Quantification of Cerebral Blood Flow and Oxygen Metabolism with 3-Dimensional PET and ¹⁵O: Validation by Comparison with 2-Dimensional PET

Subjects

Eight healthy men, 21- to 24-y-old, were recruited and provided written informed consent. Two sequential PET studies, with a 2D scanner and a 3D-only scanner, were performed on the same day with each subject at rest. The order of the studies (2D, 3D) was randomized. All volunteers were determined to be healthy on the basis of medical history, blood screening tests, anatomic MRI (T1- and T2-weighted images), and MR angiography of the brain. MRI was performed with a 1.5-T whole-body scanner (Magnetom Vision; Siemens Medical Solutions, Inc.). Three-dimensional T1-weighted images (magnetization-prepared rapid acquisition gradient echo [MPRAGE] sequence; repetition time, 9.7 ms; echo time, 4.0 ms; inversion time, 300 ms; flip angle. 12°; isotropic voxel size, 1 mm) were also used as anatomic references in subsequent region-of-interest (ROI) analysis. The study protocol was approved by the Ethics Committee of the Akita Research Institute of Brain and Blood Vessels.

2D PET Scanner

2D PET was performed with a SET-2300W scanner (Shimadzu Corp.), which uses bismuth germanate (BGO) detectors and provides 47 sections with a center-to-center distance of 3.125 mm (17). The axial field of view (FOV) was 150 mm. The intrinsic spatial resolution was 4.5-mm full width at half maximum (FWHM) in-plane and 4.5-mm FWHM axially. Although the septa can be retracted for 3D acquisition, all data were acquired in 2D mode with the septa in place. The coincidence time window was set to 18 ns.

Image reconstruction with a filtered-backprojection (FBP) algorithm and a Butterworth filter resulted in a final in-plane resolution of approximately 8-mm FWHM. Slices were rebinned by adding sinograms in groups of 2 with a 1-slice overlap between adjacent groups, resulting in 46 reconstructed slices. Before the emission scans, a transmission scan (10 min) with a ⁶⁸Ge/⁶⁸Ga external source was performed for attenuation correction. The PET scanner was calibrated by scanning a cylindric phantom (14 cm in diameter) filled with a ⁶⁸Ga aqueous solution with a known radioactivity concentration. In this procedure, an ROI of 10 cm in diameter was placed in the center of the phantom image. Detector normalization, attenuation correction, and image reconstruction were applied identically to the calibration phantom and to the subjects. No scatter correction was applied.

3D PET Scanner

For 3D PET, a SET-3000GCT/M scanner (PET/CT; Shimadzu Corp.) dedicated to the 3D mode was used (15). It uses gadolinium oxyorthosilicate (GSO) detectors and provides 59 sections with a center-to-center distance of 2.6 mm. The axial FOV was 156 mm. The intrinsic spatial resolution was 3.5-mm FWHM in-plane and 4.2-mm FWHM axially. In this study, the scanner was operated in a static scan mode with dual-energy window acquisition. The coincidence time window was set to 6 ns. To reduce the counting rate of random coincidence and scatter coincidence attributable to radioactivity outside the FOV, we used a prototype shield module consisting of 7-mm-thick lead plates attached to the gantry bed and covering the breast and shoulder of the subject.

FBP image reconstruction followed by 3D gaussian smoothing with 6-mm FWHM resulted in a final in-plane resolution of approximately 7-mm FWHM. Before the emission scans, a transmission scan (3 min) with a ¹³⁷Cs point source was performed with a BGO transmission detector ring coaxially attached to the GSO emission detector ring (15). The PET scanner was calibrated by scanning a cylindric phantom (15 cm in diameter) filled with a ⁶⁸Ga aqueous solution with a known radioactivity concentration. In this calibration procedure, an ROI of 10-cm diameter was placed in the center of the phantom image. Detector normalization, attenuation correction, scatter correction, and image reconstruction were applied identically to the calibration phantom and to the subjects.

A hybrid scatter-correction method, based on acquisition in the dual-energy window combined with a convolution-subtraction method in the upper energy window (UEW), was used (16,18). In this article, this method is referred to as the hybrid dual-energy window (HDE) method. Coincidence events were assigned to the standard energy window (SEW) when both photons fell within the standard photopeak window (300–700 keV). Additionally, when both photons deposited energy in the range of 480–700 keV, the events were also recorded as UEW data. Raw data for each window were acquired simultaneously and then converted off-line to 2D sinograms by a Fourier rebinning algorithm (19). Therefore, subsequent data processing could be done in 2D calculation without the computational burden of 3D data manipulation.

