Voxel-based Dynamic FDG PET with Simplified Data Acquisition Protocols and Shortened Scan Duration

Qing Ye,Tianyu Ma,Shulin Yao,Yunhan Yu,Shi Wang,Yaqiang Liu

2018-01-01

Abstract:644 Objectives: Dynamic FDG PET provides more accurate quantitative information than static PET. The net uptake constant Ki estimated from dynamic PET has shown better performance in oncological diagnosis than standardized uptake value (SUV). However, conventional dynamic imaging requires longer scan duration. A simplified data acquisition protocol with shortened scan duration could make dynamic FDG PET more clinically flexible. Our previous work has shown good results in lung cancer study using ROI-based dynamic FDG PET with simplified protocols (presented in the 2017 IEEE NSS/MIC conference). The objective of this study is to investigate the voxel-based dynamic FDG PET with simplified data acquisition protocols and shortened scan duration. Methods: The voxel-based dynamic FDG PET analysis was performed through indirect and direct ways. A population-based input function (PBIF) template was used in Patlak analysis. Patient-specific PBIF was generated by scaling the PBIF template with an image-derived input function (IDIF) estimated from the blood pool of left ventricle in dynamic images. Dynamic images were reconstructed with an OSEM algorithm, incorporating attenuation, crystal sensitivity normalization, random, decay and scatter corrections. Voxel-based Patlak analysis with scaled PBIF was used to indirectly generate Ki images. As for direct dynamic PET analysis, parametric images were reconstructed directly from the acquired dynamic data. We used existing dynamic images with known Ki to simulate sinograms and performed direct parametric reconstruction. A nested algorithm was applied to accelerate the convergence. 10 inner loops of parametric image update were nested into each outer loop of dynamic image update.The developed dynamic image methods were validated on simulated and clinical data with simplified protocols. We used single-bed FDG PET lung cancer patient data to generate preliminary results. Accurate simulation and more patient study has been planned. Four acquisition protocols with different scan duration (t) and start time (t[asterisk]) were compared in the indirect analysis: (a) t=70 min, t[asterisk]=20 min; (b) t=30 min, t[asterisk]=20 min; (c) t=30 min, t[asterisk]=30 min; and (d) t=40 min, t[asterisk]=30 min. Results: The 8-mm lung nodule could be obviously observed in the parametric images generated from indirect voxel-based Patlak analysis. The images became noisier with shorter scan duration. Direct Ki images were smoother than indirect Ki images. The mean Ki values of the nodule in direct image showed higher bias than indirect images, compared to the ground truth generated from ROI-based irreversible 2-tissue modeling. The larger bias may be caused by the inaccuracy in preliminary simulation and slow convergence in direct reconstruction. Further phantom simulation is ongoing to validate the accuracy of direct reconstruction.Conclusion: The voxel-based dynamic analysis with simplified protocols were validated to be feasible. Preliminary parametric images were reconstructed in both indirect and direct approaches. The optimized design of acquisition protocols requires further investigation.