Abstract
1963
Objectives Transmission-less PET attenuation correction has been an active research topic in recent years due to the clinical introduction of PET/MR systems. One of the promising solutions is TOF-MLAA (1, 2). For TOF based reconstructions, directly reconstructing from list-mode events is more efficient than reconstructing 3D TOF projection data since one does not need to store/process the TOF projection data which are enormous in size. Moreover, a more optimal initial µ-map or magnitude preconditioning; i.e. Initial Average Mu-value (IAM) approach has been proposed to further improve the efficiency and accuracy of TOF-MLAA (3, 4). In this work, a list-mode based TOF-MLAA reconstruction with the IAM preconditioning was implemented and evaluated using GATE simulation.
Methods A 3D torso phantom which consists of 3 tissue classes (water, lung, and bone) was simulated using GATE. The simulated geometry was based on Philips Gemini/Ingenuity TOF-PET scanner with TOF resolution of 588 ps. Zero activity was assigned to the lungs and spinal regions while uniform activity was assigned to the surrounding soft-tissues. ~40 million prompts which correspond to a 90 second acquisition of 68.3 MBq of activity (typical clinical whole body scan for a single bed position) were simulated within the torso phantom and reconstructed using the list-mode TOF-MLAA based on (1) with voxel-wise update in the MLTR instead of regional-wise update. Monte Carlo simulated scatter correction and randoms correction based on delayed coincidences were included along with the IAM preconditioning. List-mode TOF-MLAA was also performed with the conventional preconditioning (i.e. filling the initial µ-map with the µ-value of water). TOF-OSEM reconstruction of the emission data using the perfect µ-map was used as the reference emission image for comparison purposes.
Results It was observed that the reconstructed µ-values reached closest to the reference values at 9th iteration of list-mode IAM-TOF-MLAA. Slightly overestimated µ-values were observed after 10 iterations likely due to the increased noise as was observed previously from 2D simulations (4). For the TOF-MLAA reconstruction with the conventional preconditioning, the error in the reconstructed µ-values is +8% in bone, +14% in water, and +54% in lung at the 9th iteration. On the other hand, the error in the reconstructed µ-values is <1% in bone, -2% in water, and +18% in lung at the 9th iteration of IAM-TOF-MLAA. The reconstructed emission image profile using IAM-TOF-MLAA also showed good agreement with the reference, whereas that obtained from TOF-MLAA with the conventional preconditioning showed overestimated activity as expected.
Conclusions This work demonstrated that the unconstrained and fully corrected list-mode IAM-TOF-MLAA can produce quantitative µ-maps with a relatively low number of iterations and generate accurate emission images as based on results from GATE simulation. The IAM preconditioning uses the average µ-value which has a much lower dependency on the tissue classification than most rescaling or calibration techniques used for TOF-MLAA to produce a quantitative µ-map; i.e. the proposed IAM approach is much less sensitive to the error due to mis-classifications of tissue types. Further validations of list-mode IAM-TOF-MLAA will be performed using clinical patient data.