Skip to main content
  • Original research
  • Open access
  • Published:

Preliminary protocol for measuring the reproducibility and accuracy of flow values on digital PET/CT systems in [15O]H2O myocardial perfusion imaging using a flow phantom



Several factors may decrease the accuracy of quantitative PET myocardial perfusion imaging (MPI). It is therefore essential to ensure that myocardial blood flow (MBF) values are reproducible and accurate, and to design systematic protocols to achieve this. Until now, no systematic phantom protocols have been available to assess the technical factors affecting measurement accuracy and reproducibility in MPI.

Materials and methods

We implemented a standard measurement protocol, which applies a flow phantom in order to compare image-derived flow values with respect to a ground truth flow value with [15O]H2O MPI performed on both a Discovery MI (DMI-20, GE Healthcare) and a Biograph Vision 600 (Vision-600, Siemens Healthineers) system. Both systems have automatic [15O]H2O radio water generators (Hidex Oy) individually installed, allowing us to also study the differences occurring due to two different bolus delivery systems. To investigate the technical factors contributing to the modelled flow values, we extracted the [15O]H2O bolus profiles, the flow values from the kinetic modeling (Qin and Qout), and finally calculated their differences between test-retest measurements on both systems.


The measurements performed on the DMI-20 system produced Qin and Qout values corresponging to each other as well as to the reference flow value across all test-retest measurements. The repeatability differences on DMI-20 were 2.1% ± 2.6% and 3.3% ± 4.1% for Qin and Qout, respectively. On Vision-600 they were 10% ± 8.4% and 11% ± 10% for Qin and Qout, respectively. The measurements performed on the Vision-600 system showed more variation between Qin and Qout values across test-retest measurements and exceeded 15% difference in 7/24 of the measurements.


A preliminary protocol for measuring the accuracy and reproducibility of flow values in [15O]H2O MPI between digital PET/CT systems was assessed. The test-retest reproducibility falls below 15% in majority of the measurements conducted between two individual injector systems and two digital PET/CT systems. This study highlights the importance of implementing a standardized bolus injection and delivery protocol and importance of assessing technical factors affecting flow value reproducibility, which should be carefully investigated in a multi-center setting.


Several factors may decrease the accuracy of Positron Emission Tomography (PET) quantification within one system or between different PET systems. These factors involve injection protocols, practical data acquisition with different kind of clinical protocols, different PET detector technologies, software implementations as well as reconstruction and data corrections [1,2,3]. The contribution of these factors is an additional challenge in myocardial perfusion imaging (MPI) quantification, as it relies on kinetic modelling to extract the myocardial blood flow (MBF) in absolute quantitative values, requiring accurate and reproducible image quantification.

The European Association of Nuclear Medicine Research Ltd. (EARL) has proposed guidelines for PET acquisition harmonization in order to achieve reproducible [18F]FDG tumor standardized uptake value (SUV) quantification [4, 5]. Similar procedural guidelines have been proposed for MPI, which frequently uses short lived tracers, such as [15O]H2O [6]. Implementing harmonization measures for MPI is extremely important. Any test used for clinical decision-making and patient management needs to be accurate and should have adequate test–retest repeatability to enable management decisions for individual patients [2, 7]. As MPI PET is increasingly being used for the detection of myocardial ischemia [6], there is a growing need for more consistent and standardized evaluation of the technical factors contributing to the variation of MBF values in MPI studies [3].

The lack of suitable test objects has hindered the implementation of standardized protocols for harmonization of MPI. Phantoms used in PET harmonization studies are usually static phantoms, such as the National Electrical Manufacturers Association (NEMA) image quality phantom [8] or simple cylindrical phantoms [9]. Few suitable phantoms have been available for harmonization purposes in dynamic MPI until recently. Gabrani-Juma et al. validated a flow phantom simulating MBF for MPI PET standardization purposes where image-derived flow values can be evaluated against a ground truth reference value [10]. This flow phantom offers an optimal basis for creating standardized protocols for assessing the accuracy and reproducibility of MBF values across PET systems as well as for investigating the contribution of purely technical factors on kinetic modelling and MBF quantitation.

Several studies have investigated harmonization measures of accurate SUV quantification for oncological [18F]FDG and neurological studies. Akamatsu et al. list several investigations of the latest PET harmonization strategies [11]. For example, SUV harmonization can be based on the image uniformity and spatial resolution with tracer-specific phantoms [12]. EARL has also performed extensive work for evaluating reproducible SUV quantitation and accreditation, and the investigations include harmonization protocols with standard acquisition parameters [5]. Furthermore, SUV (more specifically, SUVmean, SUVmax, and SUVpeak) has been reported to have 10% coefficient of variation [13] and 27% reproducibility rate for SUVmean and SUVpeak, and 33% for SUVmax [14].

