Simulation of Tracer Dose Reduction in ¹⁸F-FDG PET/MRI: Effects on Oncologic Reading, Image Quality, and Artifacts

DISCUSSION

In PET, the annihilation of positrons and electrons leads to the emission of 511-keV γ-photons. This high-energy radiation might potentially harm patients and personnel in the long term (9,23,24). The increasing use of ¹⁸F-FDG PET in oncologic imaging not only in initial staging but also in follow-up examinations also results in an increase of radiation exposure for both the patients and the medical personnel. Thus, a reduction of PET tracer doses is desirable. The definition of today’s doses in ¹⁸F-FDG PET is mostly based on theoretic models, phantom studies, and the retrospective comparison of different groups of patients (10–15). Because PET is a quantitative statistical imaging technique based on the Poisson-distributed random process of positron emission, theoretic calculations are legitimate. Those models use theoretic benchmarks such as image noise (which is calculated as the SD of the counts in a region of interest) and NECR as surrogates for the assessment of image quality. They are aimed at optimizing objectified parameters such as the NECR peak and do not consider individual characteristics of the patient, physiologic tracer distributions, or the subjective requirements of the physician for an adequate reading of the examination. It is therefore not clear how representative, for example, a phantom study is for an individual patient (25). Moreover, Chang et al. (14) reported in a retrospective study with a large number of patients that a scanner with a higher NECR does not necessarily translate to lower image noise and therefore better image quality, depending on the technique used for image reconstruction. Therefore, the conjunction between NECR and the real image impression or subjective image quality might be more complex. Moreover, to evaluate image quality, many studies measure the image noise in homogeneous liver tissue (12); however, this might not be sufficient to represent the PET image quality of a whole-body examination, because the PET image quality of individual regions might be influenced by regional patient characteristics. This uncertainty is also represented in the heterogeneous recommendations for doses in ¹⁸F-FDG PET, for example, comparing the respective European and American societies (13,26). By applying the proposed method of retrospectively modifying PET list-mode data, we could perform a retrospective intraindividual comparison of PET images with different doses.

The first part of our study revealed the effects of dose reduction on objective imaging parameters. We observed variable influences of doses on the system NECR and estimated SF in different body regions, pronounced in the abdominal/pelvic region. Therefore, the influence of doses on image quality differs in different body regions. With decreasing doses, an overall nonlinear increase of the CV in organs can be seen. Additionally, an enhanced effect on the average SUV_mean of the small bowel, the psoas muscle, and the lumbar vertebra (mid abdominal region) with a pronounced decrease compared with the other investigated regions (upper abdomen or urinary bladder) was observed with decreasing doses. However, it has to be noted that those regions provide low SUV and thus small changes cause increased relative deviations. The finding of increasing estimated SF and decreasing SUV_mean with decreasing doses is likely to be connected, because an artificially growing SF results in an underestimation of measured tracer uptake.

In 2 patients, we observed large photopenic areas in the middle abdomen with decreasing dose; one of them is demonstrated in Figure 6. With BMIs of 24 and 26, those patients were not obese—a common potential reason for impaired PET image quality. In both cases, the estimated SF reached the system limit of 0.75 in those bed positions at doses of 1.75 MBq/kgBW and below, respectively. The system NECR in these cases showed a break in the decreasing curve at the corresponding doses (Fig. 7). Therefore, the high amount of scatter resulting from an erroneous scatter estimation (or more precisely from erroneous scatter scaling) hampers the image reconstruction, or the scatter correction in particular. A reconstruction of PET images without scatter correction or scatter scaling did not reveal these photopenic areas (supplemental materials). The NECR and SF reported were not measured but estimated (calculated) by the scanner. Thus, the computation of the SF and consequently the SUV is incorrect in those regions at low tracer doses. However, because no PET-positive lesions were found in those regions for either patient, the clinical impact on oncologic examinations remains unclear.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 7.

NECR and estimated SF of bed position covering abdominal/pelvic region of 2 patients with large photopenic artifacts in abdominal region. Note break in curves of NECR when estimated SF reaches internal maximum of 0.75, limited by scanner software.

Scatter estimation is usually performed by single scatter simulation (27,28) whereas the correction for multiple scatter is done via scatter scaling (scatter fitting outside attenuating regions in the emission sinogram (29)). Thus, scatter scaling can be problematic for low tracer concentrations or specific tracers (low counting rate outside attenuating regions (30)) as well as for high tracer concentrations outside the PET field of view (31). This is especially true for current PET/MRI systems, which have higher estimated scatter fractions than PET/CT systems (32). Thus, to allow PET imaging with low tracer activities, an optimization of the scatter correction is needed (e.g., via fast Monte Carlo simulations (33,34)). However, because the attenuation correction map in an essential input for the scatter estimation, an optimization of the MR-based attenuation correction might also be an important point of action to improve the reconstruction results (e.g., (35,36)). A workaround to reduce photopenic areas in the abdominal region might be an optimized positioning of PET fields of view (PET field of view not between kidney and bladder). Also, prolonged acquisition times in this region might compensate for low-dose examinations. However, this has to be further evaluated in future studies. Of course, ensuring the voiding of the patients’ bladders is important.

