This single-institution study was approved by the Institutional Review Board of the Kofu Neurosurgical Hospital and Yamanashi PET imaging clinic in accordance with the Declaration of Helsinki. Because of the retrospective study design and the use of anonymized patient data, the requirement for informed consent was waived.
Ring-shaped dbPET scanner
The dbPET scanner (Elmammo, Shimadzu Corp., Kyoto, Japan) has received approval from the Japanese Pharmaceutical Affairs Law and is commercially available in Japan. It consists of 36 detector modules arranged in three contiguous rings, has a diameter of 195 mm and a transaxial length of 156.5 mm, and has depth-of-interaction measurement capability . The transaxial effective field-of-view (FOV) is 185 × 156.5 mm2. Each detector block consists of a four-layered 32 × 32 array of lutetium oxyorthosilicate crystals (1.44 mm × 1.44 mm × 18 mm in size) coupled to a 64-channel position-sensitive photomultiplier tube via a light guide. Attenuation correction was calculated using a uniform attenuation map with object boundaries obtained from emission data . Scatter correction was performed using the convolution-subtraction method with kernels obtained by background tail fitting . Performance metrics included 1.5-mm FWHM resolution in standard mode in the transverse, sagittal, and coronal views, detector sensitivity of 0.09–0.13 cps/Bq at the centre of the detector, and the sensitivity at 39.5 mm from the edge of the detector (depth of 1/4) is 0.05–0.08 cps/Bq. The peak noise equivalent count was 600–800 kcps. The characteristics and standard performance of this scanner have been reported in detail previously .
Whole-body PET/CT scanner
PET/CT scans were obtained using a Biograph Horizon TrueV FDG-PET/CT system (Siemens Medical Solutions, Knoxville, TN, USA). This system has 52 detector rings consisting of 160 blocks, with each block containing an array of 13 × 13 lutetium oxyorthosilicate crystals (4 mm × 4 mm × 20 mm) covering an axial FOV of 221 mm and a transaxial FOV of 690 mm. A CT scan was performed for attenuation correction (130 kV; 15–70 mA; tube rotation time, 0.6 s per rotation; pitch, 1; a transaxial FOV, 700 mm; and section thickness, 5 mm).
Development and preparation of the breast phantom
A cylindrical breast phantom containing four plastic spheres of different diameters was used. The inner and outer diameters of the cylinder were 100 mm and 140 mm, respectively, and the height was 170 mm. The diameters of the spheres, arranged in a planar circle inside the phantom, were 5, 7.5, 10, and 16 mm. Spheres smaller than 5 mm in diameter were not used because they could not be detected by PET/CT. Furthermore, in our previous studies with low TBR phantoms, the smallest 5-mm-diameter sphere could not be visually detected on dbPET images when the sphere-to-background activity concentrations was less than 8:1 . Therefore, the visibility of lesions smaller than 5 mm was not evaluated in this study. The cylinder and four spheres were filled with 18F-FDG solution at a sphere-to-background radioactivity concentration ratio of 8:1 in accordance with a previous study . The background radioactivity at the start of data acquisition by dbPET was set to 2.46 kBq/mL. One scan was performed under each position as detailed in the next section.
Data acquisition and image reconstruction
The breast phantom was positioned such that the spheres were precisely located in the same transverse plane at different positions in the transverse field of view. The spheres were positioned at 8 mm, 13 mm, 19.5 mm (1/8 of axial FOV), 39 mm (1/4 of axial FOV), and 78 mm (1/2 or halfway point of the axial FOV) below the top edge of the detector (Figure 1). Since it is unlikely that a breast lesion is located at the bottom edge of the detector, only the chest wall side of the detector was evaluated. Sphere placement at each position in the detector was confirmed visually and by measurement on the image. A three-dimensional list-mode dynamic row-action maximum-likelihood algorithm (LM-DRAMA) was applied to the image reconstruction of dbPET. In the ordered subset expectation maximization (OSEM) method, which is the commonly used method in PET/CT image reconstruction, the convergence speed of the iterative approximation improves when the number of subsets is increased. However, it also causes the limit cycle phenomenon wherein the measured data contains statistical noise. To overcome this limitation, the row-action maximum likelihood algorithm (RAMLA), a modified version of the OSEM method, was developed that uses the relaxation parameter λ in iterative calculations to suppress the effects of statistical noise due to later processed subsets . Subsequently DRAMA, a modified version of the RAMLA, was developed in which the relaxation parameter λ depends on the subset number, and the noise propagation from the subset to the reconstructed image is suppressed as the subset number increases, resulting in fast convergence with a reasonable signal-to-noise ratio . The dbPET images were reconstructed using a LM-DRAMA with one iteration and 128 subsets, a relaxation control parameter of β=20, and a matrix size in the transverse view of 236 × 200 × 236 with a post-reconstruction smoothing Gaussian filter (1.17-mm FWHM). For the clinical images, the extracted contour was the same as the subject's boundary and was therefore used for the attenuation coefficient map without adjustment (‘just mode’). For the phantom images, the estimated contour of the boundary was adjusted to account for the wall thickness of the phantom (‘tight mode’). The reconstructed voxel size of the dbPET images was 0.78 mm × 0.78 mm × 0.78 mm.