Multiple test-retest studies have also evaluated the reproducibility of MPI PET with [13N]NH3 as well as [82Rb] [15,16,17]. In addition, in [15O]H2O MPI PET the repeatability for stress MBF values has been reported to be 27% [18] and 25% [19]. However, to the best of our knowledge, no studies have systematically investigated the contribution of technical factors on the test-retest reproducibility and accuracy with a calibrated standard. Therefore, the first step for implementing systematic harmonization protocols should be to evaluate these technical factors contributing to both the accuracy and reproducibility of MBF values between different bolus injector and PET/CT systems against physical reference standards. These evaluations might eventually result in standard clinical acquisition protocols that are capable of producing similar quantitative values across different PET systems and reducing the variation due to technical factors in MPI.

In this study, we evaluate the different technical factors affecting the accuracy and reproducibility of MBF values using a preliminary measurement protocol with two [15O]H2O bolus injector systems of the same manufacturer, as well as two digital silicon photomultiplier (SiPM) -based PET/Computed Tomography (CT) systems of different vendors, using a flow phantom capable of simulating MBF.

Materials and methods

PET systems

The study was conducted in a single-centre setting (Turku PET Centre). The protocol was implemented on two PET/CTs in Turku PET Centre; Discovery MI with 20 cm axial field-of-view (FOV) (DMI-20, GE Healthcare, Milwaukee, US) and Biograph Vision 600 (Vision-600, Siemens Healthineers, Erlangen, Germany). The system performance parameters are presented in Table 1.

Table 1 Performance characteristics of DMI-20 and Vision-600

Flow phantom

The flow phantom used in this study is presented in detail in the paper of Gabrani-Juma et al. [10], to which an interested reader is directed for a more extensive description. In short, the flow phantom comprises a closed-circulation system, which pumps water from an exterior container into the circulation using a peristaltic pump. The system is located inside a plastic shell filled with water, shaped as the NEMA image quality phantom that simulates body attenuation and scatter.

The pump flow rate, named Qpump may be adjusted by the user to produce a desired range of flow values. An injection port is located after the peristaltic pump, from which the activity can be administered directly to the system. Thereafter, the injected activity propagates directly to an input chamber (with a volume of 15.7 ml) which simulates the left ventricle blood pool. The input chamber is connected to an exchange cylinder that contains a perforated tube from which the tracer permeates to the exchange cylinder. The exchange cylinder simulates the myocardium, and the water flow from the perforated tube into the exterior volume of the exchange cylinder simulates the blood perfusion. The volume of the exchange cylinder (Vcyl) is 160 ml.

Qcyl marks the flow rate coming out of the exchange cylinder whereas Qtube marks the flow rate coming out of the perforated tube. This is because not all water passes from perforated tube to the exterior volume of the exchange cylinder. There is a flow controller valve for both Qcyl and Qtube that allow the user to create different types of restricted and unrestricted flow inside the flow phantom. Qcyl and Qtube are measured using calibrated flow meters (Omega Engineering Inc., Norwalk, US). The reference flow value Qref is derived from Qcyl using the flow meter calibration factors and a look-up table.

The input and tissue time-activity curves (TACs) needed for the kinetic modelling can be measured from the input chamber and exchange cylinder, respectively. This analysis is performed semi-automatically using a software provided by the phantom vendor, where the regions of interest can be defined and then used to derive the TACs.

A phantom-specific two-compartmental (one-tissue) kinetic model is used, which is extensively explained in Gabrani-Juma et al. (Equations 2–6) [10]. The model includes two rate parameters, qin and qout, as well as an input signal fraction (ISF) and a delay parameter. qin [min− 1] denotes the rate parameter that describes the flow passing from the perforated tube into the exterior volume of the exchange cylinder, and qout [min− 1] the rate parameter that describes the flow coming out from the exchange cylinder. ISF [dimensionless] accounts for the exchange cylinder spill-over from the perforated tube and delay [s] defines the tracer passage time from the input cylinder to the perforated tube. qin and qout are analogous to K1 and k2, the rate constants used in a two-compartmental kinetic model and commonly used to derive MBF e.g. for [15O]H2O [20, 21]. qin and qout are multiplied by Vcyl in order convert them to flow values Qin and Qout in [ml/min]. In the phantom, for an ideal measurement, Qin = Qout = Qref.

