Dynamic imaging of organs can provide valuable information for radiotherapy applications or for studying physiology. However, time-resolved 5D (3D + respiratory + cardiac) MR imaging is time-consuming and recent approaches have reported acquisition times in the range of 15 min^1,2. To allow for acquisition times as low as 2 min, we propose a new method for 5D respiratory and cardiac motion-compensated image reconstruction, which is based on radial MR data with very high undersampling.Contrast-enhanced MR data covering the thorax of three free-breathing patients were acquired at 1.5 T (Magnetom Aera, Siemens Healthcare, Erlangen, Germany). Data acquisition and evaluation was in accordance with the local ethics committee and informed consent was obtained from each patient. We applied a vendor-provided radial stack-of-stars sequence with golden angle radial spacing and sagittal slice orientation: total acquisition time: 2.0 min, radial spokes per slice: 720, field-of-view: 385×385×300 mm³, voxel size: 1.5×1.5×5.0 mm³, TR/TE = 3.77/1.69 ms, 60 slices (60% slice resolution, 33% slice oversampling, 6/8 partial Fourier), flip angle: 12°, fat supression activated. Respiratory and cardiac motion signals used for self-gating were estimated from the k-space center for each acquired spoke. A bandpass filter was applied to distinguish between respiratory motion (filter range: 0.1 – 0.5 Hz) and cardiac motion (filter range: 0.5 – 2.5 Hz) and the signals were corrected for a baseline drift. Motion vector fields (MVFs) were estimated independently for respiratory and cardiac motion using a newly-developed algorithm, which alternates between image reconstruction and motion estimation. To increase robustness, deformable registrations were carried out between adjacent motion phases and regularized by cyclic constraints³. In a first step, respiratory MVFs were estimated (Fig. 1) using N_r = 20 overlapping respiratory motion bins with a width of Δr = 10% and neglecting cardiac motion (N_c = 1, Δc = 100%). In a second step cardiac MVFs were estimated (Fig. 2) assuming that respiratory motion in the end-exhale plateau can be neglected. Thus 25% of the raw data most consistent to end-exhale (N_r = 1, Δr = 25%, r = r_ref) divided into N_c = 10 overlapping cardiac motion bins with a width of Δc = 20% were used for estimation of cardiac MVFs. Having estimated respiratory and cardiac MVFs, a 5D MoCo image reconstruction was performed (Fig. 3). This was achieved by applying the estimated MVFs to a 5D double-gated gridding reconstruction (N_r = 20, Δr = 10%, N_c = 10, Δc = 20%, matrix size: 256×256×60×20×10, undersampling factor: 27.9). That means for any arbitrary combination of respiratory motion phase r and cardiac motion phase c, all other combinations (r’, c’) were warped onto (r, c) and averaged. Figure 3 shows the deformation path (r’, c’) → (r_ref, c’) → (r_ref, c) → (r, c) for one combination (r’, c’) as an example. Thus, 100% of the measured raw data were used for the reconstruction of each cardio-respiratory combination (r, c).Figure 4 shows representative reconstructions of 3D gridding, 5D double-gated gridding and 5D MoCo. 3D gridding reconstructions yielded motion blur caused by respiratory and cardiac motion. 5D double-gated gridding images exhibited high noise levels and severe streak artifacts, which arose from the strong radial undersampling as each combination (r, c) consisted of only 2% of the measured raw data. The here-proposed 5D MoCo reconstruction achieved high image sharpness because the images were fully compensated for organ motion. They further showed low noise and low streak artifact levels because each combination (r, c) was reconstructed from 100% of the measured raw data. As can be seen in Fig. 5, different combinations of respiratory and cardiac motion phases (r, c) were clearly resolved with identical image quality.In this study, we proposed a new method for 5D respiratory and cardiac motion compensation. The method enables time-resolved 5D MR imaging with acquisition times as short as 2 min without compromises in temporal or spatial resolution, size of field-of-view, noise and artifact level.