This study investigates image quality and micro-lesion detectability of dose-dependent PET images for total-body PET/CT scanner with a 194-cm-long axial FOV in paediatric patients. Our results from generated low-dose images suggested that an administered activity of 1/10-dose (0.37 MBq/kg) secures an optimal image quality superior to that of conventional digital PET; 1/15-dose (0.25 MBq/kg) showed a comparable image quality while maintaining acceptable PET parameters. Despite that the 1/30-dose (0.12 MBq/kg) PET images showed an arguably subjective diagnostic image quality, the variations in PET parameters were unacceptable. Despite that the image noise increased more prominently than the overall image quality degradation as the dose decreased, micro-lesion detectability was minimally compromised. Upon this, the feasibility of paediatric PET examinations at ultra-low-counts has been confirmed.
As for the subjective results, both SD and SUVmax in the liver increased gradually as the dose decreased, nevertheless, SUVmean in the liver was relatively stable in between simulated dose reduction groups. This observation is consistent with the feature of background measurement, where the statistic results at background obey the Poisson’s distribution, and the background feature is independent of acquisition time. Thus, the acquisition-time-dependent signal-to-noise-ratio of PET image is directly proportional to SD. As the SD increases, the extremum value turns more dispersive, leading to SUVmax increases with shortening acquisition time.
Compared to the recommended regimen (with a TAP of 7 MBq/kg·min/bed, 6–10 bed per total-body scan) in the current guideline of the European Association of Nuclear Medicine for 18FDG-PET/CT oncological examination , elongating the axial FOV to 194-cm-long suggesting a theoretical minimization of 18FDG regimen down to a TAP of 3.7 MBq/kg·min/bed (1 bed per total-body imaging). Upon the grade map of dose-and-time-dependent PET image quality (Supplementary Table 6), we suggest that an injected activity of 0.74 MBq/kg (estimated effective dose, 0.6–0.9 mSv) with an acquisition time of 5 min would be recommended for a routine protocol. The estimated effective doses of 18F-FDG using total-body PET to acquire an optimal image ranging 0.3–0.9 mSv (injected activity 0.37–0.74 MBq/kg) were much lower than the ~4 mSv originated from a PET/MR examination (injected activity 1.8 MBq/kg) [10, 20, 21, 29, 30], making CT the major safety concern.
High-sensitivity-induced TAP reduction provides a particular opportunity in either low-dose PET imaging or quick scan. Our results illustrate an extreme reduction in either injected dose or acquisition time, namely 0.25MBq/kg×10min/bed or 3.7MBq/kg×0.67min/bed, might be of use in special clinical scenario, i.e., 0.25 MBq/kg for patients needing repetitive PET examinations through their disease course to manage the overall exposure, or 0.67 min for patients who are claustrophobic or unable to keep still for minutes.
We concluded that the subjective image quality with an acquisition time of 40 s is comparable to that of clinical routine, which was different from the 60 s given by Zhang et al.  Such difference was mainly brought by the difference in subjects’ BMI. Our study included paediatric patients with a BMI of 16.2±3.4 kg/m2 (range, 12.2-22.2 kg/m2), rather than adult patients with a BMI of 22.9±3.3 kg/m2 (range, 18.4–28.9 kg/m2) . The quality of PET/CT images is inversely related to the subject’s fat mass, where the image quality of overweight patients often is degraded [31-33]. Besides, an acquisition time of 20 s showing an arguably diagnostic quality potentially reduces motion-induced artifacts, decreases the length of sedation, improves patient comfort, and may better assist radiation treatment planning . If the acquisition time could be further reduced to an extreme that allows breath-holding PET/CT scan, respiratory motion mismatch between the PET data and the CT data could be avoided.
In our study, the overall image quality and the lesion conspicuity have been evaluated subjectively and objectively, while the lesion detectability was characterized by diagnostic accuracy. The lesion detection rate is strongly related to the image texture, the size, shape, and surrounding environment of the lesions, and the reader’s experience. As distinct from the phantom study, we compared the lesion detection rate between the count-reduced reconstructed image with that of G600s. When dose-reduction down to 0.12 MBq/kg (1/30-dose), the micro-lesion detectability decreased 3% (1/33). Note that the undiagnosable micro-lesion had a moderate FDG uptake in a background of tumour infiltrated liver segment (Supplementary Fig. 2), which was beyond diagnostic necessity. These results agree with previous studies that showed decreased detectability with the short acquisition protocol in adults and an anthropomorphic thoracic phantom with irregularly shaped lesion simulating inserts [25, 35-37]. However, the image quality of extreme low-dose may not be fully-compensated by prolonged acquisition time. Because when acquisition time beyond a certain limit (i.e., 30 min) detected noise increases and may negatively impact the signal-to-noise ratio.
Limitations of this study include the small sample size, retrospective design, and restriction in extrapolation, i.e., limited to the total-body scanner. Limited by sample size, the age-dependent or BMI-dependent image quality analysis has not been included. The conclusion of our study was based on the 18F-FDG administering protocol. The further studies, for instance, delayed imaging, adminisetered low-dose, and various types of radiotracers, are still lacking. For lesion detectabiliy, only micro-lesions were taken into account. The effects of low-dose image on lesion shape, volume, contrast, remained to be explored. Particularly, the pediatric biodistribution of extreme-low dose of 18F-FDG remains unknown. The optimization of the reconstruction parameter was not considered in this study. TAP might be further reduced by increasing the number of iterations and applying the Bayesian penalized-likelihood reconstruction algorithm.