The radiation dose delivered to the patient by hybrid imaging is increased as both the CT component and the nuclear medicine component (PET or SPECT) are high-dose examinations. The large variation in dose levels is due to different factors such as the clinical use of the CT data and the range of CT technologies that are used in hybrid imaging systems. Since CT images can be used for attenuation correction only, attenuation correction and anatomical localisation or diagnostic purposes, lower patient exposures are often sufficient to answer the relevant clinical question.

Because of the complex anatomy of the brain, precise localisation of suspicious foci of FDG uptake is difficult. Iball et al. (UK, [8]), Kaushik et al. (India, [38]) and Marti-Clement et al. (Spain, [12]) reported a CTDIvol of around 13 mGy for a CT scan with as purpose attenuation correction and localisation, while a value of 7.1 mGy was found during a Swiss survey. A Nordic-wide survey, gathering data from facilities in Denmark, Finland, Norway, and Sweden, even resulted in a mean value of 15 mGy for the same examination [24]. As expected, higher doses were found for diagnostic CT scans, ranging from 16 to 46 mGy. The significantly wider range in diagnostic CTDIvol values clearly comes forward in our meta-analysis suggesting that there is quite some room for further dose optimisation. It is important to notice that although CTDIvol values of brain examinations should be linked to the 16 cm diameter reference phantom, they often are not. Therefore, CT dose values related to the 32 cm phantom should be converted to the 16 cm phantom when comparing CT doses. The majority of included studies reported the applied reference phantom and/or the conversion of CT doses to the 16 cm phantom. However, since not all studies reported this the observed range of CTDIvol values may be influenced by values reported to a different reference phantom. Mean DLP values for an attenuation correction and anatomical localisation examination were smaller in the Swiss study compared to the UK, Spain, and Nordic study. This is mainly due to the much lower CTDIvol while mean scan ranges are quite similar. DLP values for the diagnostic CTs, on the other hand, were comparable between the UK and Swiss study while a much higher value was reported in the Nordic study. However, the latter results from only one dataset and has to be taken with caution.

In this study, a distinction is made between whole body and half body examinations in which the body is scanned from head to mid-thigh and from neck to mid-thigh, respectively. In general, CT dose values are reported to the 32 cm reference phantom and scan ranges vary between 70 and 100 cm. The study of Alessio et al. (USA, [21]) had a significantly shorter scan range of 57 cm because the CT scans were performed from head to bladder. However, while the scan range was shorter, the mean CTDIvol of a diagnostic CT scan was higher compared to other countries such as Switzerland (7 mGy) and the United Kingdom (5 mGy). As a result, higher than average DLP values were registered. A larger scan range (85 to 124 cm), nevertheless, causes the even higher DLP values found in the Nordic-wide survey [24]. For CT scans used for attenuation correction and localisation, most dose values were comparable with or higher than the values reported by Iball et al. [8], suggesting that there is still room for improvement in whole body PET/CT imaging. This improvement includes not only dose optimisation but also the description of a whole body CT scan. Most of the included studies described a whole body CT scan as a scan from head to mid-thigh while others called this a half body CT scan. Some studies also reported a second whole body scan defined as a scan from the head to the toes. This variability in description, and thus in scan length, is a confounding factor when analysing whole body CT dose data. Therefore, data of scans performed from head to toes were not used in this analysis.

Our meta-analyses demonstrated that the importance of the choice between a diagnostic or an attenuation correction and localisation CT scan should not be underestimated. Patients undergoing a diagnostic CT scan are exposed to significantly higher radiation doses. Depending on the nuclear medicine department, a diagnostic CT scan may also be acquired differently. In most cases they are taken with injected contrast medium and in breath-hold, as would be the case at the radiology department. However, a diagnostic CT as part of a nuclear medicine examination may be more justified if it reduces the need for the patient to attend a separate diagnostic CT on a dedicated diagnostic CT scanner for the same clinical indication.

