Three-dimensional whole-heart MRA is a promising tool for comprehensive coronary characterization and detection of luminal narrowings. Free-running self-navigated techniques have recently been introduced (1-2) to enable the simultaneous time resolved evaluation of both cardiac function and coronary anatomy throughout the entire cardiac cycle. These techniques, however, intrinsically suffer from streaking artifacts due to undersampling and from sensitivity to respiratory motion. In order to address these hurdles, 5D (x-y-z-cardiac-respiration) XD-GRASP (eXtra-Dimensional Golden-angle RAdial Sparse Parallel MRI) has successfully been applied to reconstruct cardiac and respiratory motion-resolved 3D volumes in healthy adult subjects (3). In this study, we extended our investigations to evaluate the potential of the 5D XD-GRASP approach in patients with confirmed coronary artery disease.Data acquisition: Data acquisition was performed in N=4 patients with myocardial infarction using a prototype free-running (non ECG-triggered) bSSFP 3D golden-angle radial trajectory (2) on a 1.5T scanner (MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany). An additional superior-inferior (SI) projection was acquired at the beginning of each data segment (4), and was used for respiratory motion detection (Figure 1a). Data were acquired during free-breathing (acquisition time ≈14 minutes) and during slow infusion of contrast agent (Gadovist, 0.03 ml/sec, 0.1 ml/kg). Imaging parameters of the free-running, fat-saturated acquisition were as follows: TR/TE 3.1/1.56ms, FoV (220mm)³, voxel size (1.15mm)³, matrix dimension (192)³, radiofrequency excitation angle 90°, data segments 5’749, lines per segment 22, total acquired lines 126’478. Image reconstruction: A 1D respiratory motion signal (Figure 1b) was extracted using the self-navigated respiratory motion detection algorithm described in (5) and images were reconstructed using two different approaches: a) For the first approach, hereafter referred to as “4D reconstruction”, the 1D respiratory signal was used to correct for respiratory motion in the SI direction (5). Then, the recorded ECG signal was used to sort the data into 15-20 3D cine frames (duration 100ms, view sharing 80%, ≈12’000 radial lines per frame). Finally, images were reconstructed with a standard re-gridding algorithm (2). b) For the second approach (“5D XD-GRASP”), the 1D respiratory and ECG signals were used to sort data into 6 different respiratory phases or bins (Figure 1c), ranging from end-expiration to end-inspiration and into ~20 cardiac phases (80ms duration, 50% view sharing), respectively. The cardiac- and respiratory-resolved images with dimensions of 192×192×192×20×6 were then reconstructed by solving (3): $$\underset{m}{\text{arg min}}{\parallel{F\cdot{C}\cdot{m}-s}\parallel }_2^2 + \lambda_{1}\parallel{D_{1}m} \parallel_{1}+ \lambda_{2}\parallel{D_{2}m} \parallel_{2}$$ where F represents the NUFFT operator, C the coil sensitivity maps, m the 5D image set to be reconstructed, s the sorted radial k-space data, D₁ and D₂ the finite difference operators applied along the cardiac and respiratory dimension, respectively, and λ₁ = 0.02-0.04 and λ₂ = 0.02-0.04 the regularization parameters, which were empirically selected after normalizing the signal intensity from 0-1. Image quality assessment: A diastolic 3D cine frame was selected from the 4D reconstruction and compared with a diastolic 3D cine frame selected from an end-expiratory position of the images reconstructed with 5D XD-GRASP. Two experienced reviewers scored the reconstructed volumes with grades ranging from 0 (non visible) to 4 (sharply defined), by considering overall image quality, sharpness of the myocardium, and coronary delineation. Subsequently, visible coronary vessel length was quantified by using the software described in (6).Data acquisition and reconstruction was successful in all cases; 5D XD-GRASP provided 3D volumes resolved for both cardiac and respiratory motion (Figure 2). All the quantified results, as summarized in Table 1, confirmed the improvement in image quality, sharpness of the myocardium, and coronary visualization when the 5D XD-GRASP approach was used for image reconstruction, with respect to the 4D reconstruction.5D XD-GRASP has been successfully applied in a clinical setting to a small cohort of patients. A major advantage of the 5D XD-GRASP approach is that it reconstructs images in different cardiac and respiratory phases, and does not require any model for motion correction as is the case for the 4D reconstruction. While 1D SI motion correction performs well for some subjects and specific locations of the anatomy, it only corrects for respiratory motion in the SI direction and does not account for the more complex 3D respiratory motion of the heart. Furthermore, 5D XD-GRASP enables the reconstruction of cardiac- and respiratory-motion resolved datasets at the same time (Figure 2). Investigation in a larger number of patients is now warranted, in order to further corroborate our preliminary results, and ventricular volume, ejection fraction and mass will have to be quantified using a 2D gold standard breath-held comparison.