Performance Characteristics of the Digital Biograph Vision PET/CT System

Spatial Resolution

NEMA NU 2-2012 specifies using a ¹⁸F-FDG point source smaller than 1 × 1 × 1 mm. However, for measuring the spatial resolution on the Vision with smaller crystals, a smaller point source could improve test results (14). NEMA NU 2-2018 therefore recommends purchasing a ²²Na point source (11). Thus, a 74-kBq, 0.25-mm-diameter spheric ²²Na point source (Eckert and Ziegler Isotope Products) was used.

To comply with the NEMA NU 2-2012 standard, measurements were also performed using an ¹⁸F-FDG point source. At the start of data acquisition, a point source of 3.7 kBq of ¹⁸F-FDG with a length of about 0.3 mm was prepared in a 0.5-μL syringe (Hamilton Co.). The activity at the start of acquisition was sufficiently low to keep dead-time losses and randoms below 5% of total events. Data were acquired at several positions in the FOV (in x, y, and z directions of 0, 1, and 3.3 cm [≈⅛ FOV_Z]; 0, 1, and 13 cm [≈½ FOV_Z]; 0, 10, and 3.3 cm; 0, 10, and 13 cm; 0, 20, and 3.3 cm; and 0, 20, and 13 cm, respectively). At least 2 × 10⁶ coincidence counts were acquired in each position.

The obtained sinogram data were Fourier-rebinned and reconstructed by filtered backprojection using only a standard ramp filter into a 880 × 880 × 307 matrix with a 0.8 × 0.8 × 0.8 mm voxel size. The data were reconstructed without attenuation and scatter correction. The spatial resolution was determined according to NEMA NU 2-2012 as the full width at half maximum (FWHM) of the point-spread function (PSF) (7).

Scatter Fraction, Count Losses, and Randoms Measurement

The phantom used for these measurements was a 70-cm-long polyethylene cylinder (20-cm diameter), with a line source inserted axially into the cylinder 4.5 cm radially from the center. At the start of data acquisition, the line source was filled with 1.2 GBq of ¹⁸F-FDG to achieve count-rates beyond the expected peak of the NECR. Data were acquired for over 12 h, resulting in 35 frames, each with a 240-s acquisition time and an interframe delay of 960 s. To account for randoms, online randoms subtraction was applied using the delayed-coincidence-time-window technique (15). Subsequently, scatter fraction and NECR were determined (7).

Sensitivity

Sensitivity was measured using a 70-cm-long polyethylene tube (inner diameter, 1 mm; outer diameter, 3 mm) filled with 5.07 MBq of ¹⁸F-FDG (at acquisition start) and placed inside 5 concentric aluminum sleeves of equal length with known diameters (7). Five data sets were acquired associated with each of the 5 sleeves (starting with all 5 aluminum sleeves and finishing with a single sleeve) for 300 s each. The measurements were conducted at the center of the transaxial FOV and repeated at a 10-cm radial offset. Randoms subtraction was applied using the delayed-coincidence-time-window measurement. Next, the system sensitivity was computed (7).

Accuracy of Count Losses and Randoms Corrections

This evaluation used data acquired from the scatter fraction and count-rate measurements. Data were corrected for dead time, randoms, scatter, and attenuation (7). The parameters of the low-dose CT scan used for attenuation correction were an x-ray tube current of 80 mAs, a tube voltage of 120 keV, and a spiral pitch factor of 0.8. Scatter was corrected using the extended single-scatter simulation algorithm (16), which discriminates the scattered annihilation radiation according to its differential TOF.

Subsequently, the corrected data were reconstructed using the standard provided whole-body reconstruction algorithm, that is, an ordinary Poisson ordered-subset expectation maximization (OP-OSEM) 3-dimensional (3D) iterative algorithm (17) with 8 iterations, 5 subsets, and no filtering. An image matrix size of 220 × 220 was used. By extrapolating the true rate for low activity concentrations (where count losses and randoms can be neglected), count-rate accuracy was estimated.

Image Quality, Accuracy of Attenuation, and Scatter Corrections

The PET NEMA NU2 image-quality phantom (IQ phantom) (PTW) was used to evaluate image quality. The background activity concentration at the start of data acquisition was 5.7 kBq/mL ¹⁸F-FDG. The 4 smallest spheres were filled with a sphere-to-background ratio of 8:1 for the first set of scans and 4:1 for the second set of scans. The remaining 2 largest spheres were filled with nonradioactive water. The IQ phantom was positioned with all spheres aligned in the axial and transaxial center of the FOV. For simulation of a clinical situation with activity outside the FOV, the cylindric scatter phantom was placed axially next to the IQ phantom (7). The line source inside the scatter phantom was filled with approximately 116 MBq of ¹⁸F-FDG at the start of both data acquisitions.