Measurement protocol

The measurement protocol assessed 12 different flow values, from low to high. Qpump values were set to 150 ml/min, 200 ml/min, and 250 ml/min. Qcyl values were adjusted to 20%, 40%, 60%, and 80% of Qpump using the constrictor valves, to simulate reduced perfusion. Qcyl and Qtube values were recorded before and after the measurement and their mean values were applied as the reference flow values. For Vision test 150 − 20%, 150 − 40%, and 150 − 80% measurements the recorded Qcyl and Qtube were recorded before the measurement. Table 2 presents all recorded flow values from the phantom measurements.

Table 2 Acquisition parameters for all measurements. Qcyl marks the flow rate coming out of the exchange cylinder and Qtube marks the flow rate passing out of the perforated tube to the exchange cylinder

All measurements were repeated twice within two weeks to one month on each PET/CT system, in order to ensure test-retest reproducibility of the flow values, without changing the phantom set-up. All measurements were performed in the same order. Before each measurement session, the peristaltic pump was calibrated using Qpump 200 ml/min. The [15O]H2O bolus was automatically administered into the flow phantom circulation via the [15O]H2O RadioWaterGenerator (RWG, Hidex Oy, Turku, Finland) system. Both the DMI-20 and Vision-600 system have their individual RWG systems installed next to the PET/CT system gantry.

Both RWGs are daily cross-calibrated with the radionuclide calibrator for [15O]H2O. In the calibration procedure, the detected counts are time-corrected to the reference point, i.e., the time-point when the peak reaches the end of the infusion line. Based on the corrected counts, decay time, and radionuclide calibrator data, the software generates a calibration coefficient for the detector. Thereafter, the injected activity in units of [MBq] is calculated from the count-rate curve based on the calibration factor. The manufacturer guarantees 15% accuracy within the requested activity production level on the system. The requested activity for each measurement was 500 MBq on both systems.

All injected activities with relative differences between test- and retest measurements are given in Table 3. To investigate the reproducibility of the activity administration between the measurements and the system, the bolus curves were extracted from both injector systems. Overall, due to using two different injector systems, the injected activities for DMI-20 were generally lower than 500 MBq, whereas the activities were higher than 500 MBq for Vision-600, although all the measurements were within the manufacturer specified 15% limit.

Table 3 Injected activities for all measurements. The manufacturer guarantees 15% accuracy to 500 MBq.

PET/CT acquisition

PET acquisition and image reconstruction was conducted in dynamic acquisition mode and followed the clinical MPI acquisition protocol used at our institute (Turku PET Centre) [22]. The acquisition duration was divided into 24 time frames, 14 × 5 s, 3 × 10 s, 3 × 20 s, and 4 × 30 s. Prior to each PET acquisition, a measurement-specific CT-based attenuation correction (CTAC) was acquired. The values used in the CT acquisitions were: tube voltage of 120 kV, exposure time of 500 ms, pitch factor of 1.375, tube current of approx. 96 mA (test) and approx. 61 mA (retest) on DMI-20, and tube voltage of 100 kV, exposure time of 500 ms, pitch factor of 1.2, and tube current of approx. 30 mA (test), and approx. 30 mA (retest) on Vision-600.

All data was decay corrected for the injected activity at the start of the measurement, which was derived directly from the data from the injector system. All quantitative corrections were applied for the PET data. Image reconstructions followed the clinical protocol used at our institute [22]. The reconstruction parameters are presented in Table 4.

Table 4 Reconstruction parameters used in this study

Image analysis

All PET images were analyzed using the QuantifyDCE software (Shelley Medical Imaging Technologies, Ontario, Canada), which is specifically developed for the analysis of the flow phantom data. The details of the image analysis and kinetic modelling are presented in Gabrani-Juma et al. [10]. In short, the software requires the user to define the input chamber and exchange cylinder volumes-of-interests (VOIs) from which the software automatically extracts the input and tissue TACs. Thereafter, the software automatically applies a two-compartmental kinetic model for the TACs, which outputs Qin, Qout, delay, and ISF values. The image analysis was conducted similarly for all data sets by a single operator, and ROI locations were fixed for each measurement session.

Data analysis

For quality control, the flow meter readings from Qcyl and Qtube were extracted for all measurements and compared in terms of relative difference to the expected value from Qpump. We report the relative differences for the sum Qtube + Qcyl in comparison to expected Qpump for all measurements.

The comparison of the bolus curves and TACs was conducted visually as well as quantitatively by evaluation of the areas-under-the-curves (AUCs), whereas the modelled flow values were compared quantitatively and statistically.

The extracted bolus activity curves from the [15O]H2O generator were inspected visually for each measurement to investigate their reproducibility. Bolus AUCs were computed from each bolus curve to investigate their contribution to the measured input TAC as well as the tissue TAC from the phantom. Both input and tissue TACs with the modelled TACs were inspected visually to assess the reproducibility between the systems.

