Performance Characteristics of the Biograph Vision Quadra PET/CT System with a Long Axial Field of View Using the NEMA NU 2-2018 Standard

Spatial Resolution

Spatial resolution was measured at 6 different positions (Table 2; Supplemental Fig. 1; supplemental materials are available at http://jnm.snmjournals.org) using a point source with a 0.25-mm diameter containing 393 kBq of ²²Na (Eckert and Ziegler).

After acquiring at least 4 × 10⁶ true counts for every position, images were reconstructed in MRD 85, without the use of a postreconstruction filter, and with 3-dimensional TOF direct inversion Fourier transform backprojection, an analytic backprojection reconstruction method (18). Corrections were applied for detector normalization, dead time, radial-arc-correction decay, and randoms, but no scatter or attenuation correction was used.

Resolution was reported as the full width at half maximum and full width at tenth maximum of the point source’s spread in the radial, tangential, and axial directions. For each direction, average values over the 2 axial positions were calculated.

Count Rates: Trues, Randoms, Scatters, and NECRs

For count rate measurements, we used a solid polyethylene cylinder with an outside diameter of 20.3 cm and a 700-cm length. A 3-mm-wide and 70-cm-long polyethylene capillary was filled with 894 MBq of ¹⁸F and inserted into a 6.4-mm-wide hole running parallel to the central axis of the cylinder at a radial offset of 45 mm.

The cylinder phantom was placed onto the patient table in the center of the FOV and axially aligned with the PET/CT system. The line source insert was positioned close to the patient table, and foam blocks were used to elevate the phantom above the table to an axially aligned position. Data acquisitions in list mode were performed over the course of 700 min. However, the NEMA NU 2 criterion of waiting until true-event losses are less than 1.0% could not be achieved because of the intrinsic radioactivity of lutetium-oxoorthosilicate. Therefore, a previously described different methodology had to be used (25): count rates were measured using delayed coincidence windows, and the scatter fraction was calculated as a function of count rate.

Every 20 min, data were acquired for 240 s, and the acquisitions were binned into 35 individual sinograms of equal duration. Data were not corrected for variations in detector sensitivity, randoms, scatter, dead time, or attenuation effects.

Rates of total, true, scatter, and noise-equivalent counts were calculated as specified by section 4 of the NEMA NU 2-2018 protocol. Prompt and random sinograms were generated for each acquisition and each slice. Because of the extended AFOV of the Quadra, only slices within the central 65 cm of the AFOV were used for histogram generation.

Sensitivity

For sensitivity measurements, we used the same 70-cm-long polyethylene capillary as described above and filled it over a total length of 68 cm with an aqueous solution of 4.56 MBq of ¹⁸F. The line source was surrounded by 5 concentric aluminum sleeves of matching length and with known radiation attenuation. The setup was bedded on foam holders with negligible attenuation. One sensitivity measurement series was performed with the capillary axially aligned at the center of the AFOV, and the other series was performed with a 10-cm radial offset added to the first placement. The supports for the capillary stayed outside the FOV. By measuring the count rate while consecutively removing sleeves, we extrapolated the attenuation-free count rate, for example, the count rate of the naked line source (26). Data were acquired for 300 s for each sleeve.

Accuracy: Correction for Count Losses and Randoms

Data acquired for count rate measurements were used to estimate the accuracy of the correction of count losses due to detector dead time and due to random counts (randoms). Corrections for randoms, scatter, dead time, and attenuation were applied. For attenuation correction, a low-dose CT scan of the phantom was acquired with a 120-keV tube voltage, a 80-mAs tube current, and a pitch of 0.8. The CT image was reconstructed into a 512 × 512 matrix. Scatter was corrected for as described by Watson (27).

The PET image was reconstructed from MRD 85 data using ordered-subsets expectation maximization (OSEM)-TOF with 4 iterations, 5 subsets, and 2-mm gaussian postreconstruction filtering.

Image Quality, Accuracy of Corrections

A NEMA International Electrotechnical Commission body phantom (28) of 180-mm interior length was used for assessing image quality and the accuracy of attenuation and scatter corrections. The gravimetrically determined volume of the background compartment was 9,742 mL, and the fillable 6 spheres had internal diameters of 10, 13, 17, 22, 28, and 37 mm. The central lung insert filled with polystyrene beads was void of any activity.

