Free-breathing cardiac imaging is a desirable tool for clinical routine as well as in the context of MR-guided PET motion correction [1, 2]. Due to the desirably high number of cardiac and respiratory states (usually 20 cardiac phases and 6-8 respiratory gates) complete k-space coverage is not feasible in a reasonable scan time. In this work, we combine a radial golden angle sampling scheme with piecewise rigid motion correction to achieve high sampling efficiency. We demonstrate, that this enables the acquisition of images which are suitable for precise respiratory and cardiac motion estimation in 3 minutes or less.

MR data are acquired under free-breathing for several minutes using a fat-suppressed (spectral IR) axial spoiled gradient echo sequence (BW = 83 kHz, TE/TR = 1.4/3.5 ms, α = 5°, 256 points/readout, 64 slices, FOV = 500x500x256 mm³) and an 18-channel anterior/posterior body coil array on a Signa PET/MR scanner (GE Healthcare). A 3D k-space is filled in a golden angle stack-of-stars fashion [3]. All z-encoding steps (Cartesian) for a given projection angle are acquired before rotating the projection angle by 111.25°.

Image data is reconstructed by applying an IFFT in the fully-sampled z-direction and subsequent in-plane CG-SENSE [4] reconstruction (5 iterations). The radial IFTs and FTs in the CG-loop are carried out by the NUFFT library [5]. Optionally, the FFT and IFFT in z-direction can be moved inside the CG loop for additional SENSE acceleration along z.

During the scan, physiological data (respiratory belt and peripheral pulse gating (PPG) sensor) are recorded. A self-navigation signal is extracted from the projection data and used to synchronize the acquisition to the recorded physiologic data.

The radial golden-angle trajectory allows for flexible retrospective gating of the rawdata into respiratory bins (gating based on self-navigation amplitude) and cardiac phases (based on normalized position in the cardiac cycle, defined as the time between two subsequent peaks in the PPG signal). Each set of z-encodings is treated as a single time-point, thus the gating causes no under-sampling along z. Combinations of both, cardiac and respiratory gating can be used; this, however, can lead to severe under-sampling, especially in inspiratory gates. Instead, we propose a two-step approach, similar to [6].

In the first step, respiratory-gated images are reconstructed while disregarding the cardiac phase, thus sufficient data for high-quality reconstruction is available. From these images, the respiratory motion is estimated using non-rigid registration. The obtained motion fields are clustered into regions of similar motion using fuzzy-c-means (see Figure 1). Thus, the motion in each region can be approximated by a rigid mean motion, which can be corrected in k-space by phase ramps. In the second step, instead of using dual-gating to obtain motion-free multi-phase cardiac images, only cardiac gating and appropriate phase correction for each region is done to obtain N_phase x N_regions k-spaces. Compared to dual-gating, each of these k-spaces contains much more data, which increases sampling efficiency. After individual reconstruction, the images for a given cardiac phase are combined according to the regions determined by the clustering.

Figure 2 shows the CG-SENSE reconstructions of different k-spaces in expiration (top) and inspiration (bottom), created from an 8 min scan. The first 5 images in each row show images reconstructed from k-spaces, which have been corrected for the mean rigid motion in the 5 different image regions, obtained by the clustering. Note that each of these images consists of sharp and blurry parts. The sixth image is the region-wise combination, combining the sharp parts of all these images. For reference, the last image shows a dual-gated image. Because of the longer duration of the expiratory phase of the respiratory cycle and the long scan time, the expiratory gate carries enough information to achieve good image quality even with dual-gating (top). The inspiratory gate, however, suffers from a high under-sampling rate due to dual-gating (bottom). Figure 3 (animated) shows 20 cardiac frames, reconstructed using the described method and data only from only the first 3 minutes of the scan. Note, that all images were acquired in axial orientation and are shown through-plane.

We propose a method for free-breathing abdominal/thoracic imaging with a high sampling efficiency using piecewise rigid motion correction. The high sampling efficiency can be used to either improve image quality or enable cardiac gating in arbitrary respiratory states (i.e. not limited to end-expiration) at moderate scan times, or to significantly shorten the required scan time needed to estimate motion fields for MR-guided PET motion correction.