%0 Journal Article %A Wentao Zhu %A Tao Feng %A Yun Dong %A Hongdi Li %A Jun Bao %T Enhancing lung lesion detectability with a bias-control reconstruction method in low dose PET/CT %D 2017 %J Journal of Nuclear Medicine %P 1360-1360 %V 58 %N supplement 1 %X 1360Objectives: In PET imaging, lungs are of low activity and usually converge much slower than other regions when iterative reconstruction algorithms are applied. The positive bias in lungs leads to reduced contrast recovery of small lesions. The scope of this work is to accelerate convergence in lungs to improve small lesion detectability in lungs.Methods: An iterative modified EM algorithm for reconstruction of PET image was derived, by easing the non-negativity constraint for voxels only in the low-uptake regions with a hybrid Poisson & Gaussian model. Meanwhile, the model for other regions in the image is kept the same as ordinary Poisson. The proposed method therefore dedicatedly improved the convergence rate for low-uptake regions and the noise property for other regions compared with previously proposed AML method, which imposes hybrid Poisson & Gaussian model for the entire data space. The procedure of our method is as follows: segment lungs in the coregistered CT image; apply a 3D smoothing filter to the segmented lungs region; forward project this CT-segmented lungs region to sinogram space; apply the proposed algorithm with the above inputs as well as the routine EM and AML algorithm to reconstruct the PET image. Simulation and clinical data were used to evaluate the algorithm. In simulation, a phantom with a cold cylinder in a uniform background was investigated. In clinical study, ten patient datasets with torso PET/CT scans were analyzed. The images obtained with the two methods were compared and numerical analysis such as CRC and SUV based metrics was evaluated.Results: Phantom simulation demonstrated that contrast recovery ratio of the point source was much higher with our method compared with routine EM reconstruction using same reconstruction parameters. At equal noise level the CRC was 0.93 for ours v.s. 0.81 for routine EM. In clinical study, SUV mean in ROIs in the lungs were around 8% lower with our method compared with standard EM method (population SUV mean: 0.55 for ours v.s. 0.60 for standard EM and 0.57 for AML). Meanwhile, SUVmax for the small lesions was higher with our method. The population mean of SUVmax of a few lesions for the 10 patients was: 4.6 for ours v.s. 4.1 for standard EM and 4.0 for AML. These results were obtained with the fact that SUV mean in multiple ROIs out of lungs were almost equivalent for our method, standard EM and AML (relative error <1%). A significant advantage of the proposed method compared with the AML approach was that streak artifacts out of body boundary were removed.Conclusion: The proposed method expedites the convergence rate of low-uptake regions such as lungs to reduce positive bias in iterative reconstruction. Contrast recovery of hot small structure in the lungs is improved while other regions are not affected. In conclusion, this CT guided EM-like method could be used in PET/CT to enhance small lesion detectability in lungs. Research Support: %U