Attenuation correction in MR-PET scanners with segmented T1-weighted MR images

Attenuation correction of PET data acquired in new hybrid MR-PET scanners which do not offer the possibility of a measured attenuation correction can be done in different ways. A previous report of our group described a method which used attenuation templates. The present study utilizes a new knowledge-based segmentation approach applied on T1-weighted MR images. It examines the position and the tissue membership of each voxel and segments the head volume into attenuation-differing regions: brain tissue, extracerebral soft tissue, skull, air-filled nasal and paranasal cavities as well as the mastoid process. To examine this new approach three groups of subjects having MRI and PET were chosen, the selection criterion being the different MR scanners, while the PET scanner was the ECAT HR+ in all cases: 1) four subjects with 1.5T MR images and CPFPX PET scans, 2) four subjects with 3T MR images and Altanserin PET scans, and 3) three brain tumor patients with 3T MR images from the hybrid MR-BrainPET scanner and FET PET scans. Furthermore, a single subject had 3T MR images, a FDG PET scan, and an additional CT scan. All segmented T1-weighted MR images were converted into attenuation maps for 511 KeV photons with coefficients of 0.096 1/cm for brain tissue, 0.146 1/cm for skull, 0.095 1/cm for soft tissue, 0.054 1/cm for the mastoid process, and 0.0 1/cm for nasal and paranasal cavities. The CT volume was also converted from the Hounsfield units into attenuation coefficients valid for 511 keV photons. The 12 segmented-based attenuation (SBA) maps as well as the CT-based attenuation (CBA) map were first filtered by a 3D Gaussian kernel of 10 mm filter width and then used to reconstruct the corresponding PET emission data. These were compared to the PET images attenuation corrected using the conventional PET-based transmission data (PBA). Relative differences (RD) were calculated from ROIs. For the single subject the RD of CBA data exhibit a mean of 1.66%?0.84% with a range from -0.88% to 3.42%, while the RD's mean of SBA data is 1.42%?2.61% (range from -4.12% to 4.66%). Comparing the results obtained with the SBA correction only, the RD for 1) range from -6.10% to 2.56% for cortical regions and from -6.99% to 5.64% for subcortical regions; for 2) they range from -7.33% to 2.33% for the cortical regions, subcortical ones being not drawn due to the not significant tracer uptake; for 3) the mean over the three subjects resulted in 0.89%?1.10% for ROIs at 48% threshold of the image's maximum and in 2.25%?1.50% for ROIs at 72% threshold. ROIs on the healthy contra-lateral grey matter show a mean of -3.24%?0.87%. In conclusion, the first attenuation correction results obtained with the new segmented-based method on a strongly heterogeneous collective are very promising. Further improvements of the method will be focused on the delineation of the skull.