The main challenges in fetal cardiac MRI are both the lack of an ECG signal to synchronize data acquisition with the fetus’ heart beat and spontaneous fetal motion. Recent advances led to the development of self-gating techniques which do not require an ECG signal(1,2). However, these particular approaches are unable to detect bulk motion of the fetus and suffer from sub-optimal spatial and temporal resolution. To overcome these shortcomings, we adapted a self-gating acquisition and reconstruction framework originally designed for adult coronary imaging(3). This self-gating framework integrates a golden-angle radial acquisition together with a k-t sparse SENSE reconstruction algorithm(3). In a first step, this reconstruction generates real-time cine images from which a cardiac gating signal is extracted. Simultaneously, data corrupted by fetal motion can be detected and excluded from the next reconstruction step. During the second reconstruction step, the image-derived gating signal is used to retrospectively regroup the k-space profiles from multiple cardiac cycles and cine frames are reconstructed with k-t-sparse SENSE(4). In this preliminary report, we investigate the feasibility of this new paradigm to visualize fetal cardiac cine images with high temporal and spatial resolution.Data acquisition: Cardiac fetal cine imaging was performed in 3 patients (27, 30 and 32 weeks of gestational age) using a 2D non-ECG triggered bSSFP radial golden-angle sequence on a 1.5T scanner (MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany). Sequence parameters included TE/TR=1.99/4.1ms, field-of-view=260x260mm², matrix size=256x256pixel, pixel size=1.0x1.0x4.0mm³, RF excitation angle=60°-70°, bandwidth=1028Hz/pixel, acquisition time=15-20s in a breath hold of the mother. ECG estimation: Real-time cines were reconstructed with a temporal resolution of 18.5ms (15 readouts per frame, 70% view sharing) using a k-t sparse SENSE model, with wavelet and total variation transformations for spatial and temporal regularizations, respectively (fig.1.ab). Data processing was performed with MATLAB (MathWorks, Natick, MA). A signal characterizing the periodic contraction of the heart was then extracted from the real-time cines by selecting a reference frame and computing the Pearson correlation through time as described in (3) (fig. 1.c). Fetal motion was identified by visual inspection of both the real-time cines and the correlation signals, and motion corrupted readouts were excluded from the subsequent reconstruction step. Self retro-gated reconstruction: Data from multiple cardiac cycles were retrospectively reordered into cardiac cine frames using the gating signal obtained from the correlation results (fig.1.d). The reconstruction of these retro-gated cine images was also performed with the k-t sparse SENSE algorithm with a temporal resolution of 12.5ms (bin width of 25ms, 50% view sharing). Targeted structures: To preliminarily test the proposed framework, we targeted three specific views that are of interest to assess the fetal heart and related anatomical structures: a mid-ventricular short-axis view, an axial three-vessel view, and a four-chamber view.Data acquisition and real-time cine reconstruction were successfully performed in all patients. Despite the absence of magnetization preparation due to the continuous acquisition, the bSSFP sequence led to a good contrast between blood and myocardium. Fig.2 shows an example of real-time cines for the different views obtained in the same subject. Although compression and streaking artifacts are visible, the quality was sufficient to extract a periodic cardiac gating signal and to detect fetal movement (fig.2.right). When spontaneous fetal bulk motion (translation, rotation and limb movement) was identified, corresponding k-space profiles were excluded and were not used for the final cine reconstruction step. It was empirically observed that a window of 4 seconds without fetal motion was sufficient to reconstruct self retro-gated cine images with good quality. The corresponding retro-gated reconstructions are shown in fig.3. Ventricles and papillary muscles are visible in both short-axis and four-chamber views (fig.3.left&right). The right ventricular outflow track can be observed in axial views (fig.3.mid): ascending and descending aorta, pulmonary arteries, the superior vena cava and the ductus arteriosus (fig.4). In some specific views, the valves were also successfully visualized (fig.5).We present a novel framework that uniquely enables the reconstruction of self retro-gated fetal cardiac cine images with high temporal and spatial resolution and without any external triggering device. Real-time imaging enabled by the combined golden-angle acquisition and k-t sparse SENSE reconstruction does not only support the extraction of a gating cardiac signal, but also allows for the flexible detection and rejection of motion corrupted k-space data. The gating signal is of critical importance as k-space segments acquired from different cardiac cycles can freely be regrouped and combined to reconstruct high-resolution cine images. Based on the successful preliminary in vivo results, a quantitative Gold Standard comparison with fetal echocardiography is now warranted.