Array Compression for 3 D Cartesian Sampling

Introduction: Array compression (AC) [1,2] is a technique to reduce data size and reconstruction computation for large coil arrays. The original multi-channel data can be compressed, by a linear combination in k-space, into a few virtual channels, on which the reconstruction is performed. Among different AC methods, data-driven array compression [2] has the advantages of not requiring coil sensitivity measurement and having fast calculation of the compression matrix, and it has been demonstrated for fast reconstruction with 2D or multi-slice data acquisition. However, simply extending data-driven AC to 3D datasets will lead to non-optimal compression (signal loss), and is problematic with autocalibrating methods using 3D synthesis kernels, such as SPIR-iT [3] and ARC [4]. In this work, an optimal data-driven AC for 3D Cartesian sampling is proposed.