4D cardiovascular MRI provides full assessment of the cardiovascular system and is particularly useful for treatment planning and post-surgical evaluation. However, such scan requires long acquisition time and motion remains a major issue. Traditional motion gating approach only accepts data acquired in certain periods, results in either low-efficiency (20-30%) scan or insufficient data for reconstruction. Some previously proposed regularization methods [1,2] exploited data redundancy along respiratory and cardiac dimension, reconstructing high quality motion resolved images under high acceleration factor, but come with the price of vastly increased reconstruction time and computation burden. A recently introduced computationally light reconstruction strategy [3], soft-gating, weights data consistency according to the degree of motion corruption, and has been applied in free-breathing DCE-MRI [4,5] with improved image quality. Given this attractive feature, we incorporated soft-gating in both respiratory and cardiac dimensions in this study, and evaluated the performance of the dual soft-gating approach against the regularization strategies on 4D cardiovascular MRI application. Data Acquisition: A 3D Cartesian sequence was developed to enable 1) k-t sampling scheme and 2) self-gating. Ky-Kz plane was sampled with (ky,kz) phase encodings grouped in spiral-like arms with rotating gold-angle ordering (Fig 1). For self-gating, each arm started at k-space center (ky=kz=0), generating a series of superior-inferior (SI) projections that were used for estimation of both respiratory and cardiac motions (Fig 1). Data Sorting and Soft-Gating: Each acquired data was sorted into 5 different datasets for comparison. 1) Only the 25% data that reside closest to end-expiration were selected and further binned into 9 cardiac phases; 2) Data were first sorted into 4 respiratory motion states spanning from end-expiration to end-inspiration, and further binned into 9 cardiac phases for each state; 3) Each readout was first weighted according to the respiratory distance of its corresponding position to the reference position (end-expiration) using a Gaussian kernel that centered at reference position (respiratory soft-gating, Fig 1). The kernel parameter was chosen such that only 25% of data’s weight is larger than 0.8. Weighted data were further binned into 9 cardiac phases; 4) Same weighted data from 3) were binned into 18 cardiac phases; 5) Same weighted data from 3) were further weighted according to their cardiac position using a flat-topped Gaussian kernel that slides through each cardiac phase (cardiac soft-gating, Fig 1). Dual soft-gated data were then binned into 18 cardiac phases. Image Reconstruction: Data were reconstructed with a compressed sensing and parallel imaging combined algorithm [6]: $$argmin_x ||W(Ax-y)||_2+\mu_1||R_1x||_1+\mu_2||R_2x||_1,$$ where $$$x$$$ are the recovered images, $$$A$$$ is a linear model that includes coil sensitivity, Fourier transform and under-sampling mask, $$$y$$$ is acquired k-space data, $$$R_1$$$ is the spatial wavelets, $$$R_2$$$ is the finite difference operator (TV) along either respiratory or cardiac dimension, $$$\mu_1$$$ and $$$\mu_2$$$ are regularization parameters, and $$$W$$$ is the weighting matrix. Specific reconstruction configuration for each dataset is listed in Table 1. Experiment Setup: 8 pediatric congenital heart disease (CHD) patients were scanned under general anesthesia and controlled ventilation. Each patient received a Ferumoxytol bolus injection (4 mg-Fe/kg). Sequence parameters included: TR/TE: 2.9/0.9ms, FA: 25°, isotropic resolution: 0.8-1.0mm without interpolation, scan time: 6.8±1.2 min. Image Evaluation: Subjective image quality scores (1-4 scale, non-diagnostic to excellent) in three different anatomical regions (outflow tract, ascending aorta and pulmonary artery, and coronaries) on RespHG, RegResp (only end-expiration state was chosen) and RespSG, and myocardial border sharpness and cardiac wall motion on RegCardiac and DualSG were visually assessed by an experienced radiologist.Fig.2 shows coronal overview and coronary images derived from RespHG, RegResp and RespSG. Good definition of intra-cardiac structures and coronary arteries were obtained with both respiratory resolved, regularized reconstruction (RegResp) and respiratory soft-gated reconstruction (RespSG), while reconstruction with hard gating (RespHG) failed to recover detailed structures due to high under-sampling factor. Fig. 3 shows selected cardiac phases of reformatted short-axis view reconstructed from RegCardaic and DualSG. Using either respiratory soft-gated, cardiac regularized reconstruction (RegCardiac) or dual soft-gated reconstruction (DualSG), myocardial border and cardiac wall motion can be clearly depicted and assessed. Table 2 summarizes reader’s scores and reconstruction time per cardiac phase for each method. No significant difference was found between motion-resolved regularized approaches and soft-gated approach, but reconstruction time was significantly reduced for soft-gated approach.In this initial study, we have demonstrated that incorporating soft-gating weights along both respiratory and cardiac dimensions in the iterative reconstruction can effectively suppress respiratory motion and resolve cardiac motion. The whole 4D dataset (18 cardiac phases) can be reconstructed in around 10 minutes, suggesting the feasibility of inline image reconstruction on the MR host computer.