New compartment model for hepatic blood flow quantification in humans from 15O-water PET images

Different compartment models are commonly used to derive crucial information about blood flow, metabolism, and oxygenation from the results of a dynamic positron emission tomography (PET) scan. However, compared to blood flow in many other organs of interest, the hepatic blood flow (HBF) quantification is challenging due to the dual blood supply of liver from both the hepatic artery and the portal vein (PV). Here, we introduce a new model that can be used to estimate the HBF in combination with an automatic volume of interest selection method.

By using the $_{}^{15}$ O-water PET data of 57 patients, we extract the mean time-activity curves (TACs) from aorta, hepatic PV, liver, and spleen with help of an automated computer tomography-based segmentation tool and systematically fit our new compartment model and three earlier compartment models from literature to the TACs. After this, we compare the model performance with mean relative error (MRE), mean squared error, and Akaike's information criteria with one-sided Wilcoxon signed-rank tests. After determining the best model, we study possible HBF differences caused by age, sex, and weight with Mann-Whitney U test and Pearson's correlations coefficient.

We obtained the mean arterial HBF of 0.299±0.168 mL/min/mL, the mean portal HBF of 0.930±0.520 mL/min/mL, and the total HBF of 1.229±0.612 mL/min/mL with our new model. Based on earlier research, both these estimates and also the results of two earlier versions of the original dual-input model are realistic. Out of these three models, our proposed model performed the best in terms of MRE (p-values $\le$ 0.001). According to our results, there are no significant sex- or age-based differences but there is moderate positive correlation between the arterial and portal HBFs and negative correlation between the total HBF and the weight of patients.

The HBF can effectively be estimated from $_{}^{15}$ O-water PET data with our new model in combination with robust segmentation by TotalSegmentator. Due to the fact that potential underestimation of the PV concentration caused by the small size of this vessel might lead to overestimation of the HBF, more research would be beneficial to validate these methods further. Our results suggests that there is a negative trend between the HBF and the weight, though this might be related to the underlying conditions of the patients.