The effect of high count rates on cardiac perfusion quantification in a simultaneous PET-MR system using a cardiac perfusion phantom

The availability of hybrid PET-MRI scanners gives the ability to perform simultaneous PET-MR cardiac studies, combining the functional perfusion quantitation from PET with excellent soft tissue contrast and high spatial and temporal resolution of cardiac MRI. Although the technique of simultaneous PET-MR myocardial perfusion imaging (MPI) studies is still in its infancy, multiple simultaneous image sets of structure and function of the same area of tissue may aid the clinician in the understanding of the pathophysiology of a wide range of cardiovascular diseases.

In order to provide reliable quantification of perfusion using PET, technical requirements should be considered, such as the presence of respiratory or motion artifacts [1, 2], analysis software for calculating perfusion [3, 4], prompt gamma correction [5] and image reconstruction [6]. A further subtle requirement is an upper limit on the injected activity in order to limit the count-rate losses experienced by the scanner due to all (or a high percentage) of the radiotracer being present in the field of view (FoV) of the scanner during the initial phases of imaging (where the arterial input function—AIF—is derived). This dead-time effect may adversely influence the reconstructed images and subsequently the calculated perfusion parameters resulting from PET kinetic modeling. Reducing excessive count rates and hence dead-time effects can be achieved either by a limit on the activity injected as a bolus or by performing a slow infusion of the radiotracer [7].

Previous work used decaying radiation source (13-ammonia and 82-rubidium) phantoms to identify the dynamic range of the PET system and also performed MPI in a cohort of 21 patients [8]. Their results proposed a weight-based administration approach (9 ± 1.5 MBq/kg) in order to limit the prompt count rates and dead-time losses (in turn the amount of activity) required for good image quality and reliable perfusion calculation.

Recently the group also performed decaying phantom experiments on 10 modern PET-CT scanners using either 1100 MBq of [¹³N]NH₃ or 3000 MBq of [⁸²Rb]Cl [9]. Their results identified a range of scanner-dependent performance metrics to avoid bias in quantification of activity concentration in reconstructed images, such as limiting of maximum singles count rates to between 22 and 64 Mcps and also recommended weight-based injected activities in the region of 3–14 MBq/kg. They further showed the clinical effect of using 2 different weight-based administrations of [⁸²Rb]Cl (2 and 10 MBq/kg, injections of 227 and 1022 MBq respectively) to the same patient, demonstrating underestimation of the peak blood and myocardium activities on the higher injected activity scan and also an overestimation of the quantified perfusion [9]. Other work has noted that although weight-based administrations may represent an oversimplification, it is a useful starting point in using phantom data to explore strategies to count-rate losses on clinical acquisitions [10].

Similar work has determined that detector blocks are less saturated with 1110 MBq infusion of [⁸²Rb]Cl rather than their previous standard protocol of 1480 MBq. They noted detector block saturation in 33/217 cases, although for infusions of 1110 MBq, 99% of scans presented without saturation effects [11]. Other work using [¹³N]NH₃ attempted to correct for underestimation in dead-time-affected input functions using phantom studies, noting an 8.9% overestimation of perfusion in 4 patients [12]. Recent attempts have been made to optimize injected activity to 11 MBq/kg (800 MBq) in [¹⁵O]H₂O studies; however, these studies focused on brain imaging where lower count rates are expected [13].

Previous studies characterized the performance of simultaneous PET-MR scanners at count rates specific to the NEMA performance test or during clinical FDG imaging [14–16]. Although MPI studies have been performed on PET-MR systems using relatively low injected activities (i.e. 400 MBq 15-oxygen [17], 214 MBq [¹³N]NH₃ [18], 396 ± 26 MBq [¹³N]NH₃ [19] and 259–296 MBq 18F-flurpiridaz [20]), there are as yet no detailed studies of the performance characteristics of these systems at count rates expected during MPI with traditional perfusion radiotracers. Use of these tracers introduces added considerations in PET-MRI imaging, as the field of view is typically larger and ring diameter typically smaller than traditional PET-CT systems. These factors lead to higher sensitivity and hence higher count rates presenting challenges on the preservation of count-rate linearity. Further potential considerations involve increased scatter due to the presence of MR receiving coils [21].

We aim to investigate the performance of a PET-MR scanner at high count rates encountered during MPI studies using a decaying-source experiment to examine count-rate effects on quantitative accuracy, and also to investigate count-rate effects on quantification of perfusion values using a cardiac perfusion phantom to experimentally simulate clinical MPI studies.

