The study included 11 patients (9 men and 2 women, mean age 65 years, median 68 years, range 29–88 years) submitted to whole-body PET examination for staging of suspected lung cancer. Imaging was performed in the early morning, after 12 h of fasting. After measurement of body weight and serum glucose level, an antecubital vein was cannulated, and each patient was positioned on the bed of a Siemens Flow mCT40 system (Siemens, Erlangen, Germany) to undergo the preliminary X-ray CT scanning performed according to the conventional procedure [2].

A list-mode acquisition was started soon before the bolus injection of FDG (4 MBq/Kg body weight) with the field of view focused on the heart for 20 min. Immediately thereafter, acquisition mode was shifted, and eight whole body passages were performed from the skull to the mid-tights. In Supplementary Table 1 the scan time (expressed in minutes) per WB pass is reported. A last equilibrium scan completed the acquisition procedure to be analyzed for the clinical report.

Exploited reconstruction algorithm was the PSF + TOF 2i21s that combines an iterative reconstruction considering point spread function (PSF) correction and time-of-flight (TOF) information, with 2 iterations and 21 subsets; voxel size was: (2.03642 × 2.03642 × 5) mm^3.

The chest-centered part of the dynamic acquisition was binned according to the following frame sequence: 12 × 5 secs, 12 × 10 secs, 8 × 15 secs, 6 × 30 secs, 2 × 60 secs, 5 × 120 secs. By contrast, the eight subsequent whole-body scans were characterized by a slightly variable time sequence due to the length of the desired field of view. The last WB acquisition had an average duration of 14 min. Accordingly, the acquisition time of each slice was defined based on the DICOM metadata to perform an accurate correction of 18F physical decay.

An expert nuclear physician thus identified a ≥ 5 mL VOI on the descending aorta to estimate the input function (IF) defined by the activity concentration in the arterial blood at all times of the dynamic chest-centered frames and in the subsequent whole-body acquisitions.

A further VOI was drawn to loosely surround the tumor lesion in the last scan, and a mask was created to set all the outside voxels to 0. Data were transformed into standardized uptake value (SUV) images according to the conventional formulation [19] and the cancer lesion was defined as the set of all voxels with radioactivity concentration > 40% of the maximum value within the identified VOI.

For each patient, two sets of parametric images were set up. For the former, a Time Activity Curve (TAC) was generated in each tumor voxel throughout the eight whole-body acquisitions and six regression lines were computed considering all eight frames (1–8), the last seven ones (2–8), and so on, up to the last three (6–8). This analysis provided a first voxel-resolved description of FDG kinetics, made of a set of six parametric TAC images denoted, from now on, as TAC1–8, TAC2–8, …, TAC6–8. A regression analysis performed for each parametric TAC image allowed defining the involved voxels as accumulating (with a positive slope of the regression line) and releasing (with a negative slope of the regression line).

This preliminary evaluation was then compared with the conventional counterpart represented by the graphical approach to the compartmental analysis described by Patlak et al. [20]. According to this largely adopted model, the tracer is freely exchanged between the blood and a series of intermediate reversible pools that act as an entry gate for the irreversible compartment in which entered radioactivity can never escape. Once the equilibrium between blood and reversible compartments is reached at time t*, the irreversible accumulation within any given voxel is described by the equation:

$$_\left(t\right)=__^_\left(\tau \right)d\tau +__(t)$$

(1)

where \(_\) represents the net accumulation rate in the irreversible compartment (clearance), \(_\) (t) and Cb (t) indicate the tracer concentrations in tissue and blood, respectively, while \(_\) is the volume of the reversible compartments.

Dividing both sides for Cb (t), Eq. (1) can be re-written as:

$$\frac_\left(t\right)}_\left(t\right)}=_\frac_^_\left(\tau \right)d\tau }_\left(t\right)}+_$$

(2)

A standard Patlak analysis, performed against all the dynamical FDG-PET images, allowed the computation of the standard Patlak parametric image whose voxels contain the corresponding \(_\) values. However, reproducing the descriptive evaluation reported above, this same analysis was repeated six times, once again considering tumor radioactivity concentration at all frames, in the last seven, up to the last three. For each analysis, the IF was estimated by considering the value of the monoexponential function fitting its later measured values. Accordingly, these time-resolved Patlak analyses provided six parametric images, each one reporting the \(_\) slope of the corresponding regression line, which was denoted as \(_)}_\) up to \(_)}_\). Once again, each voxel was identified as accumulating or releasing according to the sign of the corresponding regression slope.

All data are reported as mean ± standard error of the mean (SEM). The statistical significance of the regression analysis in the case of the time-resolved approach was assessed by means of R2 > 0.3 for both the TAC parametric images and the Patlak parametric images. In the correlation analysis p < 0.05 was considered statistically significant.