Estimation of the true (scatter-free) component of the SEW data was performed on a sinogram basis as follows:

(i) The convolution-subtraction method was applied to the UEW data for scatter correction (20). The scatter component of the UEW data, UEW_scatter, which is thought to be quite small in comparison with that of the SEW data, was estimated by convolving a 2D scatter kernel with the 2D projection (coronal view) of the UEW sinograms. The scatter kernel was defined as a 2D low-pass filter in the spatial frequency domain, 1/(1 + α exp(β |f|²)), where f represents spatial frequency and α and β are optimization parameters. This function was empirically selected to represent the scatter tail of objects. We optimized the parameters (α, β) by matching the calculated scatter distribution with the measured scatter tail of cylindric phantom (15 cm in diameter). The true (scatter-free) component of the UEW data, UEW_true, was calculated as:

(ii) The scatter component of the SEW data, SEW_scatter, was calculated as a subtraction of UEW_true scaled by factor f from the SEW data as:The scaling factor f between the 2 energy windows was determined from the phantom data according to the manner of Ferreira et al. (18). For the current configuration of energy windows (480–700 keV for UEW, 300–700 keV for SEW) the scaling factor was approximately 2.5. The energy width of the UEW (480–700 keV) was optimized as a compromise between the number of detected events in the UEW and the accuracy of the scatter correction (convolution-subtraction method) for the UEW data.

(iii) In the final step, the true component of SEW, SEW_true, was estimated as a subtraction of smoothed SEW_scatter, (SEW_scatter)_smoothed, from the SEW data as:A boxcar moving average with a 50-mm square was used for smoothing. As a result, a scatter-corrected sinogram (corresponding to SEW_true) and an uncorrected sinogram (corresponding to SEW) were created and were cross-calibrated with cylindric phantom data processed with and without scatter correction, respectively.

PET Procedure

Each PET study included a transmission scan for attenuation correction and 3 static emission scans with the inhalation of C¹⁵O, the inhalation of ¹⁵O₂, and the injection of (21). The interval between scans was approximately 15 min. A head fixation system was applied for each subject to minimize head movement during scanning. The subjects inhaled the gases (C¹⁵O and ¹⁵O₂) through a disposable plastic face mask under the condition of natural respiration. The gases were supplied at a flow rate of 500 mL/min and at a constant concentration of radioactivity. Blood gases were measured at the beginning and at the end of the scanning.

C¹⁵O PET studies were performed to measure CBV (22). Static PET was initiated 3 min after 1 min of continuous inhalation of C¹⁵O gas. Scanning lasted 4 min. The total doses of radioactivity supplied by mouth were 5.5 ± 0.7 GBq (average ± SD; n = 8) for 2D PET and 1.3 ± 0.2 GBq for 3D PET. Effective dose estimates to each subject for 2D and 3D PET were 2.8 and 0.7 mSv, respectively. Three arterial blood samples were taken during scanning to measure whole-blood radioactivity. The cerebral–to–large vessel hematocrit ratio was assumed to be 0.85 (23).

PET studies were performed to measure CBF. The protocol consisted of 3 min of static scanning initiated simultaneously with 2 min of intravenous infusion of . Injected doses at the start of scanning were 1.5 GBq for 2D PET and 0.4 GBq for 3D PET. Effective dose estimates to each subject for 2D and 3D PET were 1.4 and 0.3 mSv, respectively. An arterial input function was determined with a β-detector system that continuously measured radioactivity in arterial whole blood taken from the radial artery. Delay and dispersion occurring in the β-detector system and the internal arterial line were corrected as described previously (24,25). CBF was calculated by the autoradiographic method (7,26,27).

¹⁵O₂ PET studies were performed to measure OEF and CMRO₂ (6). The protocol consisted of 3 min of static scanning initiated simultaneously with 1.5 min of inhalation of ¹⁵O₂. The total doses of radioactivity supplied by mouth were 12.2 ± 0.5 GBq for 2D PET and 3.0 ± 0.3 GBq for 3D PET. Effective dose estimates to each subject for 2D and 3D PET were 4.7 and 1.2 mSv, respectively. The arterial input function was determined in the same way as that for the PET scan. The contribution of ¹⁵O-labeled metabolic water was estimated according to the method of Iida et al. (28).

