Hybrid PET/MR scanners offer the opportunity to obtain spatially and temporally co-registered images with the advantages of both modalities including molecular function and metabolism from PET and anatomical structure and tissue function from MRI. Hybrid PET/MR has significant potential in cardiac applications. Recently, 18F-fluoride PET/CT has been shown to identify micro-calcification in atherosclerotic plaques associated with recent myocardial infarction[1]. PET/MR offers the ability to investigate active coronary atherosclerosis while reducing the radiation dose compared to PET/CT allowing repeated and longitudinal studies. PET image reconstruction requires knowledge of the PET-photon attenuation of the object in order to produce accurate images of PET tracer activity. The current standard approach for MR-based attenuation correction (MRAC) is breath-hold volumetric imaging to freeze motion of the chest and abdomen. However, for imaging the heart, alignment of anatomy during PET data collection and attenuation measurement is crucial. In this work, we propose mapping attenuation using a free-breathing golden-angle radial gradient echo sequence and compare the PET images produced with this novel approach and the standard breath-hold approach.Six patients with diagnosed cardiovascular disease or risk factors were imaged using a Siemens Biograph mMR. PET and MR data were acquired simultaneously between 40 and 90 minutes after injection of 10 mCi 18F-sodium fluoride[1]. The standard approach reconstructs PET data using an end-expiration, breath-hold, MRAC map (3D-DIXON-VIBE gradient echo) with scan parameters: coronal orientation, FOV 500 x 400 x 260 mm³, resolution 4.1 x 2.6 x 3.1 mm³, TR/TE1/TE2 3.6/1.23/2.46 ms, FA 10^o, scan time 19 s. Our motion-averaged approach used an MRAC map derived from a free-breathing, golden-angle radial (GAR) stack-of-stars sequence (3D-GAR-VIBE gradient echo) [2] with scan parameters: coronal orientation, FOV 500 x 500 x 240 mm³, resolution 3 x 3 x 3 mm³, TR 4.5 ms, in-phase TE, FA 9^o, no fat suppression, 1600 radial acquisitions, scan time ~7 min. GAR-VIBE was reconstructed with adaptive coil-combination to give a constant-signal multi-coil-combined image for later image segmentation. In addition, GAR-VIBE-MRAC data were self-gated according to center-of-k-space intensity[3] (as in Siemens WIP 793) to reconstruct an end-expiration GAR-VIBE-MRAC volume. Finally, a DIXON-VIBE-MRAC was acquired at end-inspiration. GAR-VIBE-MRAC maps comprised a single soft tissue component with linear attenuation coefficient (LAC) 0.1 cm^-1. GAR-VIBE pixel values were plotted as a histogram. A clear peak at zero was discernible from image values in all images (Fig. 1). The first trough was used to automatically segment soft-tissue from background. This approach was chosen to be independent of user interaction to find the noise level. For comparison, in the DIXON-VIBE-MRAC maps, both soft tissue and fat were reassigned with LAC 0.1 cm^-1 and lung was set to background LAC 0 cm^-1. MRAC maps are shown in Fig. 2. PET images were reconstructed offline with the same emission data using each of the four different MRAC maps using e7_tools (Siemens) with parameters: PSF-OP-OSEM, 3 iterations, 21 subsets, matrix 344 x 344 x 129, 4-mm FWHM Gaussian post-reconstruction filter. Images were analyzed for image quality by an expert panel (blinded at time of image scoring) to assess the presence of attenuation correction image artifacts. One mark was given to each image for the following artifacts: i) dark or ii) bright region on lung/liver boundary compared to average lung and liver values, iii) dark or iv) bright regions on heart/lung border compared to average signal within the heart, v) bright signal localized within the bronchi, vi) spurious non-anatomical PET signal anywhere in the image. Average scores for each MRAC method, and paired t-tests were used to compare methods.Attenuation-corrected PET images for each MRAC method are shown in Fig. 3. Mismatch between PET signal and attenuation maps was typical with both end-expiration and end-inspiration breath-hold DIXON-VIBE-MRAC producing artifacts in the heart/lung and lung/liver boundaries. Using the proposed motion-averaged GAR-VIBE-MRAC these were significantly reduced (p = 0.002) compared to end-expiration DIXON-VIBE-MRAC (Fig. 4). In one patient who had experienced a myocardial infarction, increased 18F-fluoride uptake is evident and co-localized to the left anterior descending artery (gadolinium contrast-enhanced MRA) corresponding to the infarct territory on delayed contrast enhanced MRI (Fig. 5).Motion-averaged MRAC is necessary for artifact-free PET images in the heart. Using the novel golden-angle radial MRAC approach is superior to the standard breath-hold DIXON-VIBE-MRAC approach. Future work will assess using additional tissue segments in the MRAC map including fat and lung based on signal from the T1-weighted GAR-VIBE image. Finally, we were able to identify increased 18F-fluoride uptake in the coronary artery plaque of a patient who had recently had a myocardial infarction.