Other investigated regions showed a more stable behavior, with a pronounced decrease of SUV_mean from below 1.25 MBq/kgBW. A typical photopenic artifact in PET images was located around the urinary bladder. This artifact was severely increasing with decreasing doses in 3 patients (patients different from the previously described cases). As a consequence, in 1 of those cases, a PET-positive carcinoma of the rectum could not be distinguished from the background at a dose of 0.5 MBq/kgBW (supplemental materials).

The oncologic evaluation according to PERCIST revealed that the calculated threshold in the liver shows a relatively stable behavior compared with the other examined parameters even with a slight increase of the threshold at low doses. This behavior can be explained by the increase of SD in lower-dose images (threshold = 1.5 × SUV_mean + 2 × SD), which therefore countervails the trend of decreasing SUV_mean (Figs. 2 and 4). In some target lesions, the SUV_peak in the highest dose was just close above the liver threshold. In 3 of those lesions, the SUV_peak was below the threshold in lower doses (maximum, 1.75 MBq/kgBW). This might have severe influence on oncologic readings, because those lesions would not have been rated as target lesions. Moreover, considering the Deauville criteria for lymphoma (7) and the overall trend toward quantitative assessment in PET, a reliable measurement of SUV is of utmost importance, especially in clinical trials.

The SUV_max and the MTV were revealed to be unstable parameters with a strong correlation to doses. The MTV can influence the oncologic reading in follow-up examinations, in texture analysis, as well as on therapy planning in radiation therapy (8,37,38).

With a simulated dose of 2 MBq/kgBW, the subjective image impression as well as the diagnostic performance of oncologic reading did not differ substantially from higher doses in most cases: the evaluation of the subjective image impression revealed that over a wide range of doses, no substantial differences compared with 2 MBq/kgBW were found. The reading of the image impression represents an individual subjective evaluation by an experienced nuclear medicine physician at a certain time; however, 5 PET images with a simulated dose of 1.75 MBq/kgBW were rated better than the 2 MBq/kgBW PET images. With a dose of 2 MBq/kgBW, most lesions could be identified correctly and no false-negative or-false positive findings in an organ or patient-based analysis were observed. Moreover, Likert scores regarding diagnostic confidence of lesions did not differ significantly from those of the maximum doses.

Previously conducted studies investigated the effects of simulated dose reduction in dependence on predefined true count levels on image characteristics of ¹⁸F-FDG PET images of the lung (21,39). We chose a more clinical approach by calculating whole-body PET images in dependence on patients’ characteristics, because count levels differ for different body regions and cannot be predicted. De Groot et al. (12) suggest that the signal-to-noise-ratio in PET is more strongly correlated to the body mass than to the lean body mass, the BMI, or the fat mass; thus, we decided to calculate our doses in relation to kilogram of body weight. Yan et al. (39) determined a lower limit of 5 million true counts for the Biograph mMR, which would roughly correspond to a dose below 0.5 MBq/kgBW in our setting. However, the dose-determining factor is the lower abdomen, which was not assessed by their work. Schaefferkoetter et al. (21) found similar results for PET/CT. Interestingly, they also found an increase in the estimated SF below 1 million true counts in the lung. Unfortunately, the abdominal region was not assessed here either.

Our study has several limitations. The low number of patients with a relative homogeneous BMI and with different types of tumors does not allow for general recommendations. The investigated regions were mostly located in the abdomen; other regions such as the thorax or the brain were not evaluated and are beyond the scope of our work. In the oncologic reading according to PERCIST, only target lesions were investigated; however, nontarget lesions can also have a significant influence on the evaluation of PET examinations in oncology. The initial doses in our study are already relatively low compared with guidelines, as it is generally accepted that prolonged acquisition times have the potential to compensate for lower doses (19). Bed positions of a patient were classified in different categories; because our study was performed in a clinical setting, patients show different body heights and body proportions. Therefore, the different beds do not always cover exactly the same body region or organs. Also, the random deletion of PET events from list-mode data assumes a linear dependence of true and random events to the single counts (similar to Schaefferkoetter et al. (21)). However, it is known that randoms show a quadratic dependence. Even though results from Gatidis et al. (20) indicated that there is no significant difference between measured and simulated PET images for the activity concentration range used here, our results represent a worst-case scenario; image quality might be improved for measured low-dose PET images because randoms and dead time should be lower at lower doses. Finally, all PET images in all simulated doses were reconstructed with the same parameters. However, for PET images of lower doses an adjustment of reconstruction parameters such as gaussian filtering or the number of subsets might lead to improved image impressions or lesion detections.