Investigation of the characteristics of reconstruction algorithms in positron emission tomography

Abstract

Background:<br />PET/CT imaging became increasingly important in radiation oncology for target volume delineation and response assessment. For absolute quantification of PET images, the standardised uptake value (SUV) is a commonly used parameter, which strongly affects volume segmentation and activity estimation based on the reconstructed image. Therefore, the influence of reconstruction algorithms and settings on the SUV is crucial. In this study, the impact of filter settings and permutations of iterations (i) and subsets (ss) was determined for the Ordered Subset Expectation Maximisation (OSEM) algorithm and the Siemens-specific TrueX algorithm. Material and Methods:<br />Phantom measurements were performed with different signal-to-background-ratios (SBR) (ranging from 2:1 to 9:1) on a Siemens Biograph 64 PET/CT Scanner (Siemens, Medical Systems) using a NEMA IEC/2001 Image Body Phantom. This phantom consists of a cylinder with inserts of plexiglass spheres (0.27, 1.15, 2.57, 5.58, 11.49 ml) both filled with 18F-Fluordeoxyglucose (FDG). Measurements were reconstructed using the OSEM and the Siemens-specificTrueX algorithm, with all permutations of iterations (i = 2, 4, 8) and subsets (ss = 4, 8, 14, 21) and compared to 4i21ss, which is the default protocol applied in our department. In contrast to the OSEM algorithm, the TrueX algorithm additionally uses a measured point spread function as system matrix [1].<br />For better comparison the concept of expectation maximisation equivalent iterations (EI) [2] (product of iterations and subsets) was used. The Gauss-filter was used for both algorithms while the "Allpass-filter" was applied only for the TrueX algorithm. The reconstructed PET images were visualised and evaluated using the software Rover (ABX, Dresden) and the generated data were analysed with R (v.2.12.1). For activity evaluation, the recovery coefficient RC (measured activity/true activity) was determined. Additionally, PET acquisitions of eight stereotactic lung patients were reconstructed using OSEMGaussian, TrueXGaussian and TrueXAllpasswith 4i21ss. The tumour lesions were delineated with a varying threshold method (42% for OSEMGaussian, 36% for TrueXGaussian and 29% for TrueXAllpass) and compared to the clinical target volume (CTV) and the volume determined in OSEM reconstructed images.<br />Results:<br />The threshold was calculated according to TH=a+b/SBR and the values for the parameters for the TrueXGaussian and OSEMGaussian were published in Knäusl et al [3]. For TrueXAllpass using 4i21ss (84 EI) a lower threshold and different fitting parameters (a=23.6±0.8, b=55.7±3.0) than in the previously mentioned study were found. Comparing all possible EIs to the standard configuration of 84 EI resulted in a higher deviation of the threshold for the TrueX (29%) than for the OSEM algorithm (18%). Increasing the number of iterations and subsets lowered this difference.<br />The RC was fitted according to 1/(RC)_"algorithm"=A*exp (B/(V_"sphere" n)) RCmax for the OSEM and RCmean for the TrueX algorithm were used, as recommended by the manufacturer. For the TrueXAllpass with 4i21ss, the coefficients were A=1.31±0.03 and B=0.23±0.01. The deviation from the 84 EI configuration was also higher for the TrueX (>20%) than for the OSEM (13%) algorithm for EI > 8. The CTV of the lung patients was on average underestimated by 50% using the OSEM and even more for the TrueX. Applying a threshold of 36% and 29% for the different filter settings of the TrueX algorithm decreased the difference to the OSEM-volume to 5% at most. Compared to the results from the OSEM algorithm, the SUVmax for TrueXGaussand TrueXAllpass were on average higher by 20% and 55% respectively. Conclusion and Outlook:<br />The results for threshold or RC from TrueX reconstructed images, especially with "Allpass-filter", are more sensitive to permutations of iterations and subsets than for the OSEM. Using the TrueX SUVmax for tumour classification in lung patients leads to an overestimation of the activity in the reconstructed image.<br />1 V.Y. Panin, F. Kehren, H. Rothfuss, D. Hu, C. Michel, M.E. Casey; Pet reconstruction with system matrix derived from point source measurements. IEEE Trans Med Imaging, 53:152 - 9, 2006 2 T. Sera, K. Erlandsson, A. Varrone, K. Tatsch, B.F. Hutton, J.C.<br />Dickson, L. Tossici-Bolt; The impact of reconstruction method on the quantification of datscan images. Eur J Nucl Med Mol Imaging, 37:2335, 2010.<br />3 B.Knäusl, A.Hirtl, G.Dobrozemsky, H.Bergman, K.Kletter, R.Dudczak, D.Georg; Pet based volume segmentation with emphasis on the iterative truex algorithm. Zeitschrift für Medizinische Physik, January 2011.<br />