Decaying-source experiment
In order to observe the count-rate response of the scanner at high activities, a decaying phantom experiment was performed. An initial activity of 1160 MBq of 11-carbon was placed in the myocardial compartment of an anthropomorphic torso phantom (phantom model ECT/TOR/P, cardiac insert model ECT/CAR/I, both Data Spectrum Corporation, Durham, USA), which represents the upper torso of average-to-large male and female patients. Two fillable defects (3.8 ml each, filled with water) were fixed in position in the myocardial chamber in the septal and lateral walls in order to evaluate if defects could be viewed over the range of activities imaged. The volume of the myocardial chamber with 2 defects attached was 110 ml, giving an activity concentration of 10.54 MBq/ml at the beginning of the scan. All other compartments of the phantom were filled with water to provide a scattering environment similar to a clinical situation.
PET scanning was performed on a 3T Siemens Biograph mMR scanner (Siemens Healthcare GmbH, Erlangen, Germany). CT images were employed for attenuation correction purposes rather than the inbuilt MR-based routine. CT scanning (140 kV, 100 mA, 0.5-s rotation, 40-mm collimation) was performed on a PET-CT scanner (GE Discovery 710, GE Healthcare, Waukesha, USA), and the resulting CT images were registered to the MR-based attenuation map using a rigid registration algorithm (Niftyreg software, University College London, UK). The registered CT data were converted to attenuation values at 511 keV using a calibration curve from the PET-CT scanner and used on the PET-MR scanner for attenuation correction of PET emission sinograms.
PET list-mode data was acquired over a 100-min acquisition. List-mode data was rebinned into 5-s frames (a frame time typically used in PET MPI) with a 2-min gap between each frame to reduce data storage requirements. Sinograms were reconstructed on the scanner front-end (OSEM 3 iterations, 21 subsets, 4-mm post-smoothing filter, 344 × 344 image matrix). All corrections for decay, dead time, scatter and randoms were utilized as implemented on the scanner.
Cardiac perfusion phantom
In order to experimentally simulate a clinical MPI study using simultaneous PET and MR imaging, we used an in-house designed and built myocardial perfusion phantom, the construction and operation of which has previously been described in detail [22, 23]. Briefly, water is pumped through an MR-safe myocardial perfusion phantom placed in the PET-MR scanner. The phantom is representative of the large thoracic vessels and of the heart of a 60-kg body weight subject. It is composed of 4 cardiac chambers (120 ml each) and associated thoracic vessels (aorta, pulmonary artery, pulmonary vein, vena cava). Supporting precision pumping and monitoring mechanisms control the transport of water through the phantom. Myocardial perfusion is controlled in real time by continuously sampling the flow rate by means of high-precision digital flow meters (Atrato, Titan, Sherborne, UK) and providing re-adjustment of the rotation speed of roller pumps through a feedback mechanism. True perfusion values were obtained by means of measurements of the distribution volume for the radioactive tracer and gadolinium-based contrast agent (GBCA) and dividing the flow rate by this value. All controls are handled remotely from a custom-written LabVIEW application (LabVIEW Professional Development System 2014, National Instruments, Austin TX, USA) running on a dedicated workstation and remotely controlled using an iPad application (Dashboard for LabVIEW, National Instruments, Austin TX, USA). We utilized a non-recirculating model; therefor, after injection, no radiotracer or GBCA re-entered the system after transit through the phantom.
The MR perfusion sequence consisted of a clinically utilized protocol, namely a 2D TurboFLASH saturation recovery gradient echo sequence (TE = 1 ms, TR = 164 ms, Flip angle = 10°, slice thickness = 8 mm, pixel spacing = 1.875 mm, matrix size 144 × 192 voxels, with temporal resolution of 1 image per cardiac beat). MR images were acquired in a single transverse plane identified by markings on the phantom, the locations of which correspond to a known dispersion volume for the GBCA and radiotracer. Cardiac output flow rate was set to 3 l/min, with true myocardial perfusion rates (P
T
) set to 3 ml/g/min. Total PET-MR scan time for each value of P
T
was 4 min. We utilized a dual-bolus injection technique developed by our group in order to avoid signal saturation effects [24], whereby a prebolus of GBCA is injected via a contrast injector with concentration of 0.0075 mmol/kg, followed 30 s later by the main GBCA bolus of 0.075 mmol/kg simultaneously with the radiotracer. The prebolus is free from saturation and T2* effects (due to the lower concentration) and is used in perfusion calculations as an arterial input function (AIF). All injections to the phantom were performed from the contrast injector at a rate of 4 ml/s.
