Minimally Invasive Method of Determining Blood Input Function from PET Images in Rodents

microPET R4

The microPET R4 was the first commercial PET scanner for rodents. Its design is based on the UCLA prototype (1) with improved sensitivity (20). The reconstructed-image resolution of the microPET R4 is 2.3 mm in full width at half maximum close to the center of the field of view. The volumetric resolution is ∼5 μL at the center of the field of view on the basis of the 10-ns timing window and is less than ∼14 μL within a 1-cm radius. A complete performance evaluation of the scanner was done by Knoess et al. (20).

Animal Studies

To investigate the accuracy of the FA method in determining the blood input function and myocardial time–activity curve, 2 groups of rodents were studied. The rodents had free access to standard rat chow and water. In group 1, 6 Sprague–Dawley rats (weight [mean ± SD], 305 ± 91 g) were studied with ¹⁸F-FDG (n = 3) and 1-¹¹C-acetate (n = 3). In group 2, 6 BALB/c mice (weight, 20.8 ± 1.4 g) were studied with ¹⁸F-FDG (n = 3) and 1-¹¹C-acetate (n = 3). The radiotracers (¹⁸F-FDG and 1-¹¹C-acetate) were produced in our laboratory with a commercially available module (CTI Molecular Imaging).

Animal Preparations and Small-Animal PET Imaging

The use of animals was approved by the Animal Studies Committee at Washington University School of Medicine. Approximately 30 min before undergoing PET, rodents were anesthetized with inhalant anesthesia (1.5% isoflurane). Catheters were placed in the jugular vein for radiotracer administration and in the carotid artery for blood withdrawal during the scan. Rodents were placed in a supine position in the small-animal PET scanner.

All dynamic small-animal PET images were obtained with a microPET R4 housed in a temperature-controlled suite. After an intravenous bolus injection (100−150 μL) of ¹⁸F-FDG (26.8 ± 4.5 MBq for rats and 24.1 ± 3.8 MBq for mice), list mode data were acquired for 60 min. The dynamic small-animal PET image data were sorted into 3-dimensional (3D) sinograms as follows: 24 frames of 5 s and 58 frames of 60 s. The frame durations were chosen to allow small intervals at early time points to show the rapid changes in the blood-pool input function and sampling at larger intervals at later time points to show the uptake characteristics of the myocardium. After an intravenous bolus injection of 1-¹¹C-acetate (32.7 ± 4.8 MBq for rats and 20.5 ± 5.5 MBq for mice) and because of the rapid kinetics of the tracer in the heart, list mode data were acquired for 20 min. Dynamic sorting of the data into 3D sinograms was performed as follows: 24 frames of 5 s, 12 frames of 15 s, 10 frames of 30 s, and 10 frames of 60 s. The frame durations were chosen on the basis of the dynamic imaging protocols used for 1-¹¹C-acetate in humans and large mammals studied at our institute.

The images were reconstructed by Fourier rebinning of the 3D sinograms followed by 2-dimensional filtered backprojection with a ramp filter (cutoff at Nyquist frequency). To obtain sufficient pixel density for analysis, the original images were enlarged with zoom factors of 3 and 4.225 for the rat and mouse studies, respectively, resulting in pixel sizes of 0.28 × 0.28 mm for the rat studies and 0.20 × 0.20 mm for the mouse studies. The slice thickness was 1.2 mm. At the time of the study, the only attenuation correction method available for this system was based on a ⁶⁸Ge point source and led to excessive noise in the transmission scans. Therefore, attenuation and scatter corrections were not applied. In general, attenuation correction factors are constant over time and should not change the shape of the time–activity curves. The time–activity curves were adjusted for their dynamic range on the basis of the maximum and minimum of the dynamic images and rescaled at the end of the FA procedure, as will be described in the next section. These steps effectively accounted for the scale introduced by attenuation and partially compensated for the bias introduced by scatter. Although the system is now equipped with appropriate attenuation and scatter corrections and the images it provides are more quantitative, we do not believe the absence of attenuation and scatter corrections significantly changed the results of the present study.

To compare blood-pool input functions obtained with the FA method to blood activity concentrations measured directly, arterial blood samples (mean ± SD volumes of blood were 6.1 ± 0.4 μL for rats and 5.9 ± 0.3 μL for mice) were withdrawn from the carotid artery catheter into Wiretrol II precision micropipette tubes (Drummond Scientific Co.). The dead volume of the catheter and sampling line was 50–100 μL, depending on the length of the catheter line and the size of the rodents used. During blood sampling, the catheter was not flushed, but a void sample was taken to ensure true blood sampling at each time point. For both types of animals and both radiotracers, blood samples were rapidly drawn for the scan duration (8–10). The micropipette tubes were preweighed and postweighed, and counting was done with a Beckman 3000 γ-counter for true blood activity concentration determinations.

FA Algorithm

FA assumes that the measured images are constructed from the superposition of a limited number of structures (or factors). For the present application, these structures correspond to the right ventricle, left ventricle, and myocardium. Each structure has a different time–activity curve, which is dependent on the kinetics of the tracer used. In the present study, we used the methodology of Di Paola et al. (15), which was extended by several groups (16–19). The FA method creates physiologic time–activity curves and their corresponding images in 4 steps. The flow chart in Figure 1 shows the processing steps used for FA.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

Flow chart of FA procedure used in this study. TACs = time–activity curves.

First, 3 or 4 transaxial slices encompassing the heart were manually selected, and the data were added together. To remove or minimize the noise, data outside the specified organ area were masked from the images. The sizes of the masked transaxial images were 32 × 32 pixels for the mice and 40 × 40 pixels for the rats. The numbers of pixels as a function of time (time-dependent pixels) thus were 1,024 (32 × 32 pixels) for the mice and 1,600 (40 × 40 pixels) for the rats. Each time-dependent pixel in the masked images was normalized to the individual maximum value for each time-dependent pixel.

Second, after image processing, a principal components analysis was performed. The covariance matrix, eigenvectors, and eigenvalues (loadings) of the normalized time-dependent pixels were calculated for the 3 structures.

The third step was the determination of new factor axes that corresponded to the best representation without negative components for the 3 structures describing the dynamic data. These axes were calculated from the set of loading values for which the triangle subtended by the combination of any 3 loading values within the loading value space led to the maximum area. Thus, triangles were formed by any 3 loading values chosen out of 1,024 loading values for the mouse studies or 1,600 loading values for the rat studies. The normalized time–activity curves and the corresponding factor images were generated with these new factor axes. The final time–activity curves then were normalized by linear interpolation of the maximum and minimum values in the original images.

Finally, a constant given by the difference between the last data point for the FA blood input function and the last blood sample was added to left ventricle and right ventricle time–activity curves.

Statistical Analysis

Correlations and percentage errors for input functions extracted from blood samples and the FA method were calculated with linear least squares regression analysis.