The background activity concentration of ¹⁸F was 5.3 kBq/mL at the start of image acquisition, constituting our low-activity-concentration benchmark. A first measurement was taken with all spheres filled with a concentration 4 times that of the background as stated in the NEMA NU 2-2018 protocol (24). A second measurement was taken with a concentration 8 times that of the background. The phantom was axially aligned, with the spheres positioned around the center of the FOV. The cylindric scatter phantom was positioned adjacent to the sphere-containing phantom, and its line source was filled with 100 MBq of ¹⁸F at the start of the acquisition.

A single bed position was acquired for 30 min in list mode. Data were corrected for decay, normalization, scatter, randoms, and attenuation. The required attenuation CT scan was acquired before the PET measurements as described above. Images were reconstructed in MRD 85 using OSEM-TOF and point-spread function (PSF)-TOF with 8 iterations and 5 subsets. Both reconstructions were also performed using 4 iterations and 5 subsets. No postreconstruction filtering was applied. Activity spill-in into the cold lung insert was used to calculate an average residual error.

TOF and Energy Resolution

To measure the positional uncertainty of the coincidence event localization, we used the same CT and PET data as previously acquired for the NECR experiment, without corrections applied.

To determine the position of the line source, the first frame with activity below the peak NECR was reconstructed in MRD 85 using OSEM with 10 iterations and 5 subsets, with scatter, random, and attenuation correction but without decay correction. The method to calculate TOF resolution is described in section 8 of the NEMA NU-2 2018 standard and was also described by Wang et al. (29).

For measuring the energy resolution of the scanner, we used the same data but without any corrections applied. This measurement is not part of the NEMA NU 2-2018 standard, but it is based on the same method as for the TOF resolution and was previously described (30). An image reconstruction was performed for determining the line source centroid, with scatter, random, and attenuation correction but without decay correction. Trues were assumed to be within a perpendicular distance of ±20 mm of line source data, and thus counts at ±20 mm were assumed to come from scatter, randoms, and background. For each crystal, an energy histogram was generated using all events within a distance of −20 and +20 mm. The weighted combination of counts at −20 mm and +20 mm, as done in NEMA count-rate studies, was used to estimate the background (scatter and randoms). All crystal peaks were aligned and added in a common energy histogram (Supplemental Fig. 2). The energy resolution was defined as the full width at half maximum of the energy spectrum so obtained. For comparison, the energy resolution was also measured using a more conventional method, by placing a 19-cm-long line source containing 19.19 MBq of ⁶⁸Ge without a scattering medium at the center of the FOV.

PET/CT Coregistration Accuracy

Coregistration accuracy between the PET and the CT images was measured with a vial of 13.3-mm diameter and conical bottom and filled with an aqueous solution of 0.2 mL of 370-MBq ¹⁸F and 1 mL of CT contrast medium (Ultravist 370; Bayer Vital) according to the NEMA NU 2-2018 document (24). CT images were reconstructed into a 512 × 512 matrix and slice thickness of 0.6 mm, and PET images were reconstructed using OSEM-TOF with 10 iterations and 5 subsets, without attenuation correction or postreconstruction filtering.

Human Studies

An oncologic female patient (age, 81 y; height, 160 cm; weight, 57 kg) participating in a clinical study (31) was scanned 60 min after administration of 191 MBq of ¹⁸F-FDG. A single bed position was acquired for 10 min. Eight images were reconstructed by binning the list-mode data into 10-min, 6-min, 4-min, 3-min, 2-min, 1-min, 30-s, and 15-s frames. Images were reconstructed using PSF-TOF with 4 iterations, 5 subsets, and a gaussian postprocessing filter of 2 mm in full width at half maximum.

An isocontour threshold of 40% delineated the volume of interest of an ¹⁸F-FDG–avid lesion in the 10-min frame, and a spheric volume of interest with a diameter of 5.1 cm was placed in the center of the liver in the same frame. Both volumes of interest were then copied into the remaining frames. SUVs and coefficients of variation were computed for each volume of interest in every frame.

The human study (31) had been approved by the regional ethics committee, and the patient had signed an informed consent form.