This work describes the effect of high count rates on the quantification of cardiac perfusion using standardized experimental simulations in an mMR PET-MR system. We have shown with a decaying-source experiment that above a total singles rate of approximately 64.9 Mcps (specific for our setup), data is subject to increasing levels of count loss. A distinct change is observed in the scanner response at 64.9 Mcps (Fig. 1). This point also corresponds to a change in recovered activity concentration, showing that at singles rates above 64.9 Mcps, quantification of images begin to be affected by count losses. For this phantom setup, a 5% reduction in NECR (NECR_95%) can be achieved with a singles rate of 34.1 Mcps and a 10% reduction in NECR with 24.7 Mcps. Figure 2 details that a 10% bias in image activity concentration is reached in the reconstructed images between singles rates of 78.2 and 82.9 Mcps; however, bias in the images can be avoided at singles rates lower than 60 Mcps. Response in the image activity concentration in PET-CT systems typically shows an exponential increase with decreasing activity rather than a sharp change at a single activity point as observed in this work, the cause of which may be related to the avalanche photo-diode detection system on the PET-MR system rather than photomultiplier tubes on traditional PET-CT systems. We opted to choose this water-filled torso phantom with cardiac insert in order to provide a more realistic geometrical setup for scatter and tracer distribution, rather than the standard NEMA phantom used during standardized performance testing of the system. Typical NEMA measurements do not include scanner performance as a result of a bolus of a high activity radiotracer.

Our value of the maximum singles rate from the decaying-source experiment is in line with that noted by Renaud et al. [9] of 64 Mcps in their survey of PET-CT systems. Of note is that their value was determined on a scanner with a similar FoV (22.1 cm) to the mMR system (25.8 cm). Activity in the field of view between 745 and 790 MBq leading to a bias of 10% is also comparable to their work for a similar PET scanner using a similar phantom setup.

Although this phantom is a clinically unrealistic scenario for the calculation of myocardial perfusion due to a lack of redistribution of the tracer, it provides the ability to measure metrics for evaluating the count-rate performance of PET systems under a semi-clinical distribution of a radiotracer and a range of count rates observed during a clinical acquisition. Although the reconstructed images demonstrate quantitation issues at singles rates above 64.9 Mcps (Fig. 2), the fixed defects could be clearly observed in the same image regions regardless of count rate and activity present in the source (Fig. 3). Thus, the main issues with cardiac imaging relate to aspects of quantification of the PET activity concentration and thus calculation of perfusion. Our results on maximum permissible count rates differed from those provided by Delso et al. [14], as the trues rate from our experiment was linear up to 1700 kcps, rather than 700 kcps. We also noted a higher NECR_max; however, of note is that our setup was not the standardized NEMA performance measurement as reported by Delso, and as such, our results are not directly comparable.

We used 11-carbon in the phantom rather than 15-oxygen or 13-ammonia primarily due to radiation safety aspects to the operator with filling the phantom, which would have required a much higher initial activity of a shorter-lived isotope. However, other tracers used in MPI would produce similar count rates due to similar positron branching ratios (11-carbon 99.8%, 15-oxygen 99.9%, 13-ammonia 99.8%, 82-rubidium 95%, 18-fluorine 96.9% [27]).

In addition to the decaying-source phantom, we utilized a perfusion phantom model to evaluate effects of high count rates in a semi-clinical setup. Previous work by our group has detailed a high level of repeatability using the phantom, showing consistent PET and MR perfusion calculations with a manually set perfusion rate [23]. In the work detailed here, MR perfusion was not calculated using a specific MR semi-quantification method to ensure that TAC and TIC data were analysed consistently using the same procedure. All factors (i.e. image acquisition, geometry, injection timings) for each scan were consistent. We have also previously demonstrated that PET quantification is unaffected by the presence of MR contrast agents at low concentrations [28].

Although we used MRAC for the perfusion phantom to mimic a clinical scenario, differences of 15% between the linear attenuation coefficients of MRAC and CTAC have been noted for simple water-filled phantoms [28], which would have a global scaling effect on all PET activity concentration curves.

Figure 4 details the rate of true events (i.e. events that contribute directly to the PET signal) and corresponding prompt rates for each level of activity injected into the phantom (A _INJ). The trues rates of the myocardial transit phase show the expected increases with higher activity used and are unaffected by count losses as indicated by the live-time fraction. It should be noted that as in clinical studies, the amount of activity in the field of view for perfusion phantom simulations changes rapidly over time, and therefor, of primary interest to our work is the peak activity during the input phase.