Thereafter, comparison for Qin and Qout values was made between all test and retest measurements for both PET/CT systems. The absolute error of Qin and Qout with respect to Qref was calculated as

$$Flow\,value\,error\, = \,\left| {\frac{{Flow\,value\, - \,Qref}}{{Qref}} * 100\% } \right|,$$

and is reported in percentage (%) units.

The relative error of modelled flow values was calculated thereafter using the following equation:

$$Repeatability\,error\, = \left| {\frac{{Flow\,value\,\left( {retest} \right) - Flow\,value\,\left( {test} \right)}}{{Flow\,value\,\left( {test} \right)}} * 100\,\% } \right|,$$

and is also reported in percentage (%) units. Thereafter, correlation and agreement between Qin and Qout values was compared on both systems using correlation and Bland-Altman plots.


Figure 1 shows the differences of the flow meter recordings from the sum of the Qcyl and Qtube values with respect to the set Qpump values on the measurements from DMI-20 and Vision-600. The measurements performed on DMI-20 showed similar difference ranges between test and retest measurement for different flow settings, whereas the measurements performed on Vision-600 produced larger differences on test measurements. In all flow phantom measurements, the flow meter readings showed smaller than 15% relative difference. Overall, the magnitude of the errors was similar for DMI-20 test and retest measurements as well as retest measurements on Vision-600. On Vision-600 200 − 40%, 250 − 40%, 250 − 60%, and 250 − 80% test measurements the relative differences were larger than 10%.

Fig. 1
figure 1

Qcyl + Qtube relative errors with respect to Qpump presented for all measurements on DMI-20 (left) and Vision-600 (right)

Figure 2 shows the bolus curves extracted from test and retest measurements on both RWG dispenser systems installed on the DMI-20 and Vision-600 PET/CT systems. The RWG installed on the Vision-600 system produced higher bolus peaks compared to the RWG installed on the DMI-20 system. Subsequently, the measurements performed on Vision showed higher injected activities (Table 3) and higher peaks for the input TACs (Fig. 3).

Fig. 2
figure 2

[15O]H2O bolus curves extracted for each measurement on the RWG system installed on DMI-20 (left) and Vision-600 (right)

Fig. 3
figure 3

Input, tissue, and modelled time-activity curves (left), and zoomed tissue and modelled TACs (right) recorded on Vision-600 from all measurements

Figures 4 and 3 show TACs from all measurements recorded on both PET/CT systems. There are amplitude differences in input and tissue TACs between the DMI-20 test and retest measurements. On Vision-600 the amplitude differences in input TACs between test and retest measurements are smaller than on DMI-20 but the tissue TACs show differences in shapes between test and retest measurements. The tissue TACs clearly intersect in measurements 150 − 20%, 150 − 60%, 200 − 40%, 250 − 20%, 250 − 60%, and 250 − 80%. The peak amplitudes of all TACs on the measurements on Vision-600 are higher compared to the measurements on DMI-20.

Fig. 4
figure 4

Input, tissue, and modelled time-activity curves (left), and zoomed tissue and modelled TACs (right) recorded on DMI-20 from all measurements

Table 5 shows Qin and Qout flow values, as well as their flow value errors with respect to Qref from Eq. (1) for test and retest measurements. Overall, most of the flow value errors fall below 10% on both test and retest measurements performed on DMI-20. However, the Vision-600 test and retest measurements show higher flow value errors especially with Qout compared to DMI-20, several being larger than 10%.

Table 5 Modelled Qin and Qout values as well as their absolute relative errors with respect to Qref from all measurements on both DMI-20 and Vision-600

Figure 5 shows repeatability errors of Qin and Qout between test and retest measurements (Eq. 2) on the measurements on DMI-20 and Vision-600. All DMI-20 measurements fall below 15% repeatability errors whereas Vision-600 shows larger than 15% errors on four measurements (150 − 20%, 200 − 40%, 250 − 20%, and 250 − 80%). The mean ± standard deviation of errors for DMI-20 Qin and Qout are 2.1% ± 2.6% and 3.3% ± 4.1%. The corresponding numbers for Vision-600 Qin and Qout are 10% ± 8.4% and 11% ± 10%, respectively.

Fig. 5
figure 5

Qin and Qout differences between test and retest measurements on DMI-20 (left) and Vision-600 (right)

Figure 6 shows that Qin and Qout values are close to the reference flow values (Qref) on DMI-20 and show resemblance between test and retest measurements. In addition, the linear polynomial fits are close to the reference line for both DMI-20 Qin and Qout. In comparison, on the measurements performed on Vision-600 both Qin and Qout values diverge from the reference line (Fig. 6) but the linear polynomial fit is similar for both Qin and Qout.

