Fetal Cardiac Magnetic Resonance Imaging (MRI) has recently shown promise as a viable tool for evaluating the fetal circulation clinically [1]. For evaluating cardiac function, one of the main issues when performing CINE cardiac fetal imaging is gating the acquisition to the fetal ECG, which has low voltages and is contaminated by the maternal ECG. Solutions for retrospectively sorting the acquired data into the proper cardiac phases have been proposed using routine clinical pulse sequences [2]. For non-fetal MRI, self-gated approaches have also been proposed [3]. The aim of this study was to demonstrate the feasibility of CINE fetal MRI within a breath-hold with self-gated retrospective binning applied to continuous golden angle radial sampled data and reconstructed with iGRASP (iterative Golden-angle RAdial Sparse Parallel) [4].

Institutional Review Board approval and written informed consent from the study participants were obtained. Exams were performed on a 1.5T Magnetom Aera system with two 18-channel phased-array coils (Siemens Healthcare GmbH, Erlangen, Germany). Short-axis fetal cardiac data were acquired using a radial trajectory with a constant azimuthal increment of small golden angle of order 7 (ψ₇ in [5]) corresponding to a rotational angle of approximately 23.1^o. This radial angle increment does not cause strong eddy currents, which are a known source of reconstruction artifacts. Thus, no gradient delay correction was performed. Balanced Steady State Free Precession images were acquired with the following parameters: TE/TR=1.88/3.9 ms, flip angle 60^ο, pixel size 0.7 x 0.7 mm², slice thickness 5 mm. A total of 3800 radial spokes were acquired within a breath-hold of just under 15s. The cardiac triggering signal required for the binning stage was estimated using the cardiac motion as it was captured by the magnitude of the center of each k-space spoke. The cardiac triggering signal extraction was based on Principal Component Analysis (PCA) [3,6]. Through PCA the most common signal variation mode among all coils was determined. The principal component representing fetal cardiac motion was then selected as the one with the highest peak in the frequency range of 2–3 Hz, considering normal fetal heart rates being 110-160 beats per minute. The number of cardiac phases depended on the number of available cardiac cycles, with typical values ranging between 18-24 phases. The final images were obtained by using iGRASP reconstruction, formulated as the optimization problem:

d* = arg min { || F S d - m ||²₂ + λ || T d||₁} (1)

where d was the fetal cardiac image series to be reconstructed in x-y-t space, T was the temporal total-variation operator, m = [m₁,…,m_c] were the acquired multi-coil radial k-space data with c coils, F was the NUFFT gridding operator defined on the radial acquisition pattern, S = [S₁,…,S_c] were the coil sensitivity maps and λ was the regularization parameter. Typical λ values were in the range of 0.01 to 0.02. The coil sensitivity maps S were estimated by means of sum of squares. The optimization problem in Eq. (1) was solved by the nonlinear conjugate gradient implemented in MATLAB (The Mathworks, MA, USA) [4]. The reconstruction time was approximately 20 minutes per slice on a medium performance laptop.

Fetal cardiac images acquired from a normal volunteer during the third trimester are shown in the figures. Figure 1 shows a mid-ventricular short-axis view of the heart at end diastole. In this image, the entire fetus and the placenta are visualized with good image quality. In Figure 2 the twenty reconstructed cardiac phases are shown cropped and magnified in order to better appreciate image quality for potential delineations.In this study, we demonstrated that self-gated fetal cardiac MRI is possible with good image quality. The method was based on iGRASP, which combines sparse sampling with parallel imaging. Eddy currents that are generated due to gradient switching associated with the golden angle were reduced by using the recently proposed small angle increments. The cardiac trigger signal was extracted through multi-coil PCA. In the future these methods can be expanded and refined to allow for free breathing and sensitization to flow thus promoting the use of fetal imaging in the clinic. For clinical fetal cardiac MRI to become generally feasible, image reconstruction speed needs to be increased by an optimized computational implementation.