J Nucl Med. 2007; 48 (Supplement 2):156P
Instrumentation & Data Analysis: Image Generation
Multi-modality and Multi-tracer Imaging |
An approach for MR-based attenuation correction for combined MR/PET: Effects of ignoring bones
Axel Martínez-Möller1,
Michael Souvatzoglou1,
René Botnar1,
Nassir Navab2,
Sibylle Ziegler1,
Markus Schwaiger1 and
Stephan Nekolla1
1 Nuklearmedizinische Klinik der TU München, München, Germany;
2 CAMP-AR, TU München, Munich, Germany
532
Objectives: Based on the clinical success of PET/CT systems, MR/PET devices are currently under development as MRI would offer improved soft tissue contrast. However, the MR signal is typically not related to 511 keV photon attenuation as is data from CT. Especially there is a lack for the delineation of bones. In this study, PET/CT data was used to assess the necessity of bone segmentation for attenuation correction, and the option of using two-tissue maps (air/tissue) was studied. Spatially registered PET/CT and MRI data from initial patient studies were used to investigate this approach. Methods: 25 patients were examined using FDG in a Biograph16 PET/CT. Two bed positions (hip, abdomen) were analyzed for each patient. Emission data was attenuation corrected using 3 different attenuation maps (AM): the original CT-based AM, the AM after bone removal (factors above 0.12cm-1 replaced by 0cm-1), and the AM after segmentation in two classes: air (0cm-1) and soft tissue (0.1cm-1). These maps represent an estimation of the AM which is obtained by a proton-weighted image assuming spatial coincidence of protons (i.e. water, fat) and photon attenuating tissue (CT). In preliminary patient studies, a heavily proton-weighted, ECG gated fast gradient echo sequence was implemented (effective TR>500 ms, TE:0.7ms, time per PET bed position 1min). After spatial registration with the CT, AM were generated from the MRI. The reconstructed PET images were compared by means of the correlation computed on a voxel-by-voxel basis excluding air, and by means of ROIs manually defined in high uptake regions. Results: The correlation between images corrected with AM without bones and with original AM was 0.99, and the correlation between images corrected with two-tissue AM and with original AM was 0.98. The average absolute difference in ROI activity was 1.4% when using the AM without bones and 8.6% when using the two-tissue AM. Patient data corrected using the MR-based AM showed excellent agreement (R=0.95) with the CT-corrected PET. Conclusions: The influence of bones in PET attenuation correction was negligible for the abdomen and even for the hip region. Thus, MR-based AM can be obtained from adjusted MRI sequences which do not delineate the bones.