Proposed national DRLs (nDRLs) for CTDIvol of brain PET/CT examinations are quite similar for attenuation correction and localisation CT scans (Table 2). However, only a few countries reported nDRLs values. For the whole body PET/CT examinations some variance in proposed nDRLs is observed. Although some countries used a different approach to obtain nDRLs, calculating the 75th percentile of the distribution of all data instead of the 75th percentile of the distribution of mean dose values as proposed by the ICRP, this is probably not the cause for the observed variability. It is probably due to differences in the description and classification into clinical purpose of these examinations. Generic descriptions used to describe some CT procedures can thus be a potential source of ambiguity since a protocol with the same name can be applied to different body regions [8, 13]. The observed variability in both mean dose values and nDRLs suggests the need of CT dose optimisation together with a breakdown into the different descriptions of whole body examinations. Due to the lack of data needed to calculate the standard errors, no pooled estimate could be made for the nDRLs.

Screening for bone metastasis is a common indication for skeletal scintigraphy or bone imaging and therefore 99mTc-diphosphonate is used. Because many benign skeletal processes demonstrate an increased radionuclide uptake, anatomic CT imaging can help differentiate benign from malignant lesions. 99mTc-diphosphonate also plays a role in detecting or excluding the presence of infections. Localising the site of infection to soft tissue or bone with the help of CT data has a crucial impact on the type of therapy. As expected, dose values are higher for CT acquisitions with a diagnostic purpose. The Nordic study, however, reports a lower CTDIvol for the diagnostic CT scan than for the localisation CT scan. However, because only one dataset reported CT doses of a diagnostic CT scan compared to 30 datasets in case of a localisation CT this result has to be taken with caution. It only suggests that a diagnostic CT as part of SPECT/CT bone examinations is not common in the Nordics [24]. In general, our meta-analyses suggest that mean CT doses are almost twice as high when the patient undergoes a diagnostic CT scan compared to an attenuation correction and localisation CT. However, since only three studies with a diagnostic CT scan are included of which one resulted in much lower dose values because of the above mentioned reason, mean diagnostic CT doses may be higher than observed in this meta-analysis. The lower observed dose values for the Nordic and Japanese study may also be explained by the use of a dedicated low-dose diagnostic CT protocol instead of a default diagnostic CT protocol. For the attenuation correction and localisation case, small pooled 95% confidence intervals are observed indicating that there already exists some consistency in SPECT/CT bone scan examinations between different countries.

SPECT cardiac imaging using 99mTc is a sensitive modality to assess many cardiac diseases such as ischemia and infarctions but is affected by the attenuation of other organs in the field of view, including the breasts and diaphragm. Applying CT attenuation correction overcomes this issue to a large extent and offers high quality, artefact-free radionuclide images. In general, the CT component of a SPECT/CT cardiac examination is only used for attenuation correction. Nevertheless, a large variation in DLP values is found in literature with reported doses in Switzerland [13] being 10 times larger than in Canada [3]. CTDIvol values resulting from national surveys are quite similar, suggesting that variations in DLP are due to differences in scan length.

MIBG and octreotide are both used to localise neuroendocrine tumours. Although reported dose values of the attenuation correction and localisation CT scan are scarce, doses between different countries seem to be similar. This is also demonstrated in our meta-analysis where the pooled mean CT doses, taking into account the weight of the different studies, are similar to the country specific doses.

Precise localisation of the (para)thyroid adenoma, which can be acquired by a 99mTc SPECT/CT scan, is crucial for the success of minimally invasive (para)thyroidectomy. For parathyroid imaging, DLP values are almost all similar in the case that the CT is used for localisation and CTDIvol values range from 4.1 to 7.4 mGy, while for diagnostic purposes CT doses are higher to achieve the required image quality. This is also demonstrated by our meta-analysis. Larger variations in DLP are seen for thyroid imaging suggesting a variation in scan length because CTDIvol values are comparable. Following ablative therapy, the standard current practice in patients with well-differentiated thyroid cancer after radioiodine therapy is 131I SPECT/CT imaging. This allows accurate staging of the disease and tailors the management for the patient appropriately. Here as well, the CT doses are comparable between examinations performed in Bulgaria and the United Kingdom. The mean DLP found in the Canadian study is twice as high, suggesting larger scan lengths. However, this value is the result of a study on one SPECT/CT system and has to be taken with caution.

The SPECT/CT nDRLs for CTDIvol (Table 3) are all in reasonable agreement with each other, expect those of the bone imaging protocols in Switzerland which is due to the discrete categorizations of individual bone protocols according to the different anatomical structures commonly scanned [13]. For all examinations, high variability in nuclear medicine practice was observed. Because of missing data needed to calculate the standard errors, no pooled estimate could be made for the nDRLs.

