Contrast-enhanced 3D magnetic resonance angiography (CE-MRA) is routinely used to diagnose thoracic aortic disease. Without ECG gating, high spatial resolution comes at the expense of motion-induced blurring of aortic wall. Alternatively, the longer acquisition times with ECG gating to suppress motion-induced blurring typically require a sacrifice in spatial resolution. Currently, neither approach has emerged as a clear winner. Both approaches require a gadolinium-based contrast agent, which is contraindicated for patients with impaired renal function. Non-contrast MRA is an alternative method that could be used for all patients and can be rescanned immediately as needed. A previous approach reported GRAPPA acceleration factor (R) = 6 using an ECG-gated, T2-prepared, and fat saturated NC-MRA sequence with a 32-element cardiac coil array (1). For this study, we developed a 6-fold accelerated ECG-gated NC-MRA, with timing as shown in Figure 1, using a standard coil array, by a combination of 3D radial stack of stars with tiny golden angles (2,3) and compressed sensing (CS). We compare its performance versus ECG-gated CE-MRA. Informed consent was acquired for 8 subjects who were scheduled to undergo clinical cardiovascular MRI scans (4 female, 4 male, mean age = 57+/-16). All imaging was performed on a 1.5T scanner (Siemens, Avanto). Figure 1 shows a diagram of the stack-of-stars trajectory. Non-contrast ECG-gated MRA was performed prior to ECG-gated, breath-hold CE-MRA. Imaging parameters for NC-MRA were TR/TE 438.7/2.2 ms FA 100°, FOV 350mmx350mmx76.8-83.2mm, voxel size 1.3mmx1.3mmx3.2mm (interpolated to 1.6mm), 48 radial rays acquired per slice, T2 prep duration 50 ms. Coil sensitivity maps were self-calibrated using the densely acquired center of k-space (central 32x32 lines) using a previously described method (4,5). Image reconstruction with 50 iterations was performed offline using non-local means (NLM) as a constraint (6). Prior to denoising k-space data were normalized to maximum value. 3D NLM was performed with comparison and search windows of 3x3x3 and 5x5x5 respectively, and data fidelity and NLM weights of 0.4 and 0.7 respectively. These weights were determined empirically based on visual inspection of training data. Vessel dimensions were measured at 3 locations as shown in Figure 2 (see arrows). We also measured normalized signal difference [(vessel-background)/vessel] as a surrogate for contrast-to-noise ratio (CNR) at these locations. We note that CNR is not easily measurable from GRAPPA and CS data. Paired sample t-test was performed to compare mean vessel dimensions and normalized signal difference between CE-MRA and NC-MRA.Figure 2 shows representative multi-planar reconstructed images for CE-MRA and NC-MRA. For the 8 subjects studied there was no significant difference in vessel dimensions between the CE-MRA and NC-MRA tests (p>0.08). Corresponding CE-MRA and NC-MRA for another patient are shown in Figure 3. In all 8 patients, NC-MRA produced images that are comparable to CE-MRA. There was no significant difference between the two MRA methods at the aortic root and aortic arch (p>0.1), but there was statistically significant difference at the descending aorta (p<0.05). Acquisition times for the CE-MRA and NC-MRA scans were 22+/- 4 and 20.1+/- 3.5 sec respectively.We have demonstrated feasibility of NC-MRA using a combination of 3D radial stack of stars and CS. In 8 patients, NC-MRA produced clinically acceptable image quality with aortic dimensions that are not different from those measured from ECG-gated CE MRA. Future studies include clinical evaluation in patients with aortic diseases.