This study was approved by our institutional review boards, the Human Study Committee (approved on Sept 7th, 2021) and was conducted in accordance with the Declaration of Helsinki. In this study, two major evaluations were included: A, the utility of histogram analyses using 18F-FPYBF-2 comparing with visual evaluation and conventional ROI analyses using SUVR; B, comparison of the histograms analyses between two different PET tracers (18F-FPYBF-2 and 11C-PiB). For each purpose, we used the PET/CT data obtained in our previous studies: the first-in-human study using 18F-FPYBF-2 for subjects with cognitive disorders or healthy volunteers, and the comparative study between 18F-FPYBF-2 and 11C-PiB PET (13). These PET studies were approved by our institutional review boards, the Human Study Committee (approved on Sep. 25, 2013, and on Jan. 18, 2016). Each subject gave a written informed consent for the previous studies. Because of the retrospective study design of this study and the way of using anonymized subject data, requirement for additional informed consent was waved for the purpose of this additional analyses using histogram for the patients who had undergone the PET studies previously.
Subjects
The subjects were cognitively normal healthy volunteers (HV) and patients with suspected or diagnosed dementia, who were the subjects for our previous first-in-human studies using 18F-FPYBF2(13), Out of all the subjects (n = 116) of the previous studies, those who had undergone PET/CT examination were included in this study (n = 63). The cases examined by a dedicated PET machine were excluded (n = 53).
A: Evaluation of Histogram analyses using 18F-FPYBF-2 PET/CT
For evaluation of the utility of the histogram, the data of 18F-FPYBF-2 PET/CT performed in 25 HV and 38 patients with diagnosed or suspected dementia or related diseases (27 men and 36 women; mean age: 68.7 ± 12.4 (range: 40–88) years old) were used. Of the 38 patients, 15 had been clinically diagnosed as having AD, 17 mild cognitive impairment (MCI), and 6 other disease including dementia of Lewy body, amyloid angiopathy and vascular dementia, and front-temporal lobe dementia (Others).
B: Comparison of the histogram analyses between two different tracers
Thirteen subjects, 7 men and 6 women; mean age 70.5 ± 11.7 years old, included 2 HV, and 11 patients of MCI (n = 3), AD (n = 4), and Others (n = 4). All these subjects underwent PET/CT studies using two different tracers: 18F-FPYBF-2 and 11C-PiB. The intervals of the two PET/CT studies were within 2 weeks for patients with dementia or MCI and half a year for HV.
PET/CT examination
Both 18F-FPYBF-2 and 11C-PiB were synthesized in-house. The radiosynthesis of 18F-FPYBF-2, and 11C-PiB were performed using modification of the methods described elsewhere (11) (18) with a hybrid synthesizer and cassette-type multipurpose automatic synthesizer module (JFE Engineering Corporation, Japan).
PET/CT scans were performed by a whole-body PET/CT scanner, Siemens TruePoint Biograph 16 (Siemens/CTI, Erlangen, Germany). Static head PET image was acquired 50–70 min after the intravenous injection of 18F-FPYBF-2 (3.7-4.0MBq/kg), and 11C-PiB (7.4-9.0MBq/kg), respectively. Matrix size was 256 × 256, and acquisition data were reconstructed using the back projection reconstruction, and images were blurred to 6.0mm full width at half maximum in the transaxial direction using a Gaussian filter.
A CT scan was performed for attenuation correction (tube voltage, 130 kV; tube current, 30 mA; tube rotation time, 0.6 s per rotation; pitch, 1.0). The CT data were resized from a 336 × 336 matrix to a 256 × 256 matrix to match the PET data and construct the CT-based transmission maps for attenuation correction of the PET data with a post-reconstruction Gaussian filter (5 mm FWHM). The reconstructed voxel size of the PET/CT images was 2.3 × 2.3 × 4 mm.
The data analyses
PET images are evaluated by two conventional methods: visual evaluation and quantitative analysis using automated ROI, and two new methods: whole brain histogram parameter analyses, and the evaluation of the color map named Top20%-Map.
Visual evaluation of PET images
The axial coronal and sagittal PET images were evaluated using PRISM color and gray scale expressed with the SUVR display (range: 0–3) with the reference on the cerebellar cortex. For the gray scale images, we also evaluated by freely adjusting the gradation of the DICOM image on the image viewer.
The PET images were visually evaluated by three experienced nuclear medicine experts (C.O., T.H., and K.I.) who met at least one of the following conditions regarding the experiences for the diagnosis of amyloid PET: 1) with the experience of diagnosing more than 100 cases of amyloid PET, 2) with the board-certification for the interpretation of amyloid PET by Japanese society of Nuclear Medicine. Without any clinical information of the cases, each reviewer separately interpreted the PET images and classified them into three scores: 0, no abnormal amyloid deposition; 1, equivocal finding; 2, abnormal amyloid deposition. The scores of the three reviewers were summed up, and the final visual estimation were determined as below: negative, the summed score was 0 or 1; border, 2–4; positive, 5 or 6.