Fig. 6
figure 6

Correlation of DMI-20 (left) and Vision-600 (right) test and retest measurements presented with the reference flow line

Figure 7 describes the correlation of Qin and Qout flow values between DMI-20 and Vision-600 test and retest measurements. For Qout values the retest measurements are close to the line-of-identity and Qin test and retest measurements have similar deviation from the line-of-identity. The deviation between DMI-20 and Vision-600 is higher with larger flow rates in test and retest measurements for Qin and for test measurements for Qout.

Fig. 7
figure 7

Qin (left) and Qout (right) values extracted from test and retest measurements for DMI-20 and Vision-600. The fitted lines show the correlation between DMI-20 and Vision-600 separated for the test- and retest measurements and line-of-identity the ideal correlation between DMI-20 and Vision-600

The Bland-Altman plots in Fig. 8 show that the Qout values vary more compared to Qin between DMI-20 and Vision-600. The lines-of-agreement (LoAs) are larger for Qout compared to Qin LoAs. Also, the mean differences between DMI-20 and Vision-600 are 1.59 ml/min and −17.3 ml/min for Qin and Qout. There is one outlier point for the test measurement in both Qin and Qout. The outlier is measurement 250 − 80% where DMI-20 produces test Qin and Qout values of 179 ml/min and 176 ml/ min and the corresponding values for Vision-600 are 231 ml/min and 270 ml/min.

Fig. 8
figure 8

Bland-Altman plots for Qin (left) and Qout (right) value between DMI-20 and Vision-600


This study proposed a preliminary protocol for assessing the technical factors contributing to the accuracy and reproducibility of quantitative flow values in MPI. This protocol could eventually be used for planning harmonization measurements for myocardial perfusion imaging using [15O]H2O as well as the flow phantom in the future. We evaluated the impact of technical factors on the modelled flow values, Qin and Qout, extracted from the flow phantom. The accuracy and reproducibility was evaluated between test and retest measurements, two [15O]H2O injectors as well as two digital PET systems, Discovery MI and Biograph Vision 600. There is no specific reason why the protocol could not be applied to analog PET/CT systems as well.

First, we evaluated how well the measured Qcyl and Qtube flow values inside the phantom fulfill the ideal pre-assumption that Qtube + Qcyl should be equal to Qpump. Almost all measurements on both DMI-20 and Vision-600 showed errors between Qtube + Qcyl vs. Qpump. Especially the test measurements on Vision-600 produced larger Qtube + Qcyl differences to Qpump when compared to the retest measurement on Vision-600, or test and retest measurement on DMI-20. It is still worth noticing that all Qtube + Qcyl vs. Qpump differences were below 15% on both systems (Fig. 1). This finding suggests that the flow phantom set-up is a technical factor that plays a role in the measurement accuracy.

Considering in applying the phantom for larger multi-institutional studies, the flow phantom has internal characteristics that may affect the measurement. For example, the pressure variations inside the hoses, the air bubbles within the phantom, the air pressure and humidity of the scanner room, and the phantom flow meter inaccuracies can affect the modelled flow values. In addition, the flow phantom presents only a simplified simulation of myocardial perfusion. In clinical subjects, there will be more variations due to physiological factors, including tracer dispersion and individual reactions to pharmacological stress, in addition to other physiological factors that contribute to the measured MBF. However, with correct procedures implemented the phantom can be used to investigate the technical factors independent of the physiological characteristics.

What is more, the phantom vendor advices performing quality control procedures prior conducting the measurements. One procedure is calibrating the peristaltic pump for a measurement session. We however noticed that the measurements could benefit even more by calibrating the pump whenever we altered Qpump or the constriction of Qcyl or Qtube between tests. Although noting this effect, we decided to use a single pump calibration factor between measurements. What is more, the flow meter calibration applied in the QuantifyDCE software should verify the flow meter readings and therefore a single pump calibration should be sufficient. Given that the errors were smaller than 15% and that these variations would still be accounted in kinetic modelling, we assume the Qtube + Qcyl error is not the most significant factor affecting the accuracy of the modelled flow values.

Second, we measured the administered [15O]H2O activities as well as bolus peaks in all measurements and assessed their performance between test and retest measurements as well as the PET/CT systems. The administered activities were repeatable between test and retest measurements with relative differences lower than 9% on both systems (Table 3). All administered activities were within 15% of the requested 500 MBq and fell within the vendor specifications. When comparing the PET/CT systems, the RWG installed on the Vision-600 produced systematically higher activities as well as bolus peak amplitudes. The variation in bolus peak amplitudes is possibly due to the discrepancies in cross-calibration of the two RWG systems. Therefore, the use of individual RWGs on each PET/CT system might be a technical factor affecting the measurement accuracy, especially between the PET/CT systems used.

