Abstract
526
Objectives: Neuroendocrine tumors (NET) generally refers to a series of heterogeneous tumors originating from peptidergic neurons and neuroendocrine cells. Although NET is relatively rare malignancies, studies have shown that their incidence has increased significantly in recent years. Both the 18F-FDG and 68Ga-DOTATATE PET/CT image information are commonly used to NET. Imaging 18F-FDG and 68Ga-DOTATATE simultaneously in a single scan can reduce time and cost of diagnosis, and avoids changes in biochemical or pharmacologic state. The objective of this study is to investigate the feasibility of quantification of simultaneous FDG and 68Ga-DOTATATE neuroendocrine tumor PET imaging using machine learning approach.
Methods: Twelve NET patients were recruited in the study. All patients had both 45-min 18F-FDG and 68Ga-DOTATATE single-bed dynamic PET-CT scans. Regions of interest (ROIs) were manually defined on the PET-CT images and applied to dynamic PET images to generate ROI time activity curves (TACs, CPET(t)). Aorta TACs were used as arterial input functions for kinetic modeling. The mixed aorta and tissue TACs were obtained by superimposing TACs of two tracers with time-delay injection protocols. A recurrent XGBoost (rXGBoost) was proposed to separate the mixed aorta and tissue TACs, respectively.
Results: The relative mean square error (RMSE) of 18F-FDG and 68Ga-DOTATATE were 0.0199 and 0.0240 with a 3min delay injection protocol. In addition, tracer uptake rate constants were estimated by kinetic modeling. The kinetic parameters of machine learning based dual tracer PET were highly correlated with those of single tracer PET with interlaced injection protocols, and its correlation coefficient is higher than that of parallel multi-tracer compartment modeling method.
Conclusions: Our proposed method had a good performance on dual tracer PET signals separation, and it has excellent potential for improving the accuracy of NET diagnosis.