Coronary MR angiography (MRA) provides a non-invasive tool to assess coronary lumen integrity. Fat suppression is required to visualize the coronary arteries, as they are embedded in epicardial fat. It has been shown that cardiac fat may provide important diagnostic information¹ and thus fat visualization may be desirable. Dixon-based methods yield both water and fat images from a multi-echo (in-phase and out-of-phase) acquisition. These methods have shown superior fat suppression when compared with spectral methods² and have recently been used to produce attenuation maps for simultaneous PET-MR imaging³. Respiratory motion is a major problem in whole-heart 3D water/fat Dixon MRI and leads to prolonged scan time in respiratory gated acquisitions. Moreover, retrospective translational correction of respiratory motion may be of limited use as it may introduce ghosting artifacts from static fat tissue present in the in-phase and out-of-phase images. Here, we propose a novel nonrigid motion correction framework for free-breathing 3D whole-heart water/fat Dixon coronary MRA that uses all the acquired data for reconstruction (100% scan efficiency) and therefore has a predictable scan time. The proposed approach was compared with a 2D translational correction (TC) and no motion correction (NMC) in four subjects.Data was acquired using an interleaved scanning framework⁴ with two scans acquired in an alternated fashion: a 3D segmented Cartesian whole-heart dual-echo Dixon and a 2D single shot golden-radial (GR) coronal image navigator (2D iNAV) (Fig. 1a). Motion correction was performed in two steps: beat-to-beat translational correction and bin-to-bin nonrigid correction (Fig. 1b). iNAVs were reconstructed and rigidly registered to estimate superior-inferior (SI) and right-left (RL) translational motion, allowing beat-to-beat translational correction of each echo image. A respiratory signal was derived from the SI component and data was binned according to respiratory position. Out-of-phase bins were reconstructed with soft-gated⁵ iterative SENSE⁶ and registered to estimate 3D nonrigid motion. The estimated motion fields were used in a Total Variation regularized⁷ General Matrix Description⁸ (TVGMD) reconstruction for each echo image by solving: $$$\widehat{I} = arg min_I\left\{||\sum_b \bf A_b \bf F \bf S \bf U_b \bf I - \bf K||_2^2 + \lambda TV\right\}$$$, where $$$\widehat{I}$$$ is the motion corrected echo image, $$$K$$$ the corresponding translation corrected k-space, $$$A_b$$$ the sampling matrix for bin b, $$$F$$$ the Fourier transform, $$$S$$$ the coil sensitivities, $$$U_b$$$ the nonrigid motion fields, $$$TV$$$ the spatial total variation operator and λ the regularization parameter. Coronary water/fat images were obtained from the motion corrected echo images as described in Berglund et al². Four healthy subjects were scanned free-breathing on a 3T Philips scanner using a 32-channel coil. Dual-echo Dixon data was acquired with an ECG-triggered 3D Cartesian spoiled gradient echo (1x1 mm in-plane resolution, 2 mm slice thickness, 300x300x80 mm FOV, TR/TE1/TE2 = 4.0/1.44/2.6 ms, flip angle = 20°, T2prep duration = 40 ms) interleaved with a 2D golden radial iNAV spoiled gradient echo sequence (4x4 mm in-plane resolution, 25mm slice thickness, 300x300 FOV, TR/TE = 2.4/1.07 ms, flip angle = 5°). Non-motion correction (NMC), 2D translational correction (TC) and the proposed translation plus nonrigid (TC+TVGMD) reconstructions were obtained from these data.Coronal images for two subjects with the proposed TC+TVGMD, TC and NMC for echoes 1 (in-phase) and 2 (out-of-phase) are shown in Fig.2 and Fig.3, respectively. Motion artifacts are visible for NMC in both echoes. TC brings into focus structures inside the heart (arrows in Fig.2 and Fig.3). However, this global k-space translational correction will be incorrect for static fat signal present in echoes 1 and 2, introducing significant aliasing. The proposed approach combines inter-bin (i.e. small) translational correction with intra-bin nonrigid correction. Thus, both moving and static tissues are adequately motion corrected with GMD and residual incoherent aliasing suppressed with total variation regularization. Water and fat images reformatted along the right and left coronary arteries are shown in Fig.4 and Fig.5 for the proposed TC+TVGMD, TC and NMC. Motion artifacts are visible in NMC, as in the individual echo images. Partial motion correction is achieved by TC, but considerable aliasing is also introduced (arrows in Fig.4 and Fig.5). The proposed TC+TVGMD accurately corrects the nonrigid motion of the heart and minimises ghosting artefacts from static sources.A novel framework for 3D water/fat Dixon coronary MRA with nonrigid motion correction has been proposed. The proposed approach corrects nonrigid motion of the heart, thereby minimising ghosting from static fat tissue and solving a fundamental limitation of translational correction. The proposed approach allows ~100% scan efficiency, enabling shorter and predictable scan times. Future work will compare the proposed approach with standard navigator-gated acquisition in additional subjects.