Third, we evaluated the accuracy and reproducibility of the modelled flow values. The errors of Qin and Qout with respect to Qref were smaller on DMI-20 compared Vision-600 showing higher accuracy on DMI-20. Also, we could show that on DMI-20 all measurements were highly repeatable as the repeatability errors of Qin and Qout were below 15% (Fig. 5; Table 5). In comparison, on Vision-600 the repeatability errors were higher for almost all measurements and increased as high as 40% on one measurement (150 − 20%). Moreover, the correlation plots of Qin and Qout between DMI-20 and Vision-600 showed discrepancies (Fig. 7) with more variability on higher flow rates (Fig. 8). This variability will likely have higher impact on MPI patients during pharmacological stressing due to higher MBF.

The most probable explanation for the reproducibility differences between DMI-20 and Vision-600 is the difference in tissue TACs between the test and retest measurements on Vision-600. After careful inspection we noted that whenever there was a clear shape discrepancy in the tissue TACs or the tissue TACS were intersecting between test and retest measurements, there was a larger test-retest error in Vision 600. On DMI-20 we did not observe such a trend (see Supplementary File 1).

This finding indicates that even though the measurement set-ups between test and retest sessions were as repeatable as possible, there still exists a fundamental source of variability between the two PET/CT systems, which can be attributed to be originating from technical factors. Therefore we strongly advice investigating other factors affecting the flow value modelling accuracy. For example, one possible factor affecting the discrepancy might be the reconstruction differences between Vision-600 and DMI-20. EARL has proposed guidelines in oncological imaging to apply reconstruction parameters that result into equal voxel sizes regardless of the PET system used [4]. In this study, on DMI-20 the voxel sizes were (x, y, and z) 1.82 mm, 1.82 mm, and 2.79, and on Vision-600 1.65 mm, 1.65 mm, and 3. Thus, the final reconstructed image resolution on both systems was similar. We still advise studying this phenomenon in more detail and perform reconstruction harmonization studies between DMI-20 and Vision-600 in order to eliminate this factor from affecting the modelling accuracies.

What is more, one factor that might explain the differences between the systems may be the higher sensitivity of the Vision-600 compared to DMI-20 (13.7 cps/MBq on DMI-20 and 16.4 cps/MBq on Vision-600) as well as smaller spatial resolution of Vision-600 compared to DMI-20 (4.1 mm on DMI-20 and 3.5 mm on Vision-600) (Table 1). Additional factor affecting the differences between DMI-20 and Vision-600 is in the implemented scatter correction algorithm and the difference in its behavior. The scatter correction factors especially in the Vision-600 measurements were seen to vary across time between test and retest measurements, although no such behavior was seen on the DMI-20 (please see Supplementary File 2). Finally, even though we could not show any direct impact between bolus amplitudes and flow value modelling differences, this is a factor that should be minimized between the PET/CT systems used.

In summary, this study has identified the following technical factors affecting the accuracy and reproducibility of modelled flow values between as well as within the PET/CT systems. The factors are (1) the flow phantom set-up, (2) the injected bolus, (3) the RWG producing the injected bolus, (4) the flow-rate, (5) the reconstruction parameters, (6) the sensitivity of the PET/CT systems, and (7) the scatter correction algorithm, as well as any subsequent data corrections. The contribution of these factors should be investigated in more detail.

When comparing the results of this study to the data published from MBF values measured on clinical subjects, the mean values and standard deviations of the repeatability errors between test and retest measurements on Vision-600 (Qin: 10% ± 8.4% and Qout: 11% ± 10%) or DMI-20 (Qin: 2.1% ± 2.6% and Qout: 3.3% ± 4.1%) were within a similar or smaller range that is measured in clinical myocardial perfusion studies [23]. For example the table in Klein et al. describes that overall the stress MBF values have repeatability accuracy from 11 to 34%, and for [15O]H2O the reported values are 27% [18] and 25% [19]. In this regard, the measured values from the phantom study agree well with the results gained from the clinical subjects and brings light on the level of reproducibility achieved without biological factors. As compared to our phantom study, higher variation in patient studies reflect the contribution of differences in biological factors, such as the systemic hemodynamic state, on MBF.

In the future, there is room for validating the protocol for multiple tracers used in MPI. The present protocol was assessed only for [15O]H2O but also [82Rb] and [13N]NH3 are commonly used in MPI as perfusion tracers. Thus, at least a cross-verification to the present protocol or designing a tracer-specific protocol for these tracers is required. Moreover, modifying the protocol to be used with [18F]-labelled tracers would be highly useful, as they are widely available in different centers. Thereafter, preliminary harmonization measures could be evaluated using also 18F-tracers, as long as the protocol would have been cross-calibrated with the present [15O]H2O protocol.