Activity concentrations from the aorta and myocardium compartments (C _A(PET) and C _M(PET) respectively) are shown in Fig. 5, and it can be noted that for C _A(PET) for A _INJ = 804 and 997 MBq, the curves show reduction of the peak height of response. Although net trues for A _INJ = 594 show effects of high count-rate losses in Fig. 4, C _A(PET), conversion to activity concentration in Fig. 5 shows that this may be to some extent recoverable or shows only a minor effect potentially as the activity injected (594 MBq) is close to the 570 MBq point identified from the decaying-source experiment. It should be noted that, owing to dispersion of the radiotracer, not all of the radiotracer injected is present in the scanner FOV simultaneously.

Calculation of the area under the curve (AUC) for each C _A(PET) and C _M(PET) shown in Fig. 6 demonstrates that C _M(PET) increases linearly over all tested activities (R ² = 0.99) while C _A(PET) is linear only up to A _INJ = 594 MBq with count-rate losses above this point affecting linearity. Calculated values of perfusion using the same methodology for PET and MR data (Table 1) also show that for A _INJ > 594 MBq, increases in K ₁ can be attributed to a decreased C _A(PET) and increased C _M(PET) as evident in Fig. 4. Consistent values for K ₁ from the MR curves show that good repeatability can be obtained by use of the phantom. Direct comparison of PET and MR K ₁ values should be avoided, as MR values were calculated using a PET kinetic model from the prebolus peak and using signal intensity values as PET activity concentration.

It should be noted that the scattering scenarios between the perfusion phantom and the decaying-source phantom are different, given that the perfusion phantom is not surrounded by water. Therefor, the maximum trues rates between the setups will be different based on the lower number of scatter and randoms in the decaying-source experiment. However, even with this difference of phantom environments, insights can be gained into a limiting value of count rate for accurate quantification. Further limitations of the phantom model compared to a clinical situation are that the phantom represents an oversimplification of the cardiovascular system and cannot detail true myocardial diffusion or intracellular uptake of PET radiotracers. It is also unable to reproduce the multiple sources of image artifacts in clinical PET-MR (such as respiratory or cardiac motion). Despite these seemingly big limitations, the setup provides an environment free from these potentially confounding effects, allowing analysis to focus only on the assessment of the perfusion dynamics within the cardiac compartments, and the potential to repeatedly examine count-rate capability for determination of perfusion accuracy. Furthermore, it allows technical development of imaging techniques related to myocardial perfusion without the need to expose patients to ionizing radiation or contrast agents. In this respect, the standardized and repeatable setup is a considerable advantage over routine clinical studies for the evaluation of the count-rate losses due to the effects of high activity in the FOV.

Count rates in clinical studies are likely to be lower than those described in the perfusion phantom experiments due to the different scattering geometry of a patient, dispersion of radiotracer, and also the time taken to transit the pulmonary circulation and reach the left ventricle before being measured as part of the input function. Therefor, the singles rates provided by this work may represent an upper maximum of clinically observed count rates and will be compared to clinical studies in the future. Refinement to the acquisition and list-mode rebinning protocol should be investigated using the perfusion phantom (and in patients) using methods such as that described by Kolthammer et al. to optimize the frame time in relation to variables such as the infusion duration, scan time and total activity [29]. For simplicity, we used a constant time frame over the entire study, in a similar fashion to previous work [9].

Our findings indicate that singles rates above 64.9 Mcps will affect the quantitation of resulting PET images, potentially impacting the arterial input functions and thus calculation of perfusion. However, patient studies should be performed in order to compare the count rates observed on the phantom to those found during clinical acquisitions and also to examine the effects of lower count rates on image quality.

Our work on a decaying-source phantom has identified that a singles rate above 64.9 Mcps during the first pass of the radiotracer on the mMR system affects the quantitation of image-based arterial input functions, and a singles rate between 78.2 and 82.9 Mcps leads to a 10% bias in recovered image activity concentration. NECR_max was 286 kcps at 47.1 Mcps, and NECR_95% was observed from a singles rate of 34.1 Mcps up to 60.1 Mcps. Our cardiac perfusion phantom with a perfusion rate of 3 ml/min/g showed count losses with increasing singles rates above 64.9 Mcps, corresponding to an injected activity to the phantom of 600 MBq and higher. Above this count rate, at least 55% of the events were lost and quantification of perfusion was severely affected. Quantification of perfusion values with singles rates above 64.9 Mcps on this system may be compromised and should be avoided where possible. Validation studies in large animals could be performed to verify the results of the phantom experiments, as well as to investigate any potential effects on image quality.