Data Analysis

For each subject, 3 sets of parametric maps (CBF, CBV, CMRO₂, and OEF) were produced for 2D PET, uncorrected 3D PET, and scatter-corrected 3D PET. The 3D PET maps were registered with 2D PET maps with SPM99 (Wellcome Department of Cognitive Neurology) (29,30). All PET maps (2D and 3D) were then transformed to standard brain size and shape with linear and nonlinear parameters with SPM99 and a built-in CBF template. Thus, the resultant parametric maps of all subjects had the same anatomic format with an isotropic voxel size of 2 mm. The ROIs defined in the study by Ito et al. (31) were subsequently applied for all parametric maps. The individual T1-weighted image was examined to confirm the proper arrangement of the ROIs.

Circular ROIs 16 mm in diameter were defined for the pons, midbrain, thalamus, putamen, parahippocampal gyrus, and anterior and posterior parts of the cingulate gyrus, and elliptic ROIs of 16 × 32 mm were defined for the cerebellar cortex, centrum semiovale, and 4 neocortical regions representing the frontal, temporal, occipital, and parietal lobes. ROIs were drawn bilaterally for each brain region (if possible) in 3 adjacent slices, and data were averaged. Mean values and SD for CBF, CBV, CMRO₂, and OEF were calculated for 8 subjects. Differences in each variable between 2D PET, uncorrected 3D PET, and scatter-corrected 3D PET were examined by a paired t test with Bonferroni correction for multiple comparisons. Significance was accepted at P < 0.05. Correlation analysis between 2D PET and 3D PET was performed to examine the deviation from the line of identity. Regression lines were determined by principal component analysis. Bland–Altman analysis was also performed to confirm the agreement between 2D PET and scatter-corrected 3D PET (32).

Blood Gas Analysis

The Paco₂ (partial pressure of carbon dioxide, arterial), Pao₂ (partial pressure of oxygen, arterial), pH, blood pressure, and heart rate during each scanning arre summarized in Table 1. The Paco₂ (P = 0.03) and pH (P = 0.04) during the scanning differed statistically between 2D PET and the 3D PET, but the differences were marginal (1.3 mm Hg and <0.01, respectively). In other items, no significant difference was observed. The arterial hemoglobin concentration and hematocrit during these studies (mean ± SD.) were 14.7 ± 1.0 g/dL and 45.0% ± 3.0%, respectively.

PET

Representative images of 3D PET for ¹⁵O₂, C¹⁵O, and are shown in Figure 1. The random counting rate was much higher in ¹⁵O₂ PET compared with C¹⁵O and PET primarily because of the high concentration of radioactivity in the gas mask (outside the FOV) and in the upper airway. During ¹⁵O₂ PET, the coincidence counting rate reached up to 300 kcps (1,300-kcps prompt and 1,000-kcps random counting rate) with 3D PET and 100 kcps (550-kcps prompt and 450-kcps random counting rate) with 2D PET. During C¹⁵O PET, maximum counting rates, which occurred at the start of scanning, were 50 kcps (70-kcps prompt and 20-kcps random counting rate) with 3D PET and 20 kcps (30-kcps prompt and 10-kcps random counting rate) with 2D PET. During PET, the coincidence counting rate reached up to 300 kcps (450-kcps prompt and 150-kcps random counting rate) with 3D PET and 100 kcps (200-kcps prompt and 100-kcps random counting rate) with 2D PET. Despite the lower radioactivity dose in 3D PET, the maximum counting rate was approximately 3 times that of 2D PET, reflecting the higher sensitivity and better counting rate performance of the 3D PET scanner.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

Representative images (subject 1) of 3D PET for ¹⁵O₂ (A), C¹⁵O (B), and (C) in axial view (top), coronal view (middle), and sagittal view (bottom). Images were scatter-corrected by the HDE method. ¹⁵O₂ image shows high concentration of radioactivity in upper airway because of acquisition during inhalation of ¹⁵O₂ gases.

