PET/CT systems
Discovery MI
The Discovery MI is a combination of an LBS, an SiPM-PET scanner and a 64-slice CT scanner. The LBS includes four blocks of detectors aligned in the axial direction, each comprising 19,584 crystals (3.95 × 5.3 × 25-mm) in a 4 × 9 matrix. The scanner has 36 detector units per ring and 9,792 SiPM channels. The DMI enables axial and transaxial FOV of 20 and 70 cm, respectively, with 71 image planes spaced at 2.79-mm intervals. The timing resolution is 375 ps (16). The spatial resolution according to NEMA NU 2-2012 is 3.91 mm at full width at half maximum (FWHM) (17).
Discovery PET/CT 710
The Discovery PET/CT 710 is a combination of LBS with PMT-PET and 64-slice CT scanners. The PET scanner has 13,824 LBS crystals in a 4.2 × 6.3 × 25-mm3 block. The D710 enables a 150.42-mm axial FOV and a 700-mm transaxial FOV with 47 image planes spaced at 3.27-mm intervals. The timing resolution is 500 ps. The spatial resolution according to a NEMA NU 2-2007 is 4.52 mm at FWHM (21).
PET reconstruction condition
Data acquired using SiPM-PET and PMT-PET were reconstructed under the following conditions: three dimensional-ordered subset-expectation maximization (3D-OS-EM) with TOF; 4 iterations; 16 subsets; Gaussian filter, 2.5 mm (FWHM); 128 × 128 matrix size; FOV, 25.6 cm; 2.0 mm/pixel. Images of the Hoffman 3D brain phantom acquired by both PET scanners were also reconstructed without TOF to evaluate changes in image contrast using TOF as the contrast gain. Contrast gain was evaluated as described below.
Phantom study
Data acquisition
Images were acquired for 30 min on different days using SiPM-PET and PMT-PET in list mode from a Hoffman 3D brain phantom (Data Spectrum Corporation, Hillsborough, NC, USA) that mimicked the [18F]FDG distribution in the human brain (22) and a pool phantom (Itoi Plastics Co. Ltd., Kobe, Japan), each containing 20 MBq of [18F]FDG. Phantom conditions and the scan duration were determined according to the Japanese Society of Nuclear Medicine (JSNM) phantom test procedure (23). We extracted time frames of 0–420 for the PMT-PET and 0–380 sec for SiPM-PET from 30 min of data derived from the two phantoms. The count statistics achieved from the PMT-PET time frame were equivalent to those for [18F]FDG clinical brain images at our institution as described below. The time frame of SiPM-PET was determined based on radioactive decay during the scan interval during the second acquisition.
Data processing
Image quality was evaluated using physical indices for phantom tests proposed by the JSNM: the ratio of gray-to-white matter contrast (contrast, %) calculated from images of Hoffman phantom, image noise (coefficient of variation, CV [%]) and uniformity (standard deviation, SD) calculated from images of pool phantom (23). The SD was also calculated from the pool phantom image with a scan duration of 30 min. The contrast, CV and SD were respectively calculated as described using images acquired from Hoffman and pool phantoms (23). Contrast gain (%) was calculated as:
where ContrastTOF and Contrastnon-TOF are contrast (%) with and without TOF, respectively.
These physical indices were calculated using PETquactIE ver. 3.0 (Nihon Medi-Physics Co., Ltd., Tokyo, Japan).
Clinical protocol
Data acquisition
The present study proceeded in accordance with the Declaration of Helsinki and was approved by the Ethics Committee at the TMIG (Approval No. 28077). All control individuals and patients provided written informed consent to participate in the present study after physicians provided a detailed explanation of the study at the Research Team for Neuroimaging. The individuals rested comfortably in a quiet, dimly-lit room for several minutes, then were placed in the supine position for intravenous [18F]FDG injection and uptake. Low-dose CT images for attenuation and scatter correction were acquired before starting PET image acquisition. The first set of PMT-PET images were acquired for 10 min starting from 40 min after [18F]FDG administration, then the second set of SiPM-PET images was acquired, also for 10 min. The second scan started within 5 min of completing the first scan of 22 controls and 10 patients using the two PET/CT scanners between April 2017 and July 2018. The controls were confirmed as not having degenerative neurological disorders on [18F]FDG and brain magnetic resonance images acquired using a Discovery MR750w 3.0T scanner (GE Healthcare). The MR images were acquired under the following conditions: three-dimensional mode (spoiled gradient recalled acquisition in the steady state: repetition time, 7.648 ms; echo time, 3.092 ms; matrix size, 196 × 256 × 256; voxel size, 1.2 × 1.0547 × 1.0547 mm3). Table 1 shows the characteristics of the controls. Among the controls, four were healthy volunteers and 18 had visual issues (visual snow, n = 12; blepharospasm, n = 2; visual disturbance, n = 1; photophobia, n = 1; Charles Bonnet syndrome, n = 1; traffic injury, n = 1). Ten patients had suspected degenerative neurological disorders. Table 2 shows the characteristics of patients.
Data processing
We separately normalized [18F]FDG images of 22 controls to a standard [18F]FDG PET template using MIMneuro (MIM Software Inc. Cleveland, OH, USA). Anatomical volumes of interest (VOI) of MIMneuro were automatically placed on the caudate nucleus, cerebellum, frontal, occipital, parietal and temporal lobes, putamen, thalamus and whole brain. Mean standardized uptake values (SUVmean) were measured using these VOI (24). Images of the 22 controls and 10 patients were statistically analyzed using CortexID Suite (GE Healthcare) (25-27). Anatomical VOI of CortexID Suite comprised the lateral and medial frontal, inferior and superior parietal, and lateral and medial temporal lobes, the anterior and posterior cingulate cortex, occipital lobe, sensorimotor, precuneus, primary visual cortex, and cerebellum. The SUV ratio (SUVR) was calculated using the value of the pons as a reference region. Z-scores for anatomical VOI-based analyses were calculated from anatomically normalized SUVR images using the formula,
where SUVRindividual and SUVRnormal are the mean SUVR of the individuals and the normal database of CortexID Suite in the VOI, respectively and SDnormal is the SD of the SUVR of the normal database of CortexID Suite in the VOI.
Data analyses
Data were statistically analyzed using Prism 8 Version 8.4.0 (GraphPad Software Inc., San Diego, CA, USA). The SUVmean of all regions for SiPM- and PMT- PET acquisitions were statistically compared using two-tailed paired Student t tests. Spearman rank correlation coefficients were calculated to evaluate relationships among different SUVmean in the whole brain and intervals between acquisitions. Z-scores were statistically analyzed for both acquisitions using Wilcoxon matched-pairs signed rank tests. Values with P < 0.05 were considered significant.