Impact of Reconstruction Algorithms on the Quantitative Evaluation of 11C-PiB PET/CT Studies

1459 Introduction: Modern PET/CT with better instrumentation and software could possibly produce more accurate diagnosis for brain studies such as in Alzheimer disease. The objective of the study is to evaluate the impact of different reconstruction algorithms on the accuracy of brain 11C-PiB PET images reconstructed from a Siemens Biograph Vision PET/CT scanner. Methods: PET data included scans of NEMA IQ and Hoffman phantoms filled with 18F solutions and 60 min dynamic scans of 11C-PiB in five normal and one AD subject. All PET data were reconstructed using four different reconstruction algorithms: 3D FBP, 3D FBP with TOF (3D FBP+TOF), OP-OSEM with TOF (3D OP-OSEM+TOF) and OP-OSEM with point spread function (PSF) and TOF (3D OP-OSEM+PSF+TOF). Phantom data were reconstructed using 4-iterations and 5-subset with and without 5 mm Gaussian while the same iterations and subsets without any post filtering were applied for clinical data. We calculated the recovery coefficient (RC) of the maximum, peak and mean as a noise level and reconstruction settings using the NEMA IQ phantom. The effects of the choice of image reconstruction algorithm were evaluated using the 3D Hoffman phantom data for the calculations of mean activity concentrations, recovery coefficients and measurement accuracy. For 11C-PiB PET studies, the SUVR images, R1 and DVR (=BPND+1) parametric images were generated using a spatially-constraint simplified reference tissue model (SRTM2) [1], and DVR using conventional Logan plot with reference tissue input. Results: Analysis of theNEMA IQ phantom showed that the 3D OP-OSEM+PSF+TOF was the best RC for both short frame duration and longer frame duration for all spheres. Analysis of the Hoffman phantom indicated that 3D OP-OSEM+PSF+TOF had the best accuracy for grey and white matter respectively. Consistent with phantom studies, the SUVR and parametric images of R1 and DVR generated from 11C-PiB study with the OP-OSEM+PSF+TOF algorithm provided lowest noise levels. Highly linear correlations between the reconstruction methods (FBP+TOF, OP-OSEM+TOF, OP-OSEM+PSF+TOF) and FBP were obtained for SUVR, R1, and DVR measurements (R2: 0.93-0.98). When using 3D OP-OSEM+PSF+TOF for 11C-PiB, an increment of 10 % and 8 % can be seen for SUVR 0-20 min and SUVR40-60min respectively. R1 and DVR using SRTM2 show increment of 9% and 13% while 8% for DVR Logan. The DVR images generated from Logan plot showed noise-induced underestimation, where the FBP provided lowest DVR while OP-OSEM+PSF+TOF provided highest DVR. Conclusion: 3D OP-OSEM+PSF+TOF produced better accuracy in phantoms and for human dynamic 11C-PiB PET images and will be required further optimization in reconstruction parameters.It demonstrates that an improvement of SUVR, R1, and DVR compared to 3D FBP and other reconstruction methods. SRTM2 provides robust results that appears optimal for quantification of human 11C-PiB PET.[1] Zhou et al., Using a reference tissue model with spatial constraint to quantify [11C] Pittsburgh compound B PET for early diagnosis of Alzheimer9s disease. Neuroimage, 2007, 36(2):298-312Figure 1. 1st row shows NEMA image quality phantom while 3D Hoffman brain phantom on the second row with 5 minutes frame duration. Third row shows SUV for one AD subject. The generated parametric images of R1 and BPND derived using SRTM2 method are shown in row 4 and 5 respectively. The last row shows the parametric images of BPND generated using Logan reference for the same subject. All phantom and clinical data were reconstructed using FBP (first column), FBP+TOF (second column), OP-OSEM+TOF (third column) and OP-OSEM+PSF+TOF (fourth column).