This study evaluated the feasibility of using a clinical TB-PET/CT scanner for preclinical small animal imaging. Image quality and comparability in quantification were assessed with the NEMA NU 4–2008 IQ phantom and in a sub-cohort of three anesthetized mice, followed by simultaneous scans of nine frozen mice with the Biograph Vision Quadra scanner that were compared with sequential scans with the Inveon preclinical PET scanner.

The phantom study confirmed the expected limitations of the clinical TB-PET/CT scanner in resolving submillimeter structures, with the 1 mm rod remaining undetectable due to the inferior spatial resolution of the Quadra (3.35 mm FWHM at ½ of the axial FOV and 1 cm radial offset [18]) compared to the Inveon (1.63 FWHM at the axial center FOV and 5 mm radial offset [23]). Nonetheless, rods ≥ 2 mm could be identified, and for the 4 mm and 5 mm rods, good agreement in RCs was observed between both systems. This agreement persisted across different positions within the field of view, including transaxial and axial offsets, where potential degradation due to parallax error or reduced sensitivity showed minor impact (e.g., RC 4 mm rod 0.69 (cFOV) and 0.62 (48 cm axial offset)). Of note, PSF modeling substantially improved quantitative accuracy partially compensating for the system’s lower spatial resolution. These findings highlight that while the Quadra is not suited for resolving very small structures, still quantification for larger structures is enabled. For additional information on position dependent spatial resolution and contrast recovery, we refer to our previous studies in which the Quadra was characterized at 18 positions within the FOV, extending the NEMA NU 2–2018 protocol using a Na-22 point source [21], and contrast recovery evaluations using small hot-sphere phantoms with warm background under various reconstruction settings [25].

As a proof-of-concept to demonstrate feasibility of the experimental setup and to evaluate the impact of reconstruction parameters, a sub-cohort of three anesthetized mice was scanned on the Quadra scanner. Higher SUVmean and SUVmax values were determined when PSF modeling was applied, which is in line with the phantom data. No impact of the number of iterations was determined on SUVmean and SUVmax values.

A potential position-dependent uptake for a realistic in vivo scenario was evaluated with a single frozen mouse scanned at multiple positions within the Quadra FOV (Fig. 2). Analogue to the phantom studies, the qualitative analysis (Fig. 6) revealed a higher resolution of smaller animal structures for the Inveon scan, as expected, due to the different spatial resolutions of the two scanners.

The quantitative analysis revealed comparable liver SUVmean and SUVmax at all investigated positions within the Quadra FOV (Fig. 8), indicating that varying the spatial resolution and sensitivity along the FOV had no impact on the SUV measurements [18, 21]. The SUVmean data from the Quadra scans were also consistent with the Inveon SUVmean data. However, the SUVmax was considerably higher for the Inveon scans (0.66) than for the Quadra (0.38 ± 0.02) scans. This difference can be attributed to the lower image noise (5.1%SD) of the Quadra (cFOV, PSF, 4 iterations), supported by the blurring associated with the lower spatial resolution, compared with 9.0%SD noise for the Inveon.

In order to achieve the best possible comparability of semiquantitative uptake values between different PET scanners or even between the same scanners within multicenter studies, harmonization is required. This process requires the definition of standardized scanner-specific image reconstruction and scan protocols, including phantom scans, scanner calibration and quality control, e.g., for clinical PET scanners, according to the European Association of Nuclear Medicine (EANM) FDG PET/CT guidelines for tumor imaging [26]. However, although various attempts and strategies have been proposed for preclinical PET scanners [27, 28], preclinical harmonization remains a challenge [29]. We investigated a simplified approach by matching the small voxel size of the Inveon scanner to the larger voxel size of the Quadra scanner by applying a Gaussian filter. This method helped to obtain a better visual comparison of the two scans (Fig. 6) and a better agreement for the liver SUVmax, with values of 0.49 (Inveon matched voxel size) and 0.38 ± 0.02 (Quadra).

Interestingly, we did not observe any position-dependent differences in the SUV parameters along the Quadra FOV. In previous work, we reported that due to the parallax error, the average spatial resolution (e.g., 4.1 mm FWHM (axial center) vs. 3.6 mm FWHM (40 cm axial offset)) [21] and contrast recovery (e.g., 56% (axial center) vs. 62% (40 cm axial offset)) [25] vary along the FOV. However, in this study, although the liver is a small structure, the SUV parameters were not affected, presumably as the respective VOIs covered only a fractional central part of the liver.

An entire animal cohort of nine frozen mice was scanned simultaneously with the Quadra instrument, and the SUVmean and SUVmax measurements of different target regions were compared with sequential scans performed on the Inveon scanner. Comparable SUVmean to those of the Inveon scanner were determined for all investigated organs (Fig. 9a-c). This result is particularly remarkable for the investigated muscle region, as the respective VOIs in the Quadra datasets contained only ~ 3.5 fractional voxels compared with 129.2 fractional voxels for the Inveon dataset. Notably, this low number of voxels determined the same quantitative uptake, although the Quadra voxels were resampled to the Inveon voxel sizes (VOIs were delineated on the Inveon datasets and transferred to the Quadra scans). Furthermore, the relatively large group strength of the 9 animals utilized in this study improved the comparability between the Quadra and Inveon scanners.

