Biograph Vision 450 PET/CT
The PET component of the Vision 450 is very similar to that of the version 600. It contains 38 blocks per ring for six rings along the axial FOV. Each block is subdivided into 4×2 mini-blocks (four mini-blocks in tangential position for two mini blocks in the axial position), that each contain an array of 5×5 LSO crystals of 3.2×3.2×20 mm3. The mini block is coupled to an array of 16×16 mm2 SiPMs. Table 1 summarizes the main properties of the Vision 450.
NEMA measurements
All measurements were performed following the NEMA procedure including spatial resolution, sensitivity, scatter fraction, noise equivalent count rate (NECR) and accuracy, timing resolution, image quality, and co-registration accuracy. The different experiments were analyzed using the software provided by the manufacturer.
Spatial resolution
Because of the expected high spatial resolution of the Vision, a 22Na point source (352 kBq) with a dimension (diameter: 0.25 mm) suitable to the crystal size was used for all measurements. The source was located in the FOV at given transaxial position (x,y) where x and y, expressed in centimeters, were at the following positions in a given z-plane: (0,1), (0,10), and (0,20) at a z-position of 1/8 axial FOV and ½ axial FOV. The precise position of the source was controlled through a pre-localization step using a specific source L-fixture developed by the manufacturer to ensure that the source was within +/- 2 mm for the (x,y) positions and +/- 0.25 mm for the z positions. Two million net true coincidences (defined as prompt minus random coincidences) were collected.
Data were reconstructed using a Fourier rebinning combined with a filtered back projection without attenuation, scatter corrections, and using a ramp filter. The voxel size was 0.83×0.83×0.83 mm3 (matrix size: 880×880×237).
Sensitivity
A 700 mm long polyethylene tube (inside diameter: 1 mm; outside diameter: 3 mm) was filled with 4.9 MBq of 18F at the start of data acquisition. The source was placed inside the sleeves and positioned at the center and at a 10-cm radial offset. Five data sets corresponding to five specific wall thicknesses were acquired for 300 s each.
Scatter fraction and NECR
A cylinder of polyethylene (700 mm long and a diameter of 200 mm) was used. A line source (inside diameter: 3 mm, outside diameter: 4.8 mm) was inserted axially into the cylinder at a radial position of 45 mm from the phantom center. The initial activity in the line source was 1156 MBq of 18F. Thirty-five frames of 240 s were acquired in 11.3 hours. The random coincidences were accounted for using the delayed coincidence technique.
Timing resolution
The timing resolution was calculated based on the experiment involving the measurement of scatter fraction and NECR. It was estimated as a function of concentration activity following the method proposed by Wang et al [10].
Image quality
The torso-shaped IEC phantom with six coplanar spheres (internal diameters of 10, 13, 17, 22, 28 and 37 mm) was used to evaluate image quality. A central cylindrical insert simulating lung tissue was added to the IEC phantom. The background was filled with a concentration of 5.4 kBq/mL of 18F-FDG while the four smallest spheres were filled so that the concentration ratio between the spheres and the background was 4:1 (the two largest spheres were filled with non-radioactive water). Acquisition was performed with the spheres’ center aligned with the axial center of the FOV. The phantom used for the scatter fraction and NECR was placed axially near the IEC phantom to simulate activity outside the FOV. The activity in the line source of this phantom was 153 MBq at the start of the acquisition. The acquisition time was set to 240 s.
Data were reconstructed using the TOF 3D ordinary Poisson ordered subset expectation maximization (3D OP-OSEM) algorithm with point spread function (PSF) recovery and TOF. Two matrix sizes were considered: 220×220 (voxel size: 3.2×3.2×1.65 mm3) and 440×440 (voxel size: 1.65×1.65×1.65 mm3). The reconstruction parameters were close to those used in routine clinical practice: four iterations and five subsets without post-filtering. For comparison purposes, data were also reconstructed following the reconstruction parameters matching those used in the seminal work of van Sluis et al. using the Biograph Vision 600 [9] for image quality assessment. The percentage contrast recovery (CR) for each sphere, the percent background variability, the residual errors for attenuation, and scatter corrections were then computed as specified by the NEMA guidelines [8].
