Abstract
1187
Objectives: Dynamic modeling is an important method to quantify metabolic processes using 18F-fluorodeoxyglucose positron emission tomography (18F-FDG-PET). For liver studies, a tracer is supplied to entire liver vascular anatomic regions by both the hepatic artery (HA) and the portal vein (PV). The standard modeling using the artery blood input cannot reflect its true time domain delay and scattering of the portal vein input. Our study is to develop an improved dynamic modeling system to quantify the liver tracer kinetics in damaged hepatic vascular lesions.
Methods: The time-activity curve (TAC) was created based the drawn region of interest (ROI) in different organs (Fig 1.A) of patient using co-registration images of PET/CT. Portal vein tracer concentration was estimated according to the time domain delay and scattering of the tracer signal of the hepatic artery and corrected by the measurements of gut radioactivity. Our improved dynamic model considered two pools: input1 (CHA(t)) from the hepatic artery, which was acquired from the TAC of aorta, and input2 (CPA(t)) from the portal vein, which was corrected by gut radioactivity (Cgut(t)) (Fig 1.B). The TACs of different regions were used to estimate the parameter of the model by the equations in Fig 1.C. According to the metabolic characteristics of 18F-FDG, the model of 18F-FDG metabolism includes the 3 rate constants (K1, K2, and K3), where K1 represents the transfer constant 18F-FDG for blood-to-cell; K2 represents the transfer constant 18F-FDG for cell-to-blood; K3 represents the transfer constant for 18F-FDG to 18F-FDG-6P. Finally, we define the metabolic parameter Ki=K1K3/(K2 +K3) to represent the overall metabolic rate of the liver. The Ki values between the healthy liver and tumors vascular nodules were compared. At the end, the deep learning methodology was applied for those TACs to improve the quality of the goodness of fitting.
Results: The modified dual input function of 20 TACs of different regional ROIs from a liver cancer patient showed improved curve fitting by coefficient of variation (CV), for instance in 2TCM, the K1 value of the malignant vascular nodule is reduced from 3.09% to 0.63%, while the mean value of K1 remain the similar level (Table 1). These findings validated the improved model-based estimation of the dual input for two-tissue compartment model (2TCM) to get realistic data modeling. The peak of the dual input was delayed 30 seconds compared to the aortic input indicated that the time domain delay from the dual input is more realistic for liver vascular system (Fig 2). Regarding model selection between 2TCM and 1TCM, the CV% value were much smaller in 2TCM (0.63% vs 4.84%, Table 1), therefore, Ki values in the liver malignant and healthy vascular nodules were analyzed based on 2TCM. And the results suggested that Ki values of the malignant vascular nodules were significantly higher than the healthy liver vascular nodules (0.52 ± 0.080 vs 0.007 ± 0.002) (Table 2). Simultaneously, Ki values in the tumor bed were analyzed (the malignant vascular nodules were cut off for one year, male,62), the results show that the Ki values of the tumor bed were similar to the healthy liver vascular nodules (0.054 ± 0.006 vs 0.035 ± 0.012) (Table 3).
Conclusions: This study provided an improved dynamic model with dual input function, which has been well-fitted with the original 18F-FDG-PET dynamic data from a liver cancer patient. This dynamic model has been proved to be a quantitative method to distinguish the healthy liver vascular nodules from malignant vascular nodule. This new modeling method deserves further analysis with large amount patients to validate its function for quantifying the liver tracer kinetics in damaged hepatic vascular lesions.