The SUVmax in the liver were comparable between the clinical and preclinical scanners (Fig. 9d). Even with the limited number of voxels, e.g., 15 voxels for the liver VOI, relatively homogeneous [18F]FDG uptake [30] is assumed to result in a comparable SUVmax between the two scanners. For the muscle SUVmax, significant differences were observed between the Inveon and Quadra scans. However, by matching the Inveon voxel size, a comparable SUVmax of 0.26 ± 0.06 (Inveon matched voxel size) to the values of all Quadra scans could be obtained (Fig. 9e).

For the larger whole brain region, comparable SUVmax were also determined between the Inveon scanner and the Quadra scanner (except for the Quadra grid scan, Fig. 9f). However, inhomogeneous uptake patterns in small brain regions, especially with PET tracers targeting central nervous system receptors [17], can lead to significant spill-in and spill-out effects due to high binding in specific brain regions, such as the striatum, and low binding in surrounding areas. This pattern, combined with a low number of voxels, can result in larger deviations in the SUVmean and SUVmax for smaller brain regions.

In addition, functional PET studies of mice or rats, where differences in small brain regions are determined or a voxel-wise analysis via parametric mapping is applied, will be impaired by the lower spatial resolution of the Quadra scanner. Nonetheless, investigations of anesthesia effects or treatment responses at the whole-organ level would be feasible for mice and rats.

The quantitative analysis of the different frame durations for the Quadra (Fig. 10) revealed that for frame durations down to 5 s, the SUVmean and SUVmax was still comparable to the value of the reference frame of 2119 s, although the number of trues was highly reduced (for all 9 animals: 3.72 × 106 (5 s) and 1.40 × 109 (2119 s)). Furthermore, although the mean percentage deviation for the 5 s frame duration compared with the reference scan was 17.3%, the SUVmean and SUVmax measurements in the liver and muscle regions were still comparable. However, the muscle region exhibited a larger standard deviation, which can be attributed to the smaller VOI size with fewer voxels included and hence a greater contribution of noise due to the limited count statistics. Notably, high quantification and image quality are still present even with short 5 s frames, suggesting a potential application for dynamic imaging. These parameters could be further improved by performing future scans at 60 min p.i. (instead of > 4.5 h, as in this study). The shortest frame duration for the Inveon was 139 s for the single frozen mouse experiment. When comparing SUVmean and SUVmax values with a 600 s frame duration (typical static frame duration in our institute), the SUVmean changed by only 1.8%, whereas the SUVmax revealed a change of 35.3% (data not shown). The SUVmax is more affected by technical variations and image noise then SUVmean, and is based on a single voxel value, potentially not reflecting the entire biology of the underlying structure [31]. The SUVmean is characterized with an overall greater robustness and comparability in our study.

This work can be extended to rat imaging, where a further improvement in quantification is expected due to the larger organ structures compared with those of mice.

Finally, due to the large dimensions of the FOV of the Quadra scanner, an unrestrained, freely moving rat could be imaged in its home environment (cage), minimizing stress levels and allowing the analysis of unaffected brain responses [32], provided that an advanced motion correction method is used [33].

One constraint of our study is that, while PSF modeling was applied during reconstruction of the Quadra data—partially compensating for PVE—no such correction was applied to the Inveon data. Although this may have affected comparability, no PVE correction option was available within the standard Inveon reconstruction workflow. Furthermore, a comparison with ground truth data, e.g., gamma counting, was not available for absolute quantification and comparison of the activity concentrations, as the animals were snap-frozen to avoid motion artifacts during scanning and to allow motion-free comparison and coregistration of the Inveon and Quadra datasets. Notably, the evaluation of the single frozen mouse and nine frozen mice in this study is based on an ex vivo comparison and hence is not impacted by potential motion artifacts (cardiac and respiratory motion) that could impact the comparability of both Quadra and Inveon scans. In contrast to this, our proof-of-concept study of three anesthetized animals (see Figs. 1 and 7) is impacted by cardiac and respiratory motion accordingly and demonstrated the feasibility of the experimental setup.

Dedicated preclinical imaging systems are state-of-the-art tools for performing high-resolution and high-sensitivity preclinical research. Although the preclinical PET scanner used in the present study does admittedly not belong to the latest generation of available preclinical PET scanners, it nevertheless is widely in use and provides robust and reliable results, which is of utmost interest to ensure longitudinal comparability and reproducibility.

A major advantage of using a clinical TB-PET scanners for preclinical studies is the ability to scan entire cohorts of animals simultaneously within a single acquisition, e.g., for drug screening. This process, of course, requires an advanced anesthesia setup to provide each animal with a constant, reproducible and individual anesthesia supply, as well as advanced temperature monitoring to ensure stable animal temperatures throughout the scan.

Scanning an entire animal cohort in a single PET scanner eliminates the intrascan variability that typically occurs when scanning animal cohorts with individual sequential PET scans (e.g. due to statistical fluctuations, animals’ circadian rhythm and tracer production specificity), thereby improving the reliability of animal comparisons within a cohort. Imaging sites with a clinical TB-PET scanner that do not have access to a dedicated preclinical imaging facility may have new opportunities for preclinical research by exploiting the high sensitivity of the clinical TB-PET scanner.