Co-registration accuracy
A small vial filled with 35 MBq of 18F and a CT contrast agent (240 mg/mL) was used for this measurement. The vial was positioned at three transaxial coordinates: (0,1), (0,20), and (20,0) centimeters. A total weight of 115 kg was positioned on the table and the co-registration accuracy was evaluated at two axial positions (5 cm and 100 cm from the tip of the pallet). A CT scan followed by a 3-min PET acquisition was performed for the six positions considered. PET images were reconstructed using 3D OP-OSEM (10 iterations, 5 subsets, no post-filtering, 440×440 matrix size). The co-registration error was subsequently calculated using the software provided by the manufacturer as well as the maximum ratios defined in the NEMA guidelines [8].
Image quality comparison between Vision 450 and mCT
A direct comparison between the Vision and the analog-based mCT with the extended axial FOV [11] was conducted. This comparison will help guide how to sort patients between the two systems, and how to maintain consistency in longitudinal studies involving both systems. For this purpose, the IEC NEMA phantom with a sphere-to-background contrast of 4:1 was first acquired on the Vision for 240 s and immediately after on the mCT for 250 s to account for radioactive decay.
The noise level (described hereafter) typically observed using the mCT (with a standard clinical acquisition and reconstruction protocol) was set as the reference value for different numbers of net true coincidences corresponding to acquisition times of 4, 3, 2, 1.5, 1, and 0.5 min. The same number of net true coincidences was chosen for both systems (within 0.01%). The noise level computed using the Vision was matched to the reference noise level (described below) using different reconstruction set-ups (matrix size and reconstruction parameters) to determine the potential reduction of counts numbers for the same noise level. Indeed, a similar noise level could be reached with a lower number of counts thanks to the improved time resolution of the Vision. The noise level was computed using the image roughness (IR) as described by Tong et al [12] based on a single 27 cm3-spherical region-of-interest (ROI) so that the same computation could be used with patient data. IR was defined following:
where N is the number of voxels in the ROI, vi the value of voxel i and the mean of voxels in the ROI. The image roughness measures the pixel-to-pixel variation and is closely related to the visual perception of noise using a single image. This metric was preferred over background variability (BV) defined in the NEMA standards because BV captures more region-to-region variability, which is better adapted to quantify the variance of background measurement. As noise is not the only parameter relevant in this context, the CR for the hot and cold spheres in 3D considering the entire sphere volume (as opposed to the 2D evaluation used in the NEMA evaluation computed for a 2D cross-section of each sphere) was also computed for each final set-up so that a comparison of contrast could also be achieved. Each voxel intersected by the theoretical surface of the sphere was considered in the computation. The reference reconstruction parameters (mCT) for the 3D OP-OSEM+TOF+PSF used were three iterations and 21 subsets using a matrix size of 200×200 (voxel size: 4×4×2 mm3). A post-filtering was not applied to ease the interpretation of the results. Additionally, the CR (calculated in 3D) for each sphere size was also derived as a function of IR.
Two clinical cases were also considered to qualitatively illustrate the possible benefits derived from the proposed methodology. As a true whole-body comparison is nearly impossible with 18F-FDG given the time difference between the two imaging sessions, two patients were selected with an identical mass, size and time between injection and PET imaging. The first patient was a 73-year-old (56 kg, 1.56 m) woman evaluated for a breast cancer (injected activity: 173 MBq of 18F-FDG, time delay between injection and imaging: 63 min, system used: Biograph mCT, reconstruction parameters: 3D OP-OSEM+PSF+TOF with three iterations, twenty-one subsets and no post-filtering). The second patient was a 78-year-old (56 kg, 1.57 m) woman evaluated for a recurrence of follicular lymphoma (injected activity: 173 MBq of 18F-FDG, time delay between injection and imaging: 60 min, system used: Vision 450, reconstruction parameters: 3D OP-OSEM+PSF+TOF with four iterations, five subsets and no post-filtering). Vision 450 data were reconstructed by adapting the acquisition time to what was found by using the phantom experiment. Image roughness was calculated in a homogeneous region of the right lobe of the liver for each reconstruction.
The second clinical case referred to a patient treated with 90Y-microspheres (2251 MBq of Therapshere®) for a segment VII hepatocellular carcinoma. The radioembolization was for the management of a local recurrence including several satellite nodules. The acquisition time was 30 minutes for mCT (1 bed step) and Vision (2 beds step of 15 minutes). The reconstruction parameters were OP-OSEM+PSF+TOF, 3 iterations, 21 subsets and no post-filtering for mCT (200×200 matrix size) and OP-OSEM+PSF+TOF, 4 iterations, 5 subsets and no post-filtering for Vision as recently suggested in this specific case [13]. The time delay between injection and acquisition was 45.3 hours for Vision and 46.1 hours for mCT.