Radial k-space acquisition improves robustness of MR-based attenuation maps for MR/PET quantification in an animal imaging study of the abdomen

One of the most important steps in positron emission tomography (PET) is the correction of photon attenuation for accurate quantitative PET. Currently, FDA approved clinical MR/PET systems employ segmentation of conventional, low resolution, gradient echo (GRE) based, T1-weighted MR images to generate maps for MR-based attenuation correction (MRAC). However, these acquisitions are optimized for imaging human subjects and exhibit artifacts when used in preclinical MR/PET studies. Pronounced breathing artifacts in animal models used for preclinical imaging, impede accurate segmentation for generation of attenuation maps, impacting quantitative measurements of reconstructed PET images. We propose a radial k-space acquisition sequence designed to redistribute coherent breathing artifacts that result from Cartesian k-space trajectories into incoherent pseudo-noise spread across the image domain. PET data from five rabbits was reconstructed using the system standard MR-derived attenuation map with segmentation errors, due to breathing artifacts in the Cartesian acquisition (cartMR map), the manually segmented MR-derived attenuation map (msegMR map) and the radially acquired MR sequence used to generate an attenuation map from the system standard segmentation algorithm (radMR map). The resulting attenuation corrected PET data sets (PET
_cartMRmap
, PET
_msegMRmap
, and PET
_radMRmap
) were then qualitatively and quantitatively evaluated. Voxel-by-voxel comparison of PET values for all five rabbits showed excellent correlation between PET
_msegMRmap
and PET
_radMRmap
SUV values (R=0.999, p0.0001). Bland-Altman plots showed that the mean of the difference of SUVs between PET
_msegMRmap
and PET
_radMRmap
voxels for all five rabbits was 0.53% (0.004±0.014SD). Region-of-interest-based comparison showed that PET
_radMRmap
and PET
_msegMRmap
methods differ in SUVmean by -0.7% to 0.9% and SUVmax by -1.2% to 2.7%. Employing a radial k-space MR acquisition during preclinical MR/PET protocols facilitates highly accurate segmentation and PET quantification, without the need for subjective user input and is therefore, better suited for use in preclinical MR/PET protocols than the existing MR Cartesian acquisition.