted to each k-t plane and is updated for each estimated data point. The model is used to estimate the unacquired k-space data in each frame. A full k-t plane is composed using the estimated and acquired datasets and is used to obtain improved estimation of sensitivity map. Experiments on real dynamic parallel MRI data demonstrate that the proposed method significantly improve the quality of reconstructed MR images.