In order to analyse the effects of high activity on the quantification of perfusion, we performed injections of the radiotracer to the perfusion phantom at a range of injected activities of [18F]F- (A
INJ = 252, 398, 594, 804 and 997 MBq). The radiotracer and the main GBCA bolus were preloaded into tubing connected to the vena cava of the phantom and injected simultaneously 1.5 min after the scan start time.
PET data were acquired in list mode, rebinned into 5-s frames, reconstructed on the scanner front-end (OSEM, 3 iterations, 21 subsets, 4-mm smoothing filter, 344 × 344 matrix). MR-based attenuation correction (MRAC) was performed using the standard dual-point VIBE T1-weighted Dixon sequence available on the scanner. After each experiment, each MRAC was visually inspected in order to check for any errors in tissue classification such as fat/water tissue inversion, which can influence the linear attenuation coefficients applied to the PET emission data [25].
The terms ‘flow’ and ‘perfusion’ have been used interchangeably in both PET and MR literature. Owing to the fact that rates of liquid through our phantom were calibrated in terms of ml/g/min (i.e. units of perfusion) and K
1 values from PET kinetic modeling were in the same units, we opt to keep consistency with terminology and use the term ‘perfusion’ rather than ‘flow’ (i.e. units of ml/min).
Image analysis
Decaying-source phantom
Reconstructed images of the decaying phantom experiment were analysed in PMOD (v3.7, PMOD Technologies, Zurich, Switzerland). An automated 25% isocontour was drawn around the myocardial wall and a time-activity curve (TAC) was derived. With known activity in the phantom (as assayed using a dose calibrator within 5% accuracy), the average activity concentration was plotted against true activity in the field of view (A
FOV) of the scanner. Count-rate data (total prompts, randoms, trues, live-time fractions, etc.) for each time frame were extracted directly from a text file as part of the sinogram headers. Noise equivalent count rate (NECR) was calculated using the NEMA performance-testing equations (with delayed randoms) [26].
Perfusion phantom
Dynamic PET perfusion data were analysed in PMOD v3.7 to produce TAC of the aorta and myocardium compartments. MR images were analysed in OsiriX (OsiriX 64-bit, version 8.0.2, Pixmeo SARL, Geneva, Switzerland) to extract time-intensity curves (TIC). A single ROI of 1.6 cm (matching the tubing diameter) was placed over the aorta of the phantom, and ROIs of 4-cm diameter were placed over the left and right myocardial sections. Positioning of ROIs on the PET image plane corresponding to the MR image plane was determined from fusion of the dynamic 3D PET and 2D summed dynamic MR images in PMOD. ROIs were placed on PET images over the same spatial extent as the MR ROIs. PET VOIs were 8.1-mm thick (4 PET slices) in the axial dimension in order to match the slice thickness of the MR data (8 mm). All ROIs and VOIs were drawn on one set of PET and MR images (A
inj = 252 MBq), saved and then translated on to all subsequent scan data. Data for the left and right myocardial compartments were averaged to a single curve. We thus produced a set of TAC for PET data and TIC for MR data for the aorta (C
A(PET) and C
A(MR) respectively) and myocardial compartments (C
M(PET) and C
M(MR) respectively) over the range of activities administered to the phantom. C
A(MR) was derived from the prebolus peak of GBCA and not the main bolus. An estimate of total activity in each PET time frame was calculated by using a 10% threshold of the images via PMOD and multiplication by the VOI volume.
PET data was modeled using a one-tissue compartment model and two rate constants K
1 (uptake rate constant in units of ml/g/min) and k
2 (clearance rate constant in units of min−1) using the PKIN software module of PMOD. We assume an extraction fraction of 1.0 of [18F]F- due to the lack of any metabolic processes in the phantom, and thus, in this setup, K
1 is entirely representative of perfusion. MR data was modeled in a similar fashion using the same single tissue compartment model in PMOD in order to provide relative measurements of MR-based perfusion that are independent of the effects of PET detector dead time. Area under the curve (AUC) for each TAC and TIC was also calculated as part of the modeling process.