Summary of the findings

In this study we were able to demonstrate 15% repeatability across all measurements on DMI-20 and on 7/24 measurements on Vision-600. This relatively high repeatability is expected, as in a single-center setting there are several factors that are advantageous for minimizing the bias and variability between the measurements. First, the [15O]H2O bolus injectors were calibrated to a common reference within the center. Second, the acquisition protocols and the activity delivery were standardized between the measurements. Third, both PET systems were cross-calibrated to the common reference within the center, their acquisition durations as well as reconstruction frame times were the same, with relatively similar settings in the reconstruction parameters.

These experiments confirmed that the proposed imaging protocol with several flow values as well as imaging parameters including reconstructions can be applied between different injector and PET/CT systems to provide an understanding for technical accuracy and reproducibility of the quantitative flow values between test-retest measurements. However, after all there still resumes an underlying question: where do the higher errors on Vision-600 repeatability origin from as on DMI-20 such errors were not visible? This calls for further attention to assess and minimize the contribution of different factors between PET/CT systems.

A future multi-center study would be essential to provide upper and lower limits for the Qin and Qout values for calibration purposes and would give specific and more detailed information about the different factors contributing to the variations of MBF values between centers and PET/CT systems. Eventually these findings would provide information for establishing similar quality control measures that EARL recommends for oncological PET imaging. Before applying this protocol for multi-center setting, a careful investigation for standardizing the bolus delivery between different sites and dispenser systems needs to be conducted first. Ensuring an even more consistent tracer administration profile should improve test-retest repeatability as well as system-to-system reproducibility further [23]. However, in multi-center settings there will be several more technical factors affecting the measurements, all of which should be investigated separately.


A preliminary protocol for measuring the accuracy and reproducibility of flow values in [15O]H2O MPI between digital PET/CT systems was assessed. The test-retest reproducibility falls below 15% in majority of the measurements conducted between two individual injector systems and two digital PET/CT systems. This study highlights the importance of implementing a standardized bolus injection and delivery protocol, which should be carefully investigated in a multi-center setting. The study indicated that there still remain a number of technical factors which could be investigated further, to further minimize their effect to accuracy and reproducibility of flow values within and between PET/CT systems.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.


  1. deKemp RA, Klein R, Beanlands RSB. 82Rb PET imaging of myocardial blood flow—have we achieved the 4 rs to support routine use? EJNMMI Res. 2016. 6:.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Moody JB, Lee BC, Corbett JR, et al. Precision and accuracy of clinical quantification of myocardial blood flow by dynamic PET: a technical perspective. J Nucl Cardiol. 2015;22:935–51.

    Article  PubMed  Google Scholar 

  3. Sciagrà R, Lubberink M, Hyafil F, et al. EANM procedural guidelines for PET/CT quantitative myocardial perfusion imaging. Eur J Nucl Med Mol Imaging. 2021;48:1040–69.

    Article  PubMed  Google Scholar 

  4. Boellaard R. Standards for PET Image Acquisition and Quantitative Data Analysis. J Nucl Med. 2009;50. :11S-20S.

  5. Aide N, Lasnon C, Veit-Haibach P, et al. EANM/EARL harmonization strategies in PET quantification: from daily practice to multicentre oncological studies. Eur J Nucl Med Mol Imaging. 2017;44:17–31.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  6. Knuuti J, Wijns W, Saraste A, et al. 2019 ESC guidelines for the diagnosis and management of chronic coronary syndromes. Eur Heart J. 2020;41:407–77.

    Article  PubMed  Google Scholar 

  7. deKemp RA, Celiker Guler E, Ruddy TD. More evidence for adequate test–retest repeatability of myocardial blood flow quantification with 82Rb PET/CT. J Nucl Cardiol. 2021;28:2872–5.

    Article  PubMed  Google Scholar 

  8. Devriese J, Beels L, Maes A, et al. Impact of PET reconstruction protocols on quantification of lesions that fulfil the PERCIST lesion inclusion criteria. EJNMMI Phys. 2018;5:1–13.

    Article  Google Scholar 

  9. Namias M, Bradshaw T, Menezes VO, et al. A novel approach for quantitative harmonization in PET. Phys Med Biol. 2018. 63:.

    Article  PubMed  Google Scholar 

  10. Gabrani-Juma H, Clarkin OJ, Pourmoghaddas A, et al. Validation of a Multimodality Flow Phantom and its application for Assessment of Dynamic SPECT and PET technologies. IEEE Trans Med Imaging. 2017;36:132–41.

    Article  PubMed  Google Scholar 

  11. Akamatsu G, Tsutsui Y, Daisaki H et al. (2023) A review of harmonization strategies for quantitative PET. Ann Nucl Med.

