TY - JOUR T1 - MR-based attenuation correction in cardiovascular PET/MR imaging: challenges and practical solutions for cardiorespiratory motion and tissue class segmentation JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 452 LP - 452 VL - 57 IS - supplement 2 AU - Nicolas Karakatsanis AU - Philip Robson AU - Marc Dweck AU - Ronan Abgral AU - Maria Trivieri AU - Javier Sanz AU - Johanna Contreras AU - Jagat Narula AU - Maria Padilla AU - Umesh Gidwani AU - Valentin Fuster AU - Jason Kovacic AU - Zahi Fayad Y1 - 2016/05/01 UR - http://jnm.snmjournals.org/content/57/supplement_2/452.abstract N2 - 452Objectives Cardiovascular PET imaging may substantially benefit from the advent of simultaneous hybrid PET/MR imaging, thanks to the superior soft tissue characterization and of cardiac MRI and the capability for concurrent anatomo-functional acquisition at significantly less radiation exposure compared to PET/CT. However, MR-based attenuation correction (MRAC) is challenging as MR is sensitive to proton but not electron density. In addition, standard MRAC typically employs a breath-hold Dixon MRI sequence, thus introducing cardiorespiratory motion-induced mismatches between breath-hold attenuation and free-breathing emission maps, in the absence of motion compensation. Recently, we demonstrated the benefits of segmenting air and soft tissue from a free-breathing MRI sequence to significantly limit these mismatches. In this study, we introduce additional segmentation of lung and fat to considerably improve PET lesion quantification and evaluate the method on a larger study cohort.Methods A Siemens golden angle radial (GAR) VIBE © MR sequence was employed to acquire free-breathing 3D data followed by tissue classification to obtain the 3D MRAC maps. The tissue segmentation method now supports not only 2-class air/soft tissue (2C-MRAC), but also 3-class air/lung/soft tissue (3C-MRAC) and 4-class air/lung/fat/soft tissue (4C-MRAC) separation. The same fixed attenuation values were assigned to each class as with standard Dixon MRAC. Currently, manual segmentation of VIBE MRAC maps has been used, by qualitatively matching the tissue classes with the respective end-expiration Dixon MRAC classes, except in areas of known misclassification errors, such as the bronchus and lung interfaces. Validation was performed for a total of n=15 dynamic 18F-FDG and 18F-NaF cardiac PET/MR studies using Siemens Biograph mMR © scanner. The acquired PET data were reconstructed offline with a 3D PET OSEM (3 iterations, 21 subsets) algorithm employing Siemens e7-tools. The performance of proposed 4C-MRAC maps was assessed against respective 3C- and 2C-MRAC as well as standard breath-hold Dixon-based 4C-MRAC at end-expiration and end-inspiration positions.Results The application of standard end-expiration Dixon 4C-MRAC maps systematically produced distinct artifacts at the bronchus, due to misclassification of air as soft-tissue, and at lung-heart and lung-liver interface, due to cardiorespiratory motion. All results from our larger cohort of exams confirmed our previous findings for significant reduction of the aforementioned artifacts with free-breathing VIBE MRAC, especially for the case of 18F-NaF. In particular, quantitative analysis on two PET lesion regions (NaF coronary artery and FDG myocardium uptake) has shown 5-15% mean difference in absolute uptake, tumor-to-background ratio and contrast-to-noise scores between Dixon 4C-MRAC and VIBE 2C-MRAC, although all lesions corresponded to positions assigned the same attenuation values, thus suggesting AC artifact propagation through PET reconstruction. An additional 3-12% mean difference was also observed between VIBE 2C- and 4C-MRAC PET images for the same lesions, which is comparable with Dixon vs. VIBE MRAC PET differences, thus demonstrating the importance of both free-breathing MRAC as well as detailed 4-class segmentation. Finally, end-inspiration Dixon 4C-MRAC is not recommended, as it was associated with >50% differences from all other cases.Conclusions In the absence of motion compensation, cardiovascular PET/MR imaging is challenged by attenuation-emission mismatches and propagation of tissue misclassification errors. Nevertheless, most major PET artifacts can be removed with proposed free-breathing MRAC, while quantification in coronary plaques and myocardium lesions may be further improved with 4-tissue class segmentation, thus rendering cardiovascular PET/MR feasible. ER -