Patient population
To evaluate the repeatability of TBF measurement with static and dynamic 82Rb PET, ten patients diagnosed with prostate cancer were included in the study. Both low-risk patients in active surveillance and high-risk patients were recruited as we aimed to represent tumors spanning the range from low, intermediate, to high blood flow. The low-risk patients had undergone a clinical multiparametric magnetic resonance imaging (MRI) scan, and the high-risk patients had undergone a clinical 68Ga-prostate-specific membrane antigen (PSMA) PET/computed tomography (CT). Each patient underwent two 82Rb PET scan sessions within 1 week, each consisting of a dynamic pelvic and a dynamic cardiac 82Rb PET. No patients were excluded from the study.
Imaging
All 82Rb PET scans were carried out on a GE Discovery MI Digital Ready PET/CT (GE Healthcare, Waukesha, WI, USA). At the beginning of each scan, a bolus of 1110 MBq 82RbCl was injected directly by the Cardiogen-82 generator infusion system (Bracco, Monroe Township, NJ, USA). Details of the scan and reconstruction protocols have been described previously [22].
Image analysis
Static analysis
In seven patients, where 68Ga-PSMA PET/CT scans were available, the 68Ga-PSMA PET/CT scans were co-registered to the two 82Rb PET/CT scans using the low-dose CTs (Hybrid Viewer, Hermes Medical Solutions, Stockholm, Sweden). The tumor volumes of interest (VOIs) were drawn directly on the 68Ga-PSMA PET/CT images and subsequently transferred to the 82Rb PET/CT static images. The tumor VOIs were automatically drawn in two ways, with a fixed SUV threshold of 6 and by using a 30% threshold of max at the tumor site. An example of a high-risk patient with PSMA SUV 6 fixed threshold VOIs is shown in Fig. 1a with corresponding test and retest 82Rb PET/CT scans. Multiple different SUV values and different threshold percentages were evaluated before selecting SUV 6 and 30% as the most optimal threshold. One patient had a tumor in the prostate basis, the PSMA activity of which was confluent with the urinary PSMA activity. Consequently, it was necessary to mask the bladder manually before the automatic VOI drawing.
In the remaining three patients, T2-weighted images of the MRI scans were co-registered to the 82Rb PET/CT scans using the low-dose CT as a bridge. The tumor VOIs were drawn directly on the MRI by visual guidance. An example of a low-risk patient with MRI-guided VOIs is shown in Fig. 1b, including both 82Rb PET/CT scans of the patient.
The static image series (3 to 7 min post injection) were used for SUV analysis. Image analysis was performed using Hermes viewer (Hermes Medical Solutions, Stockholm, Sweden).
Dynamic analysis
The tumor VOIs described above were transferred to the dynamic PET series, and time-activity curves were extracted. To obtain a blood input function for calculation of K1, we utilized the method developed by Tolbod et al. [13], which uses an image-derived input function from a separated scan of the heart. In short, a separate 82Rb PET/CT scan over the heart was performed in connection with the pelvic scan with the same tracer dose and infusion profile. Image-derived input functions were extracted with cluster analysis from both the heart and pelvic scans, and subsequently, the heart image-derived input function was delay- and dispersion-corrected to the pelvic image-derived input function. This method was previously validated against 15O-H2O PET with input function obtained using arterial blood sampling [13]. Kinetic modeling was performed using a one-tissue compartment model.
Statistical analysis
Since the clinically relevant parameter is a relative change in blood flow, the repeatability data for both K1, SUVmax, SUVmean, and SUVpeak were log-transformed. The data were visually inspected for normality using Q-Q plots of the differences, and based on Bland-Altman plots, the variation does not seem to depend on the average [24]. The repeatability of the method was calculated by the method described by Bland and Altman [25]. The within-patient/within-lesion coefficient of variance, repeatability, and intraclass coefficients (ICCs) were calculated for both K1, SUVmax, SUVmean, and SUVpeak. The statistical parameters and formulas used are described in details in Lodge et al. [15]. Bland-Altman plots are presented in the original scale with back-transformed limits of agreement using the methodology of Euser et al. [26]. Sample size calculations were performed to detect relative changes of − 20%, − 30%, and − 50% using a two-sided significance test of no difference for paired log-normally distributed data with a significance level of 5% and a power of 95%. For these sample size calculations, the standard deviation for the difference between the logarithm of the test and the logarithm of the retest was used.
Study data were collected and managed using REDCap (Vanderbilt University Medical Center, Nashville, TN, USA) electronic data capture tools, hosted at Aarhus University [27]. Data analysis was performed using Stata version 15.1 (StataCorp LLC, College Station, TX, USA).