PET Crystal Efficiency Normalization Using a Stationary Line Source

1517 Objectives: Accurate crystal efficiency normalization plays an important role in accurate and artifact-free imaging in PET. Axial FOV of PET has been growing over the years for lower patient dose and to study functioning correlations between organs further apart. Crystal efficiency normalization needs to be periodically performed on clinical PET scanners. The current commonly used normalization method relies on a large uniform cylinder phantom. The cylinder phantom is heavy and presents challenges in preparation which becomes particularly difficult for a long axial FOV or large bore scanner. One approach to overcome these difficulties is to use a rotating rod source, effectively mimicking a cylinder source with a line source. Such an approach is still somewhat complicated due to the need for a rotating rod. In this work, we propose a method for crystal efficiency normalization using a stationary line source. In this approach, no phantom movement is required and the use of a simple line source is convenient in terms of preparation and storage. Methods: Using data from a well-centered line source, the overall efficiency of the detectors is decoupled into two parts: 1. The relative efficiencies of detectors transaxially within a ring are determined by the singles counts recorded at each crystal 2. The relative axial efficiencies of detectors in different axial rings are determined by the paired counts recorded at each ring. Assuming there are I axial rings and J crystals in each ring, if the number of singles events at crystal (i,j) are si,j and the number of pairs events are pi,j, the estimated crystal efficiency ηi,jLine for that crystal is given by ηi,jLine=Nsi,jPi/Mi , where si,j is the number of singles events at crystal (i,j), Pi and Mi are the total number of pairs and singles evets from ring i. N is a normalization factor. This expression for ηi,jLine cannot capture the transaxial variances in detection efficiency due to different incidence angles. To overcome this difficulty, the difference between a cylinder source and a line source scan can be evaluated using a Monte Carlo simulation. We first estimate crystal efficiencies ηi,jCylinder_MC from the cylinder simulation dataset using the commonly used Defrise method. We then estimate crystal efficiencies ηi,jLine_MC using the method described above from the simulated line source dataset. We also scan the same line source in the real system to estimate the crystal efficiencies from a real line source, ηi,jLine_real. The final estimated crystal efficiency of the real crystal ηi,j is then given by ηi,j=N ηi,jLine_realηi,jCylinder_MC/ηi,jLine_MC , where N is a normalization factor. To demonstrate the effectiveness of the proposed approach, we compared crystal efficiency maps and phantom reconstructed images from a Canon’s Cartesian TOF PET/CT scanner using the proposed method and compared the results with the Defrise method. Results: By comparing the crystal efficiency maps, it is clear that the proposed method can produce results very similar to those obtained with the Defrise method using a cylinder source: The overall pattern of the crystal maps are very similar and the proposed method also captured the detailed structures in the crystal map. Reconstructions from both methods are also very similar quantitatively and qualitatively. Conclusions: In this work, we proposed a practical method for PET crystal normalization using a stationary line source to make the normalization scan more convenient, especially for long axial FOV scanners. We decoupled the crystal efficiency into two parts using the line source: transaxial relative efficiencies from singles events, and axial relative efficiencies from paired events. In order to capture the transaxial variances, we additionally use Monte Carlo simulations of cylinder and line sources. Experimental results show that the proposed method can produce normalization results very similar to those obtained using a cylinder source and as a result, the resulting reconstructions are also very similar.