The clinical PET/CT images were reconstructed using the ordered subset expectation maximisation method and the time-of-flight (TOF) algorithm with four iterations and 10 subsets. The CT data were resized from a 512 × 512 matrix to a 180 × 180 matrix to match the PET data and construct CT-based transmission maps for attenuation correction of the PET data with a post-reconstruction smoothing Gaussian filter (5 mm FWHM). The reconstructed voxel size of the PET/CT images was 4.11 mm × 4.11 mm × 5 mm.
Analyses of phantom image quality
Visual and quantitative analysis of all PET images was performed using an imaging workstation equipped with syngo.via VB10 software (Siemens Healthcare GmbH, Erlangen, Germany). Standardized uptake values (SUVs), as a semiquantitative assessment of FDG accumulation, were extracted using this software. The SUV of a given tissue was calculated using the following formula:
SUV = body weight (g)
The maximum (SUVmax) and the mean (SUVmean) SUVs are the maximum and average value within the region of interest (ROI) (or volume of interest [VOI]), respectively.
An experienced nuclear medicine physician and two experienced PET technologists evaluated the hot spheres. Evaluations were performed using the slices displayed in the coronal image slice containing the sphere centres. The 5-mm-diameter hot sphere was visually graded as follows: 2, identifiable; 1, visualised, but similar hot spots observed elsewhere; and 0, not visualised. Spheres with visual scores ≥1.5 were deemed to be detectable. The final score for the visibility of the smallest sphere was the mean of the scores from three readers. The visual assessment was performed based on the Japanese guidelines . A circular ROI with a diameter of 5 mm was placed on the central slice of the 5-mm hot sphere. Additionally, 12 ROIs with a diameter of 5 mm were placed in the background region of the coronal image slice that contained the sphere centres, and 12 ROIs were placed in the +5 mm– and -5 mm–adjacent slices (36 ROIs in total). CNR and CRC were calculated to quantitatively compare the visibility between the different positions in the dbPET detector. CNR and CRC provide information about the visibility and how accurately the system reproduces the true activity concentration, respectively. CNR was calculated as follows:
where CH,5mm is the SUVmean in the 5-mm-hot sphere ROI, CB,5mm is the average SUVmean of the background ROIs, and SDB,5mm is the standard deviation of the background ROIs.
CRC was calculated as follows:
where aH and aB are the activity concentration in the hot sphere and the background, respectively.
We also placed 10 ROIs with a diameter of 16 mm in the background region of the coronal image slice that contained the sphere centres and its +5 mm– and -5 mm–adjacent slices (30 ROIs in total).
The CVB was calculated using the data from these 16 mm ROIs as follows:
where SDB,16mm is the standard deviation in the background ROIs and CB,16mm is the average SUVmean of the background ROIs.
These physical values were calculated according to a previous report [14, 21].
Analysis of human images
Of a total of 202 consecutive women who underwent both dbPET and whole-body PET/CT scans from August 2016 to September 2019, 62 histologically proven breast cancer tumours of 57 women with positive findings on both dbPET and whole-body PET/CT images were included in the study. Patients fasted at least 6 hours prior to administration of 18F-FDG (3 MBq/kg) and were scanned by whole-body PET/CT for 90 s per bed position and dbPET for 7 min per breast. Scans were performed at 60- and 90-min post-injection, both in the prone position. The PET/CT and dbPET images were reconstructed using the same conditions as for the phantom images.
All PET images were evaluated separately by two experienced nuclear medicine physicians (with 16 and 7 years of experience in interpreting PET, respectively). Of the 62 lesions, those in which the shortest distance from the detector edge on the chest wall side to the tumour centre was 2 cm or less on the transverse image of dbPET were defined as the “peripheral group”; the other lesions were defined as the “non-peripheral group”. Non-mass uptakes, other than focus and mass-like uptakes, were excluded because their quantitative reliability could not be established. Tumours that were exactly centred in both peripheral and non-peripheral regions and whose volume was equally present in both regions were also excluded.
The quantitative value of PET is known to be affected by the partial volume effect . To account for lesion size bias, lesion sizes were matched in the peripheral and non-peripheral groups. The non-peripheral group was reorganised such that lesion size matched the peripheral group in a one-to-one correspondence. As a result, 23 lesions in each group (total 46 lesions) were included in the final analysis.
To evaluate lesion visibility in dbPET depending on the position of the tumour, tumour-to-background ratio (TBR) was calculated as follows. All PET images were displayed in an inverse grey scale with a standardised uptake range of 0–6 for the purpose of reducing intra-reader variability. First, the smallest spheroid VOI that just contained the tumour was placed on the monitor. Second, 5-mm-diameter spherical VOIs were placed at 6 locations on the top, bottom, left, right, anterior, and posterior of the tumour, as close as possible to it, in the non-peripheral group. Five VOIs were used in the peripheral group; the posterior VOI was excluded because there was not enough space to place it posterior of the tumour (Figure 2). The TBR was the SUVmax of the VOI on the tumour divided by the average SUVmean of the five or six VOIs on the background. In PET/CT, the SUVmax and the SUVmean were equal as a 5-mm-diameter spherical VOI contained only one voxel. The TBRs were compared between dbPET and PET/CT images, and the TBR of dbPET was compared between the peripheral and non-peripheral groups.
A paired t-test was used to compare the TBR of dbPET and whole-body PET/CT for the peripheral and non-peripheral groups, respectively. The Mann–Whitney U test was used to test for differences in TBR on dbPET between peripheral and non-peripheral lesion groups. Statistical significance was defined as p<0.05. Additionally, for these PET measurements, interclass correlation coefficients (ICC) were used to evaluate the reliability between readers.