To improve the quantitative accuracy of linear attenuation coefficients measured by computed tomography (CT), we used a single scatter model to estimate the Compton scatter distribution and then a polychromatic image reconstruction algorithm, namely the iterative maximum-likelihood polychromatic algorithm for CT (IMPACT), was implemented to include scatter correction (SC). To perform the IMPACT, the X-ray spectra of a tube were estimated via an expectation-maximization (EM) algorithm with SC. To test the accuracy of the estimated spectra, the projection images of cubic phantoms containing different depths of polymethylmethacrylate (PMMA) were acquired. The percentage of root mean square errors (%RMSE) of the measured transmission data and calculated transmission values were used to evaluate the accuracy of the estimated spectra. In addition, a hydroxylapatite (HA) phantom was used to study streak artifacts and evaluate the accuracy of the linear attenuation coefficients estimated using the IMPACT with SC. The %RMSE of the EM-with-SC estimated spectra were all lower than 1% and were also smaller than that without SC. The error in the quantification of the HA linear attenuation was only about 3% after SC. Our results showed that the quantitative accuracy of the linear attenuation coefficients measured with a cone beam CT was improved when the IMPACT with SC was used.

Energy-mapping, the conversion of linear attenuation coefficients ( ) calculated at the effective computed tomography (CT) energy to those corresponding to 511 keV, is an important step in CT-based attenuation correction (CTAC) for positron emission tomography (PET) quantification. The aim of this study was to implement energy-mapping step by using curve fitting ability of artificial neural network (ANN). Eleven digital phantoms simulated by Geant4 application for tomographic emission (GATE) and 12 physical phantoms composed of various volume concentrations of iodine contrast were used in this study to generate energy-mapping curves by acquiring average CT values and linear attenuation coefficients at 511 keV of these phantoms. The curves were built with ANN toolbox in MATLAB. To evaluate the effectiveness of the proposed method, another two digital phantoms (liver and spine-bone) and three physical phantoms (volume concentrations of 3%, 10% and 20%) were used to compare the energymapping curves built by ANN and bilinear transformation, and a semi-quantitative analysis was proceeded by injecting 0.5 mCi FDG into a SD rat for micro-PET scanning. The results showed that the percentage relative difference (PRD) values of digital liver and spine-bone phantom are 5.46% and 1.28% based on ANN, and 19.21% and 1.87% based on bilinear transformation. For 3%, 10% and 20% physical phantoms, the PRD values of ANN curve are 0.91%, 0.70% and 3.70%, and the PRD values of bilinear transformation are 3.80%, 1.44% and 4.30%, respectively. Both digital and physical phantoms indicated that the ANN curve can achieve better performance than bilinear transformation. The semi-quantitative analysis of rat PET images showed that the ANN curve can reduce the inaccuracy caused by attenuation effect from 13.75% to 4.43% in brain tissue, and 23.26% to 9.41% in heart tissue. On the other hand, the inaccuracy remained 6.47% and 11.51% in brain and heart tissue when the bilinear transformation was used. Overall, it can be concluded that the bilinear transformation method resulted in considerable bias and the newly proposed calibration curve built by ANN could achieve better results with acceptable accuracy.

No abstract

No abstract