Patient characteristics
The study was approved by the Ethics Committee at Peking Union Medical College Hospital. All patients signed informed consent. Eight male patients with 12 available liver lesions (age range, 47-70 y) were recruited in this study. The pathology of these patients was confirmed by surgical resection or by needle biopsy. Among these patients, four patients had been histologically confirmed as hepatocellular carcinoma (HCC), two patients had intrahepatic cholangiocarcinoma, one patient had liver metastasis of gastric cardia adenocarcinoma and one had inflammatory granulomatous.
PET/CT scan
PET/CT scans were conducted on a PoleStar m660 PET/CT scanner (SinoUnion Healthcare, Beijing, China) at Peking Union Medical College Hospital (PUMCH)(14) for all patients. CT transmission scans (120 kV, 260 effective mA) were performed first for attenuation correction and image fusion. Then, 174-259 MBq 68Ga-FAPI-04 was administered intravenously and the dynamic PET was performed over the liver region simultaneously. Each PET scan lasted 60 min. Dynamic PET images were reconstructed using a manufacturer (SinoUnion Healthcare, Beijing, China) provided stand-alone advanced research workstation with standard ordered subset expectation maximization (OSEM) algorithm with 2 iterations and 10 subsets. The 120-frame reconstruction protocol consisted of 60 frames of 5s, 10 frames of 30s, and 50 frames of 60s.
Image analysis
Delineation of volumes of interest were done on the Hermes Hybrid Viewer tool (Hermes Medical Solutions AB, Stockholm, Sweden). The volumes of interest (VOIs) were drawn manually over all visible lesions and healthy regions within an area far away from any lesion in the liver on the CT image of each patient. For the image-derived input functions (IDIF), a VOI was drawn within the hepatic artery (here denoted as A) and one in the portal vein (V). Then VOIs were copied to the dynamic PET images. The corresponding concentration time-activity curves (TACs) were extracted. In case of healthy regions, the mean standardized uptake value (SUV) was taken from each region. In case of the lesions and the IDIFs, the voxel with the maximum SUV (SUVmax) was taken, since these values are least affected by partial volume and motion effects (19). Furthermore, the according volume sizes and the uptake of the last five minutes were exported. A representative PET/CT scan is shown in Figure 1.
Kinetic models
The kinetic behavior of tracers in the liver is usually described with a model using two input functions(15), since the liver tissue is supplied by both, an arterial and a venous blood input. All IDIFs were fitted with a tri-exponential function starting from the peak maximum and with a linear increase before the maximum. To find the optimum model, three models were applied to all TACs (Figure 2). One reversible two-tissue compartment model “model A” with one input function from the artery using, and second “model V” with one input function from the portal vein. The third “model AV” used both input functions from artery and vein according to the formulas:
with C1(t) and C2(t) as the hepatic compartment, A(t) as the artery IDIF and V(t) as the venous IDIF. Since A(t) and V(t) are both contributing to the fraction of blood volume by having a proportion of Ka/(Ka+Kv) and Kv/(Ka+Kv) to vB (see equ, 3 below), each amount was assessed from their inflow rate constants for the measured compartment Cmeasured(t):
with vB as the fraction of the measured volume occupied by blood and Ka and Kv as the influx rate constant of the aorta and the vein, respectively.
With the rate constants as fit parameters, all model fits were performed according to the least-squares method and optimized with a Levenberg-Marquardt algorithm, implemented to a Java program. Errors of the fit parameters were estimated by calculating the covariance matrix. The residual sum of squares was calculated for each TAC, as well as the average and standard deviation of all rate constants, furthermore the parameters VT = K1/k2 (1+k3/k4) and VND = K1/k2 were calculated, with K1 as either Ka, Kv or (Ka+Kv) depending on the model. Since to our knowledge no initial parameters are available for FAPI model parameters, every TAC was first fitted with a one-tissue model to obtain values for K1, k2 and vB. For the two-tissue models, these parameters were takes as initial values and a second fit was performed to obtain k3 and k4; note that the latter could become zero thus resulting in a one-tissue model for the according TAC.
Statistical analysis
In order to find the optimum model, the Schwartz Criterion (SC) (16) was applied on all models, the percentage of TACs in relation to all TACs showing a model as most preferred was calculated for each model. The ANOVA test was used to find differences between HCC, non-HCC and healthy regions, significant differences between two groups were estimated with the unpaired Student’s t-test. Due to the small sample size, a p-value of less than 0.01 was classified as significant.