Quantitative analysis using automated ROI
For the quantitative evaluation of the amyloid deposition, SUV of each region (rSUVR) was measured using the cerebellar cortex as a reference (19, 20). To obtain rSUVR, we performed automated ROI analysis using the automated anatomical labeling atlas (AAL) (21) as described in our previous manuscript (13).
The difference from our previous studies was using CT images obtained PET/CT to convert it into a standard brain. CT images were spatially normalized to the in-house CT template and all AAL ROIs in the standard space were inversely transformed to individual spaces by the inverse deformation field.
The ROIs on the cerebellar cortex were combined and used as a reference region to create SUVR images (22). Mean rSUVR values within 90 anatomical ROIs in both hemispheres were calculated by an in-house Matlab script. Then, as a value for abnormal cortical amyloid deposition of each subject, the mean rSUVR value within the frontal, posterior cingulate, precuneus, parietal and lateral temporal cortical regions was calculated as composed SUVR (ComSUVR) (23).
Whole brain histogram parameter analysis and Top20%-Map
Figure 1 shows an overview of the process creating histogram and “Top20% -Map”. For the extraction of whole brain histogram and the creation of the map, a commercially available imaging workstations, CT volume analyzer (SYNAPSE VINCENT medical imaging system; FUJIFILM Medical Tokyo, Japan) was used.
At first, using the automatic brain extraction algorithm implemented in the volume analyzer, the skull and the soft tissue of the head were removed automatically on the CT data to extract “Brain CT.” In this process, whole brain including the brain stem is extracted automatically. Subsequently, “Brain PET” was extracted from the PET images using the “Brain CT” data as a mask image. After extracting “Brain PET” data, a spherical VOI including the whole brain was set to obtain the whole-brain histogram of the SUV with a bin size of 0.05. The process of automated brain-PET extraction and the creation of its histogram from one subject took about 5 minutes.
Then, after removing the data with no counts which corresponding to the data in the ventricle or outside the brain in the VOI, we exported the histogram data as a CSV file to a statistical software JMP®14.3 (SAS Institute Inc., Cary, NC, USA) to extract parameters including skewness, mean, and mode values from the histogram and the mode/mean ratio (MMR) was calculated according to the following formula:
$$MMR=\frac{mode value}{mean value}$$
.
Subsequently, high-accumulation area corresponding to 20% volume of the whole brain parenchyma were colored on the subject’s CT images using the fusion algorithm implemented in the volume analyzer. Creating the Top20%-Map includes the following three steps: 1st, Measure of the brain volume; 2nd, Scaling of the top 20% SUV; 3rd, Color mapping on the CT images.
1) First step: Measure of the brain volume
After overlaying “Brain PET” data on the CT images, a spherical large ROI that includes the entire brain was set. The minimum value of the PET display was gradually raised from zero until tracer accumulation was visually invisible in the lateral ventricle so as to calculate the volume of whole brain parenchyma.
2) Second step: Scaling of the top 20% SUV
After determining the volume of the whole brain parenchyma, the minimum value of the PET display was increased to determine the top20% SUV at which only 20% of the total volume with high uptake was displayed.
3) Third step: Color mapping on the CT images
On the fused images, input the top20% SUV to both the maximum and minimum values of the PET display. In this way, high-accumulation areas that account for 20% of the total volume are highlighted on the individual’s CT images. These steps creating Top20%-Map for one subject took about 3 minutes.
4) Evaluation of the Top20%-Map
Figure 2a shows the representative Top20%-Map of a HV without abnormal amyloid deposition; deep white matter, corpus callosum, posterior limbs in the internal capsule, brain stem and cerebellar white matter are visualized. On the axial image of the levels of the basal ganglia, the highlighted area seems like two kissing seahorses with pointed horns and long beaks (Fig. 2b). When kissing seahorses, brainstem and cerebellar peduncles were all observed, the maps were judged as normal map. On the other hand, abnormal map had the both following two findings: (1) highlighted cerebral cortex and/or posterior cingulate gyrus, (2) invisible the seahorse or no-colored cerebellar peduncle and brainstem (Fig. 2c). The cases with one of the following findings were determined to be a borderline map: (1) the conditions for normal map are met, but also the cortex and cingulate gyrus are partially highlighted; (2) invisible seahorse or brain stem without highlighted cortex and cingulate gyrus.
Statistical analyses
All numeric values were expressed as mean ± standard deviation (SD). All statistical analyses were performed using JMP®15.3.0 statistical software (SAS Institute Japan, Tokyo) with p values < 0.05 denoting statistical significance. A comparison between each group was analyzed by the Wilcoxon or the Kruskal-Wallis analyses for unpaired data. To compare the categorical data for the visual evaluation, we used Chi-square tests. For the comparison of the ComSUVR, skewness, and MMR between 18F-FPYBF2 and 11C-PiB, Spearman’s rank correlation tests and paired Wilcoxon signed-rank sum test were performed.