Quantitative accuracy of MR-based attenuation correction in whole-body PET/MR

Objectives: In 2010, a hybrid whole body PET/MR system (Philips Ingenuity TF PET/MR) was installed at our institute. PET/MR is expected to offer many new possibilities in the field of quantitative bimodal functional imaging. Quantitative PET image reconstruction requires attenuation correction (AC) which is commonly based on a measurement of photon attenuation using either a transmission scan in stand-alone PET
(TRAC) or a CT scan in PET/CT systems. In PET/MR, however, AC is performed with a software-based approach (MRAC) using dedicated tissue segmentation and tissue type identification (air, lung, soft tissue) of an MR image. Here, we present a first evaluation of the accuracy of the vendor-provided MRAC in whole body investigations.
Methods: We performed sequential PET scans of 9 patients on a stand-alone Siemens ECAT HR+ and on the Ingenuity PET/MR with a time delay of approximately 2h. In addition to the standard reconstruction using MRAC, we performed a second reconstruction of the emission data from the PET/MR using the coregistered transmission-based attenuation maps from the HR+. For the two resulting PET image volumes, we performed a voxel-by-voxel correlation analysis and a comparison of the SUVs in different ROIs.
Results: The PET correlation analysis yielded a Pearson correlation coefficient of 0.93 and 0.96 and average deviations of 4% and 3% between the two reconstructed images for all voxels in the lung and in the torso, respectively. In one patient we observed failure of correct lung detection because of severe motion artifacts of the heart in the MR image. Accordingly we found a very large average deviation of 65% in the lung.
Conclusions: The MRAC algorithm generally yields satisfactory results with respect to soft tissue and air segmentation. The average deviation between PET images reconstructed with TRAC and MRAC is usually small and quantitative accuracy is adequate. Failure of segmentation occur rarely which necessitates manual intervention to achieve adequate segmentation.