  12. Fahey F, Christian P, Zukotynski K, et al. Use of a qualification phantom for PET brain imaging in a multicenter consortium: a collaboration between the pediatric brain tumor consortium and the SNMMI clinical trials network. J Nucl Med. 2019;60:677–82.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  13. Lodge MA. Repeatability of SUV in oncologic 18F-FDG PET. J Nucl Med. 2017;58:523–32.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  14. Koopman D, Jager PL, Slump CH, et al. SUV variability in EARL-accredited conventional and digital PET. EJNMMI Res. 2019;9.

  15. Byrne C, Kjaer A, Olsen NE, et al. Test–retest repeatability and software reproducibility of myocardial flow measurements using rest/adenosine stress Rubidium-82 PET/CT with and without motion correction in healthy young volunteers. J Nucl Cardiol. 2021;28:2860–71.

    Article  PubMed  Google Scholar 

  16. Monroy-Gonzalez AG, Juarez-Orozco LE, Han C, et al. Software reproducibility of myocardial blood flow and flow reserve quantification in ischemic heart disease: a 13 N-ammonia PET study. J Nucl Cardiol. 2020;27:1225–33.

    Article  PubMed  Google Scholar 

  17. Suda M, Onoguchi M, Tomiyama T, et al. The reproducibility of time-of-flight PET and conventional PET for the quantification of myocardial blood flow and coronary flow reserve with 13 N-ammonia. J Nucl Cardiol. 2016;23:457–72.

    Article  PubMed  Google Scholar 

  18. Wyss CA, Koepfli P, Mikolajczyk K, et al. Bicycle exercise stress in PET for assessment of coronary flow reserve: repeatability and comparison with adenosine stress. J Nucl Med. 2003;44:146–54.

    PubMed  Google Scholar 

  19. Kaufmann PA, Gnecchi-Ruscone T, Yap JT, et al. Assessment of the reproducibility of baseline and hyperemic myocardial blood flow measurements with 15O-labeled water and PET. J Nucl Med. 1999;40:1848–56.

    CAS  PubMed  Google Scholar 

  20. Iida H, Kanno I, Takahashi A, et al. Measurement of absolute myocardial blood flow with H2150 and dynamic positron emission tomography: strategy for quantification in relation to the partial volume effect. Circulation. 1988;78:104–15.

    Article  CAS  PubMed  Google Scholar 

  21. Iida H, Rhodes CG, de Silva R, et al. Use of the left ventricular time-activity curve as a noninvasive input function in dynamic oxygen-15-water positron emission tomography. J Nucl Med. 1992;33:1669–77.

    CAS  PubMed  Google Scholar 

  22. Kajander S, Joutsiniemi E, Saraste M, et al. Cardiac positron emission tomography/computed tomography imaging accurately detects anatomically and functionally significant coronary artery disease. Circulation. 2010;122:603–13.

    Article  CAS  PubMed  Google Scholar 

  23. Klein R, Ocneanu A, Renaud JM, et al. Consistent tracer administration profile improves test–retest repeatability of myocardial blood flow quantification with 82Rb dynamic PET imaging. J Nucl Cardiol. 2018;25:929–41.

    Article  PubMed  Google Scholar 

Download references


This study was conducted within the Finnish Center of Excellence in Molecular Imaging in Cardiovascular and Metabolic Research supported by the Academy of Finland, University of Turku, Turku University Hospital and Åbo Akademi University.


This study has received funding from the Insrumentarium Science Foundation, Finnish Foundation for Cardiovascular Research, EMPIR programme co-financed by the Participating States, the European Union’s Horizon 2020 research and innovation programme (15HLT05 PerfusImaging, 19SIP04 TracPETperf), the UK’s Department for Science, Innovation and Technology (DSIT), Hospital District of Southwest Finland (EVO funding, project number 11209), and the Doctoral Programme of Clinical Research in the University of Turku.

Author information

Authors and Affiliations



Conceptualization, R.S., J.T.; Methodology, R.S., J.T., H.P., L.K., T.T., Validation, A.F., N.A.S., M.T., A.S., Investigation, R.S., J.T., L.K., Writing-Original Draft Preparation, R.S., Supervision, M.T., A.S., J.T.

Corresponding author

Correspondence to Reetta Siekkinen.

Ethics declarations

Competing interests

We declare no conflict of interest.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Supplementary Material 2

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Siekkinen, R., Partanen, H., Kukola, L. et al. Preliminary protocol for measuring the reproducibility and accuracy of flow values on digital PET/CT systems in [15O]H2O myocardial perfusion imaging using a flow phantom. EJNMMI Phys 11, 54 (2024).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: