Experiment 1: Harmonization development
PET scans
The Iida brain phantom [8] was filled with F-18 FDG. F-18 FDG was mixed well with 800 ml water to create a uniform concentration solution. The mixture was poured into the phantom with an injected activity of 2.9 mCi. Once the phantom was full, it was capped and shaken to move the air bubbles to the top. Then, the cap was reopened to top off the phantom with additional F-18 FDG solution. This procedure was repeated several times to minimize air bubbles. The phantom was imaged on a single day, sequentially, on a series of PET scanners including the ECAT HR+ (HR+; Siemen’s/CTI) and the High-Resolution Research Tomograph (HRRT; Siemens/CTI). The phantom was gently moved between scanners to avoid disturbing any residual air bubbles. Static images of the phantom were acquired for 15 min each, first on the HRRT and then on the HR+. Because of decay, the radioactivity concentration in the phantom was 70.22 kBq/mL at the start of the HRRT scan and 37.83 kBq/mL at the start of the HR+ scan.
PET emission data were acquired in 3D and corrected for decay, attenuation, scatter, dead time, detector sensitivity, and randoms. The HRRT data were reconstructed with ordered subset expectation maximum (OSEM) using two iterations and thirty subsets at a voxel size of 1.2 mm × 1.2 mm × 1.2 mm and dimensions of 256 × 256 × 207 voxels. The HR+ data were reconstructed with OSEM using 4 iterations and 16 subsets at a voxel size of 2.06 mm × 2.06 mm × 2.4 mm and dimensions of 128 × 128 × 63 voxels. The reported resolution of the HRRT ranges from 2.3 to 3.4 mm FWHM [9]; the reported resolution of the HR+ ranges from 4.1 to 7.8 mm FWHM [10].
Determination of harmonization factor
The HRRT image voxel size was altered to match the voxel size of the HR+ images by down-sampling using trilinear interpolation (In-house IDL software). The resized image dimensions for the HRRT were 152 × 152 × 105 voxels and were designated as hHRRT(0 mm). Isotropic Gaussian blurring was applied to the hHRRT(0 mm) image with varying values of FWHM (from 0 to 10 mm in increments of 0.5 mm). Each blurred image was designated as hHRRT(xmm) where x indicates the FWHM of the Gaussian.
The hHRRT and HR+ images were loaded into MATLAB (2018b). The HR+ image was padded with zeros in the x-, y-, and z- dimensions resulting in dimensions of 152 × 152 × 105. The HR+ image was rigidly registered to the hHRRT(0 mm).
The intensity of each hHRRT image of the phantom was compared voxel-by-voxel to the corresponding intensity of the registered HR+ image by calculating a voxel-wise difference. The voxel-wise differences were squared and summed for all voxels across the image volume resulting in a sum of squared errors (SSE),
$${\text{SSE}} = \mathop \sum \limits_{i = 1}^{N} \left( {y_{{i,{\text{hHRRT}}}} - y_{{i,{\text{HR}} + }} } \right)^{2}$$
(1)
where yi,hHRRT is the intensity of the ith voxel in the hHRRT image and yi,HR+ is the intensity of the ith voxel in the HR+ image, and N is the total number of image voxels. The FWHM of the Gaussian kernel that minimized the SSE was deemed optimal.
Experiment 2: Validation in human subjects
Subjects
Three healthy subjects (2 men and 1 woman, mean age 33 ± 6 years, mean weight 73.7 ± 7.16 kg, mean BMI 24.8 ± 2.65) were recruited into the study. This study was approved by Yale University’s Human Investigations Committee, Radioactive Drug Research Committee, and Radiation Safety Committee. Informed consent was obtained prior to the performance of any study procedures. Subjects without significant medical issues, without current or recent smoking or nicotine use, and without current psychiatric disorders were recruited.
Study design
Each subject received a structural MRI scan on a Trio 3 T MR scanner (Trio and Prisma, Siemens Medical Systems, Erlangen, Germany). Subjects received a total of two 90 min [11C]raclopride scans, each acquired on a separate PET scanner. The MRI scan was acquired on a different day from the PET scans and was used for anatomical localization of PET scans. The pair of PET scans occurred on either the same day (1 subject) or on two separate days less than 7 weeks apart (2 subjects). One subject was scanned on the HRRT first, and the other two subjects were scanned on the HR+ first.
