NEMA test procedures for Discovery MI were used to perform NEMA tests. Almost all measurements were evaluated per NEMA Standards Publication NU2-2012 . We made use of the NEMA processing tools contained in the Discovery MI software. Some additional measurements were performed according to NEMA NU2-2018 . Before NEMA testing, a normalization scan and well counter calibration were performed. Additional timing resolution and energy resolution tests were performed.
Normalization and well counter correction
A normalization scan was performed before the NEMA tests started. We used the calibration/daily quality assurance phantom, which is a 27.6-cm long, 12.5-cm outer diameter (1.3-cm thick) annulus radioactive source filled with 68Ge in an epoxy matrix. This phantom is provided with the scanner for calibrations and daily quality assurance. A well counter calibration was performed with 18.38 MBq (at the start of the acquisition) of 18F-Fluorodeoxyglucose (18F-FDG) in a uniform cylindrical phantom with a diameter of 20 cm and a length of 18 cm. This process provided a normalization sinogram and the activity correction factor.
18F-FDG was mixed with a small amount of dye to enhance the visibility of the radioactive liquid. Little drops were suspended on a plate and drawn up by capillary tubes so that the axial length of the drop in the tubes was less than 1 mm. Three point sources were made and inserted in the spatial resolution phantom. The sources were positioned at 1, 10 and 20 cm in the Y-direction from the centre of the field of view (FOV). Their positions were adjusted to within ± 1.0 mm of the corresponding nominal positions in the PET’s scan FOV. Data were collected at the centre slice of the FOV and at one eighth from the edge of the axial FOV. Every acquisition consisted of at least 500.000 counts. For NEMA, the images were reconstructed with FBP and VPHD, non-TOF OSEM reconstruction with 34 subsets and 4 iterations without PSF modelling. An additional reconstruction was made using VPHD-S. For each spatial orientation, full width at half maximum (FWHM) and full width at tenth maximum (FWTM) were calculated for every reconstruction and every point source and averaged for the acquisition at the centre of the FOV and at 1/8th axial FOV. FWHM and FWTM were statistically compared between the 3 reconstruction algorithms using correlated sample ANOVA followed by Tukey’s HSD post hoc testing; significance was called at p < 0.05. FBP and VPHD were compared on data from the 4-ring systems in Stanford and Uppsala taken from  by use of paired t tests.
A plastic tube (70-cm long and with a lumen of 1 mm) was filled with 16.02 MBq of 18F-FDG at time of filling. The activity was left to decay until it was lower than 4 MBq, in order for count losses to be negligible and random coincidences to be low. With the aid of a dedicated source holder and dedicated software, this line source was placed at the centre of the FOV and at a 10-cm radial offset in the Y-direction. At each position, 5 1-min scans were made with the number of aluminium sleeves around the plastic tube ranging from 1 to 5. The aluminium ensures the annihilation of all positrons and provides increasing attenuating material. Results were then extrapolated to give the scanner sensitivity with no attenuation material. Data were collected directly from sinograms corrected for randoms. Randoms were subtracted from prompts to obtain trues-only sensitivity results.
Scatter fraction, count losses and randoms
This test measures the count rate performance of the scanner across a range of radioactivity levels. The scatter fraction portion of this test measures the sensitivity of the scanner to coincidence events caused by scatter.
A 70-cm-long line source with an inner diameter of 3.2 mm containing 851.20 MBq 18F-FDG at the start of the acquisition was placed in the NEMA scatter phantom, a 70-cm-long polyethylene cylinder with a diameter of 20 cm. The activity was high enough to achieve count rates beyond the expected peak of the noise equivalent count rate. The phantom was secured from rolling with rubber foam wedges and elevated with a paper stack over the patient table until its centre-line aligned with the scanner’s central axis. The acquisition started with 17 frames of 15 min, without delay between the frames, and ended with 7 frames of 25 min, each with a delay of 25 min. NEMA specifications were used to derive the trues, randoms, scatter and noise-equivalent count rate (NECR) from the prompts dataset in each frame. Randoms were estimated using singles rates and the coincidence timing window that is defined by the manufacturer for clinical use.
Quantitation accuracy: corrections for count losses and randoms
This test compares the trues rate inferred from count losses and randoms corrections with the trues rate extrapolated from measurements with negligible count losses and randoms. Calculations were done on the data acquired for the test of scatter fraction, count losses and randoms as described above, reconstructed by non-TOF OSEM with 16 subsets and 3 iterations without point-spread function modelling. In each time frame, the absolute value of the error was calculated from a linear fit of the activity concentrations measured below peak NECR using 41 of the 53 slices comprising the phantom volume (the 6 end-slices were ignored); the mean, maximum and minimum error over these 41 slices were derived. The accuracy of the corrections for count losses and randoms was expressed as the maximal absolute value of the error below peak NECR.