CBF

CBF, CBV, CMRO₂, and OEF values for 13 brain regions are given in Table 2. Gray–to–white matter ratios of these variables are also included. Anatomically standardized maps for a subject (subject 4) are shown in Figure 2.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2. 

Anatomically standardized CBF, CBV, CMRO₂, and OEF images of a single subject (subject 4) obtained with 2D PET (left), uncorrected 3D PET (middle), and scatter-corrected 3D PET by the HDE method (right). Slice positions are −36, −18, 0, 6, and 36 mm from anterior commissure–posterior commissure line.

In 8 of 13 brain regions, statistically significant differences in CBF between 2D PET and uncorrected 3D PET were observed (Table 2). Compared with 2D PET, uncorrected 3D PET gave larger CBF values in the centrum semiovale and smaller values in the gray matter brain regions, resulting in lower gray–to–white matter CBF ratios. No statistically significant differences in CBF were observed between 2D PET and corrected 3D PET. Corrected 3D PET resulted in gray–to–white matter ratios close to those of 2D PET and showed similar image contrast (Fig. 2).

CBV

In all brain regions, the CBV values were significantly larger with uncorrected 3D PET than with 2D PET (Table 2). In contrast, no statistically significant difference was observed between 2D PET and corrected 3D PET for any brain region. Corrected 3D PET resulted in gray–to–white matter ratios close to those of 2D PET and showed similar image contrast (Fig. 2).

CMRO₂

In 10 of 13 brain regions, statistically significant differences in CMRO₂ between 2D PET and uncorrected 3D PET were observed (Table 2). Although the trend was similar to that of CBF, CMRO₂ values in the pons, parahippocampal gyrus, and midbrain were larger with uncorrected 3D PET than with 2D PET—opposite to the CBF results. In almost all regions (12/13), CMRO₂ values were not statistically different between 2D PET and corrected 3D PET. Corrected 3D PET resulted in gray–to–white matter ratios close to those of 2D PET and showed similar image contrast (Fig. 2).

OEF

In contrast to CBF, CBV, and CMRO₂, differences in OEF were not large between PET studies. In 5 of 13 brain regions, statistically significant differences in OEF between 2D PET and uncorrected 3D PET were observed (Table 2). A difference between 2D PET and corrected 3D PET was observed only in the midbrain.

Graphic Analysis

Figures 3 and 4 show the scatter plots correlating 2D PET with uncorrected 3D PET and scatter-corrected 3D PET, respectively. Coefficients of correlation and linear regression lines are also shown. For CBF, CMRO₂, and OEF, good correlations were obtained, irrespective of 3D PET scatter correction, although the coefficients were slightly larger when scatter correction was applied. For CBV, scatter correction improved the correlation substantially (0.519 without correction, 0.710 with correction). For all variables, the regression line was closer to the line of identity for corrected 3D PET than that for uncorrected 3D PET.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3. 

Correlation between 2D PET and uncorrected 3D PET for CBF (A), CBV (B), CMRO₂ (C), and OEF (D). Number of data points is 104 (8 subjects × 13 ROIs) for each dataset. Regression line, determined by principal component analysis (solid line), and line of identity (dashed line) are also shown.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4. 

Correlation between 2D PET and scatter-corrected 3D PET for CBF (A), CBV (B), CMRO₂ (C), and OEF (D). Number of data points is 104 (8 subjects × 13 ROIs) for each dataset. Regression line, determined by principal component analysis (solid line), and line of identity (dashed line) are also shown.

The Bland–Altman plot of the difference in each variable between 2D PET and scatter-corrected 3D PET against their mean value is shown in Figure 5. Although there was a trend toward larger fluctuation for higher CBF and CMRO₂, no systematic bias was observed over the range of measurement for all variables. The mean differences (2D PET − scatter-corrected 3D PET) in CBF, CBV, CMRO₂, and OEF were −2 ± 9 (mL/100 mL/min), 0.0 ± 0.7 (mL/100 mL), −0.2 ± 0.6 (mL/100 mL/min), and 0.00 ± 0.04, respectively.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 5. 

Bland–Altman plot (difference against mean) with 2D PET and scatter-corrected 3D PET for CBF (A), CBV (B), CMRO₂ (C), and OEF (D). Mean difference (solid line) and ±2 SD (dashed line) are shown.