PET scans
[11C]raclopride was synthesized using methods previously described [11]. Before each PET scan, either a 6-min transmission scan (HHRT) or a 9-min transmission scan (HR +) was acquired for attenuation correction. [11C]raclopride was administered as a bolus by a computer-controlled pump (Harvard Apparatus, Holliston, MA, USA). The mean raclopride mass per body weight was 0.03 ± 0.004 μg/kg for the HRRT scans and 0.02 ± 0.009 μg/kg for the HR+ scans.
PET emission data were acquired in 3D and corrected for decay, attenuation, scatter, dead time, detector sensitivity, and randoms. For the HRRT images, motion correction was applied using a Polaris Vicra (Northern Digital Inc.) optical hardware-based motion tracking device [12]. For the HR+ images, motion correction was applied using a mutual-information algorithm [13] and the FLIRT software (http://www.fmrib.ox.ac.uk/fsl/flirt). Emission data were collected for 90 min. HRRT and HR+ data were both binned into 27 total time frames of 6 × 0.5 min, 3 × 1 min, 2 × 2 min, and 16 × 5 min. The HRRT and HR+ data were reconstructed with the same method and voxel sizing as Experiment 1.
Image analysis
Individual HRRT and HR+ PET images were co-registered to the subject’s own MR image and then nonlinearly registered to a MR template in a common space (MNI space; Bioimage Suite [14]). ROIs were extracted using automated anatomical labeling [15] (AAL) to generate regional time activity curves. The whole striatum ROI was extracted using the AAL atlas. Time activity curves were fitted with the simplified reference tissue model (SRTM) [16] using the cerebellum as a reference region to calculate non-displaceable binding potential (BPND) as the endpoint.
The harmonization method described in Methods in Sect. 1 was applied to the HRRT data. BPND estimates were made from the hHRRT(4.5 mm) using the same method as described above for the HR+ and HRRT scans.
Comparison of calculated endpoints
BPND values for the striatum were calculated and compared within subjects across the 3 sets of dynamic PET images: HRRT, HR+, and hHRRT(4.5 mm). The comparison metric was the percent difference (PD) between the HRRT/hHRRT(4.5 mm) and the HR+, which was calculated as follows:
$${\text{PD}} = 100 \times \frac{{\left| {{\text{BP}}_{{{\text{HRRT}}}} - {\text{BP}}_{{{\text{HR}} + }} } \right|}}{{\left( {\frac{{{\text{BP}}_{{{\text{HRRT}}}} + {\text{BP}}_{{{\text{HR}} + }} }}{2}} \right)}}$$
(2)
where BPHRRT is the striatal BPND for either HRRT or hHRRT(4.5 mm) and BPHR+ is the striatal BPND for HR+.
Experiment 3: Comparison of calculated endpoints with Test–Retest Data from HR+
Test–retest (T/RT) measurements of striatal BPND were calculated using data from a previous study by our group [7]. In that study, eight healthy subjects (four women and four men, mean age 31.3 ± 11.2 years, mean weight 68.7 ± 12.8 kg, mean BMI 23.9 ± 3.31) without significant medical issues were recruited. Current or recent smoking or other nicotine use or personal or family history of psychiatric disorders (including substance misuse) were exclusionary. [11C]raclopride scans were acquired on two separate days on the same subjects using only the HR+ scanner. The HR+ data were reconstructed using the same reconstruction method and voxel size as described above. PDT/RT was calculated between the two HR+ striatal BPND values as follows,
$${\text{PD}}_{T/RT} = 100 \times \frac{{\left| {{\text{BP}}_{{{\text{HR}} + ,1}} - {\text{BP}}_{{{\text{HR}} + ,2}} } \right|}}{{\left( {\frac{{{\text{BP}}_{{{\text{HR}} + ,1}} + {\text{BP}}_{{{\text{HR}} + ,2}} }}{2}} \right)}}$$
(3)
where \(BP_{{{\text{HR}} + ,1}}\) is the BPND from the first HR+ scan and \(BP_{{{\text{HR}} + ,2}}\) is the BPND from the second HR+ scan.