Hybrid imaging is the combination of two potentially high-dose investigations which justifies the need for dose optimisation research. For the nuclear medicine component, the most important factor determining the image quality is the administered activity which must respect any diagnostic reference level and is chosen based on several patient dependent clinical and technical aspects. To reduce the radiation dose to the patient many factors can be considered. Of course, the administered activity can be lowered to reduce the radiation dose. This depends not only on patient characteristics but also on the reconstruction algorithm used. Moreover, the use of radiopharmaceuticals with shorter physical and biological half-lives generally results in lower doses. Technological improvements, such as new high-sensitivity collimators and high-efficiency camera systems with more sensitive detectors, further induce dose optimisation possibilities. [39, 40]

CT dose optimisation in hybrid imaging depends first of all on the purpose of the CT acquisition. When recent diagnostic CT data are available and when follow-up studies are to be performed a low-dose CT is suggested. This is also the case for anatomical localisation or focal pathology characterisation. A diagnostic CT is only suggested when recent CT data are not available and when detailed anatomical information is needed. Because today most PET/CT and SPECT/CT systems have a CT part with full diagnostic image quality, a diagnostic CT as part of the nuclear medicine examination can avoid the need for the patient to attend a second imaging examination which reduces the radiation dose to the patient. In general, CT dose reduction principles in nuclear medicine are the same as in radiology. Optimising the CT scanning parameters such as tube potential, tube load, rotation time, beam width, pitch, reconstructed image thickness and applied reconstruction kernel is a first but necessary step. However, the acquisition for particular body regions is greatly influenced by the patient size, requiring patient specific settings. Using automatic tube current modulation further reduces the radiation dose by adapting the tube current to the patient’s size and anatomy. Some systems also provide organ-based tube current modulation reducing the dose to radiosensitive organs close to the surface of the body. Also, the use of iterative reconstruction algorithms allows different dose savings. Moreover, recent developments in artificial intelligence technology have introduced deep learning-based reconstruction techniques [41, 42]. Preliminary phantom and clinical studies have shown the potential of deep learning reconstruction for further dose reduction [42]. Throughout this optimisation process, it is important to consider the image quality as well because it will be influenced by changes in the CT scan and reconstruction parameters. Together with the integrated dose reduction tools, the clinical task and corresponding image quality requirements will eventually impose practical limits on achievable CT dose reduction. Although almost never specifically mentioned, the CT technology and image quality requirements will have an impact on the observed CT doses. [39, 40]

Although most CT protocols are determined as standard to the 16 and 32 cm IEC CT dosimetry phantom for head and body protocols, respectively, there may be some deviations. Therefore, it is important to convert the dose data, if necessary, to the dedicated reference phantom when comparing dose data between systems. However, most studies do not report the reference phantom in which the CTDIvol and DLP values are defined which may be a confounding factor.

While the Chi-square test is statistically significant for almost all included PET/CT studies, clearly indicating the presence of heterogeneity, it is in half of the cases non-significant for the included SPECT/CT studies (Additional file 1). However, this does not mean that heterogeneity is completely absent. Because the Chi-square test has a low power when only a few studies are included, the I2 statistics is better suited to assess heterogeneity. Most study groups of the included PET/CT and SPECT/CT examinations show moderate or high statistical heterogeneity, as measured with the I2 statistic. This supports the use of random-effects meta-analyses. However, for some study groups low or no heterogeneity was found. Tests for heterogeneity tend to be underpowered for detecting small differences if the number of studies and their sample sizes are low. Nevertheless, it is better to have a biased and imprecise estimate than to have no estimate at all. Attention just needs to be paid as heterogeneity cannot necessarily be excluded [43]. Next to the number of included studies, heterogeneity can also be influenced by the number of included devices per study for each type of examination and the amount of patient data collected per device and examination type. A nationwide survey includes of course more data. Finally, heterogeneity can also be explained by the PET/CT or SPECT/CT equipment used. Older systems may not contain any or the same dose reduction tools as the newest devices have. Default scan protocols may vary from manufacturer to manufacturer and from model to model. Differences in operator’s preferences influence CT radiation doses too.