Image quality, attenuation accuracy and scatter correction
The image quality (IQ) test simulates a PET/CT whole body clinical case. The 4 spheres of the IQ phantom with a diameters of 10, 13, 17 and 22 mm were filled with 21 kBq/cc 18F-FDG concentration whereas the 2 spheres with a diameter of 28 and 37 mm were filled with water. The background of the phantom was filled with 5.27 kBq/cc 18F-FDG, in order to yield a 4:1 concentration ratio between the radioactive spheres and the background volume. The phantom has a cylindrical insert with a diameter of 5 cm, containing a low-density material with an average density of 0.3 g/ml to simulate lung tissue. This insert is positioned in the centre of the phantom to have a non-uniform background. The IQ phantom was centred in the scan FOV. Additional activity (120 MBq) was placed outside the FOV (70-cm-long line source with 18F-FDG in the NEMA scatter phantom) to represent scatter radiation. Three acquisitions (with time correction for radioactive decay) were made and reconstructed with the VPFX reconstruction algorithm using a 384 × 384 matrix, CT attenuation correction, 4 iterations, 34 subsets, corrections for randoms, scatter, dead time and normalization. IQ was reported in terms of contrast recovery (CR) and background variability (BV) for the radioactive and non-radioactive spheres and averaged over the three acquisitions for increased reliability. The lung error (LE) is the average of LE from 48 slices out of the 53 slices in the PET image, per .
The same acquisitions were reconstructed with the Q.Clear reconstruction algorithm, with a beta value of 50. This low beta value, the same that was used in , was selected with the intent of matching the noise levels in the Q.Clear and VPFX images. CR and BV were compared between VPFX and Q.Clear reconstructions by paired t tests. For each sphere diameter and reconstruction method, CR and BV were compared amongst the 3-ring system at Bruges and the 4-ring systems at Stanford and Uppsala by calculation of 95% confidence intervals. Significance was called at p < 0.05.
An additional acquisition was performed according to NEMA NU2-2018. The 6 spheres of the IQ phantom were now filled with 21.9 kBq/cc 18F-FDG concentration, whereas the background was filled with 5.5 kBq/cc 18F-FDG concentration, again yielding a 4:1 concentration ratio between the radioactive spheres and the background. Phantom positioning and image reconstruction were identical to those described above for the NEMA NU2-2012 testing. An offline analysis tool was used to derive CR and BV values.
Timing and energy resolution
Timing resolution was calculated from the acquisition of a line source filled with 16 MBq of 18F-FDG and suspended in the centre of the FOV in the axial direction in the smallest aluminium sleeve used in the NEMA sensitivity test. Energy resolution was calculated from an acquisition with a 59 MBq 68Ge annular phantom (the scanner’s calibration phantom). Three hundred million counts were taken to acquire the timing spectrum. Measurement of the timing resolution FWHM was based on a 3-point fit of the peak of the timing spectra for each crystal pair after removal of the randoms. The energy spectra were smoothed with a boxcar filter. The timing and energy resolution were calculated for every detector crystal and averaged for the entire system.
According to NEMA NU2-2018, a PET/CT alignment scan was performed to analyze the registration between the PET and the CT image. An 8-min single-bed-position PET scan was made of the VQC phantom. This phantom consists of 5 point sources of 0.15 MBq 68Ge which are visible on both PET and CT images and are embedded in a moulded polyurethane foam. Images were reconstructed using VPFX, in a 256 × 256 matrix, with 16 subsets and 3 iterations and using a standard Z-axis filter with 5.0-mm filter cutoff. Dedicated software was used to determine the coordinates of every point source on both PET and CT images. The difference between the PET and CT coordinates along the 3 axes as well as the total distance between the PET and CT positions were calculated for each point source.
Clinical imaging comparison with Discovery 710 PET/CT
A patient with local recurrence of nasal melanoma was referred to PET for follow-up after chemotherapy and radiation. The patient had a BMI of 24.2 and was injected with 3 MBq/kg for a total of 180 MBq 18F-FDG. Ninety minutes after injection, a first TOF acquisition was made on a Discovery 710 PET/CT camera (GE Healthcare, Milwaukee, WI, USA). Two hundred minutes after injection, a second TOF acquisition was made on the Discovery MI 3-ring. The acquisition time at both systems was 13.5 min (1.5 min per bed position). Images were reconstructed using the Q.Clear algorithm, with a beta value of 400 for the Discovery 710 acquisition and 1000 for the Discovery MI 3 acquisition.