Highly-Accelerated Cardiac Cine MR Imaging using kats ARC (Autocalibrating Reconstruction for Cartesian Sampling with k- & Adaptive-t-Space Data Synthesis)

Introduction: Cardiac cine MRI (CMRI) is a routine exam for studying myocardium functionality. High spatiotemporal-resolution CMRI has been challenging, limited by breathhold capability. Recently, various k-t techniques [1,2] have been developed to achieve acceleration factors (AF) ≥3 in dynamic imaging by incorporating temporal correlations. Among these approaches, k-t GRAPPA possesses the advantages of self-calibration but suffers from significant temporal blurring with high AF’s. This study was aimed at developing a new ARC (Autocalibrating Reconstruction for Cartesian Sampling [3]) -based technique with k& adaptive-t-space data synthesis (kats ARC) and investigate its effectiveness for fast CMRI. Theory & Methods: Conventional k-t GRAPPA performs undersampling in k-t space with linear k-space shifting along time. It exploits temporal correlations by including the nearest two temporal neighbors with the same phase-encoding (PE) to recover a missing k-space line. Such reconstruction uses k-space data faraway in time from the target line for data synthesis and results in interpolation in a large temporal window covering 2·AF-1 frames at all cardiac phases. This greatly reduces its temporal fidelity, which is critical for cardiac motion delineation. For comparison, k-t ARC was implemented in this study by combining the same k-t scheme with ARC. Generally, cardiac motion induces heart deformation proportional to time distance. Compared to the k-space lines with the same PE, a line with similar PE and a shorter time distance should correlate with the target k-space line with higher temporal fidelity. Furthermore, the degree of temporal correlations is highly cardiac phasedependent [4]. Based on these rules, our new method utilizes the concept of k-t-space synthesis with modified acquisition and reconstruction. Similar to ARC, data were first converted to hybrid space by Fourier transforming along kx [5]. As demonstrated in Fig. 1, kats ARC data synthesis was performed using the nearest two neighbors along PE at the same phase (tn) and the sampled neighbor with the closest PE at each adjacent phase covered by a phase-specific time window (Wn). To optimize temporal correlation and fidelity, data were acquired in a nonlinear time-alternating fashion with minimal overall PE difference for data synthesis between immediately adjacent phases. Wn was determined for each phase based on local motion derived from autocalibration signals (ACS). A low-resolution image could be obtained from fully sampled ACS at each phase, roughly depicting cardiac motion. Deviation (DEV) between two ACS images was calculated to estimate the consistency between phases and was equivalently performed by direct k-space subtraction: