The study population consisted of a retrospective cohort of 30 consecutive patients who underwent [15O]H2O PET perfusion at rest and adenosine stress at Helsinki University Hospital for evaluation of myocardial ischemia. A history of coronary artery disease, diabetes, hypertension, smoking, and hypercholesterolemia was collected from hospital records. The exclusion criteria were the inability to complete the standard PET perfusion protocol. Written informed consent was obtained from all participants. The study was performed according to the Declaration of Helsinki, and the study protocol was approved by the local ethics committee (HUS/1226/2019) and the Helsinki University Hospital’s institutional research board.

All patients were imaged using a standard clinical [15O]H2O myocardial PET perfusion protocol. The patients were instructed to abstain from caffeine for 24 h prior to the PET study. Sequential acquisition of helical low-dose computed tomography (CT) for attenuation correction and PET perfusion scans were performed using a digital 4-ring detector time-of-flight PET/CT system (Discovery MI PET/CT, GE Healthcare, Waukesha, WI, USA). Rest perfusion was performed first, and after a 10-min pause for radiotracer clearance stress perfusion was obtained. The target dose was 600 MBq per PET scan (range 416–715 MBq). A standard 6 min adenosine stress was used during the stress perfusion with radiotracer injection 2 min after the initiation of adenosine infusion at 140 µg/kg/min. The scan parameters were identical in both scans. List mode PET data were acquired and reconstructed as dynamic and divided into frames as 14 × 5 s; 3 × 10 s; 3 × 20 s; 4 × 30 s (total time: 4 min 40 s). PET images were reconstructed with the iterative Q.Clear algorithm (GE Healthcare, Waukesha, WI, USA) with 256 × 256 matrix size. The pixel size and slice thickness were set to 2.73 mm and 2.79 mm, respectively.

The assessment of SSO was done retrospectively for research purposes only from the time-activity curves (TACs) extracted from the dynamic PET data using Syngo.via (Siemens Healthineers Forchheim, Germany) software’s MM Oncology package. Spherical volumes of interest (VOI) were manually drawn in the liver (mean: 116 cm3 [50, 292] cm3) and spleen (mean: 11 cm3, [6, 25] cm3), and then activity concentration (kBq/ml) TAC as average values from the VOI region were exported in a tabular format (.csv).

Further processing of the ta TAC values was performed in Python v. 3.8 using custom codes. The average activity concentration values from PET data were transformed into average standardized uptake values (SUVavg) using the administered tracer dose and patient weight. The following parameters were extracted from the SUVavg TAC: Time to peak SUVavg in rest spleen, Maximum SUVavg value from the spleen TAC, Spleen-to-liver SUVavg ratio (SLR) both during stress and rest (SLR = Maximum splenic SUVavg/Maximum liver SUVavg) and the splenic SUVavg ratio (SAR = Maximum splenic activityStress/Maximum splenic activityRest).

No independent ground truth method (such as gadolinium-enhanced CMR in PET-MR) for the SSO assessment was available. However, in our patient population, the spleen maximum SUVavg and SLR values systematically decreased during adenosine stress showing SSO. Only two patients out of 30 had an opposing slight increase in the spleen maximum SUVavg and SLR values in stress compared to rest.

Quantitative global left ventricular myocardial perfusion values were obtained for rest and stress PET scans using Carimas™ (Cardiac Image Analysis System) software package [10]. PET images were read by nuclear medicine physician with more than 10 years of experience with PET perfusion imaging (VU) who established the diagnosis of myocardial ischemia based on the quantitative perfusion data. Previously validated cut-off values for stress myocardial blood flow (MBF) of 2.3 ml/min/g and for myocardial flow reserve (MFR) of 2.5 were used to define the presence of myocardial ischemia [2]. For one patient, the myocardial rest perfusion analysis was omitted due to poor quality of data.

Continuous variables are reported as mean ± standard deviation (SD) and median [interquartile range] for normally distributed and skewed data. Categorical variables are reported as numbers and percentages. Due to sample size (N = 30), a nonparametric Wilcoxon signed-rank test was chosen for comparison of stress and rest spleen SUVs at different time points and SLRs. TACs were used to evaluate the optimal time-points for quantification of the SSO phenomenon. Spleen rest TAC was compared with spleen stress TAC to evaluate the adenosine-induced SSO phenomenon during stress imaging. In addition, boxplot analyses were performed to investigate possible differences in SSO reaction in patients with diabetes, previously known coronary artery disease, or myocardial ischemia at PET perfusion. Boxplots displayed a sample median as horizontal line, a first and third quartiles as boxplot edges, whiskers, which were set to first quantile—1.5 ⋅ interquartile range and third quantile + 1.5 interquartile range, and outliers, i.e., values outside whiskers as circles. The statistical differences between these groups were assessed using Welch’s t-test. Correlations were analyzed using Pearson’s correlation.

A receiver characteristics operating curve (ROC) analysis was used to evaluate the optimal cut-off value to differentiate rest perfusion images from the stress perfusion. Area under ROC curve (ROC AUC), sensitivity, and specificity were calculated, and the optimal cut-off point was determined from Youden's index maximum.

The limit for statistically significant differences was set to p < 0.001. All statistical analyses were conducted in Python using SciPy (v. 1.10.0) and Pandas (v. 1.5.2), scikit-learn (v.1.2.1) libraries [11].