kt SPIRiT for Ultra-Fast Cardiac Cine Imaging with Prospective or Retrospective Cardiac Gating

Introduction: Cardiac cine imaging requires ultra-fast acquisition of high spatiotemporal resolution images within a short breathhold, which is difficult for conventional parallel imaging (PI). Recent progress has increased the achievable acceleration mainly from two perspectives. First, various kt methods [1,2,3] have proven capable of providing reasonable image quality at high acceleration factors by additionally exploiting temporal correlations in dynamic imaging. Second, a few approaches combining PI and compressed sensing (CS) , such as L1SPIRiT , have shown potential in acceleration beyond coil capability. The purpose of this work was to develop a reconstruction algorithm for highly-accelerated cardiac imaging with prospective (PG) or retrospective (RG) gating based on coil sensitivity and spatiotemporal sparsity in dynamic medical images. Theory: kt SPARSE [7] first introduced CS to dynamic imaging, using Fourier transform (F) to sparsify signal along the short time dimension (t). However, Fourier transform is not an efficient sparse transform and requires periodic motion in a complete cardiac cycle which is not available with PG. Alternatively, we construct a hybrid image in [x, t·y] (Fig.1.c) from cine image series in [x, y, t] (Fig.1.b), with the synthesized dimension t·y obtained by concatenating all t and y points at the same x location using a “rewinding” ordering scheme (Fig.1.a). Owing to continuous cardiac motion, this hybrid cardiac image exhibits smooth signal transition in (x, t·y) for both PG and RG and is highly compressible. For accelerated imaging, incoherent artifacts in (y, t) originating from random kt undersampling manifests now as incoherent artifacts along t·y, which could be effectively resolved by wavelet CS (Fig. 1.d).