Two sequential measurements of 240 s each were acquired for a single bed position after a low-dose CT scan for attenuation correction. Acquisitions were done to simulate a whole-body scan (emission and transmission) of 100 cm total axial imaging distance in 30 min of emission imaging. All data were corrected for random coincidences (smoothed random correction), normalization, decay, dead-time losses, scatter, and attenuation. The data were reconstructed using an OP-OSEM 3D-iterative algorithm with 8 iterations and 5 subsets, applying PSF and TOF into a 440 × 440 matrix with a voxel size of 1.6 × 1.6 × 1.6 mm. The percentages contrast obtained for hot and cold spheres, the background count variability for each sphere, and the accuracies of attenuation and scatter corrections were evaluated.

Coregistration Accuracy

For measurement of coregistration accuracy, a vial was filled with 59.6 MBq of ¹⁸F-FDG activity (in 0.1 mL at data acquisition start) and CT contrast (240 mg/mL) to a volume of no more than 1.4 mL. In total, 115 kg in nine 11.5-kg increments (which includes the weight of the 11.5-kg L-fixture) were placed on the patient bed. The foam holders provided by the manufacturer were positioned on the L-fixture at 6 locations, 3 points on each of 2 transaxial planes. In the transverse direction (with the coordinate system origin x and y = 0 and 0 cm, respectively), the foam holders were placed at nominal locations of 0 and 1 cm, 0 and 20 cm, and 20 and 0 cm on the x and y axes, respectively. In the axial direction (with the coordinate system origin z = 0, located at the edge of the PET axial FOV), the foam holders were placed in the center of the PET axial FOV (z = ½ PET axial FOV) and at 5 and 100 cm from the tip of the patient table (11). Per location, a low-dose CT scan was performed first and followed by a 3-min PET scan. The total of 6 measurements was performed to determine the centroid within the PET and CT data sets and, subsequently, to calculate the length of the 3D vector between the CT centroid and the PET centroid (i.e., the coregistration error) (11).

Timing Resolution

The TOF resolution was calculated using the acquired scatter data used for NECR performance, according to a new proposed method (11,18). The timing resolution was calculated as the FWHM of the time distribution of events, after correction for scatter, randoms, and the position of the line source.

EARL Performance

EARL performance on the Vision was measured to evaluate its ability to meet current EARL guidelines and foreseen 2019 EARL guidelines (19–22).

Measurements were conducted according to EARL standard operating procedures (23). The images were reconstructed using an OP-OSEM 3D-iterative algorithm with 4 iterations and 5 subsets, applying TOF, into an image matrix size of 220 × 220, resulting in a voxel size of 3.3 × 3.3 × 1.6 mm. This reconstruction was repeated with 8 iterations, a 5-mm FWHM gaussian filter, and an image matrix size of 220 × 220. Additional reconstructions applied resolution modeling, with and without a 7-mm FWHM gaussian filter and with matrix sizes of both 220 × 220 and 440 × 440. Reconstructions were performed with corrections for attenuation, scatter, normalization, decay, and dead time. Moreover, all (nonsmoothed) reconstructed images were filtered using gaussian kernels with a FWHM ranging from 1 to 10 mm in 1-mm steps to derive the optimal combination of reconstruction methods, settings, and filtering to achieve EARL-compliant performance. The latter procedure will allow the definition of EARL-compliant reconstruction protocols for the Vision.

To provide the reader with insight on a possible activity or scan-time reduction that may be applied in clinical settings, EARL decay measurements were performed according to a previously published method (24).

Patient Study

A patient study is included to provide the reader with a first impression on clinical performance. We do not intend to provide a detailed and valid intersystem comparison.

A 67-y-old woman (1.64 m tall and weighing 73.1 kg) diagnosed with parkinsonism was injected with 200 MBq of ¹⁸F-FDG. At 30 min after injection, a brain PET/CT study was first performed on an mCT system for 15 min and then repeated on the Vision (∼5 min after the completion of the mCT study). Data from the mCT were reconstructed using 3D-TOF OP-OSEM with 6 iterations, 21 subsets, and resolution modeling. No filter was used, and the resulting image size was 400 × 400 with a voxel size of 2 × 2 × 2 mm. Data acquired on the Vision were reconstructed using 3D-TOF OP-OSEM with 8 iterations and 5 subsets, with resolution modeling into a 440 × 440 image matrix with a voxel size of 1.6 × 1.6 × 1.6 mm.

Another patient, a 56-y-old woman (1.54 m tall and weighing 67.3 kg) diagnosed with metastasized non–small cell lung carcinoma was injected with 215 MBq of ¹⁸F-FDG. At 60 min after injection, a whole-body PET/CT study was performed on an mCT system using 3-min PET acquisitions per bed position. Data were reconstructed using 3D-TOF OP-OSEM with 3 iterations, 21 subsets, and resolution modeling. A gaussian filter of 5 mm was applied to the reconstructed images, and the resulting image size was 400 × 400 with a voxel size of 2 × 2 × 2 mm. Subsequently, measurements were repeated on the Vision using 3-min PET acquisitions per bed position. The vendor-recommended reconstruction protocol was applied, that is, 3D-TOF OP-OSEM with 4 iterations, 5 subsets, resolution modeling, no filtering, an image matrix size of 220 × 220, and a voxel size of 3.3 × 3.3 × 1.6 mm.

The patient study was approved by the medical ethics review board of the University Medical Center Groningen, and both patients provided written informed consent.