Assessment of a digital and an analog PET/CT system for accurate myocardial perfusion imaging with a flow phantom

Abstract In Myocardial Perfusion Imaging (MPI) with Positron Emission Tomography/Computed Tomography (PET/CT) systems, accurate quantification is essential. We assessed flow quantification accuracy over various injected activities using a flow phantom. Methods The study was performed on the digital 4-ring Discovery MI (DMI-20) and analog Discovery 690 (D690) PET/CT systems, using 325-1257 MBq of [15O]H2O. PET performance and flow quantification accuracy were assessed in terms of count-rates, dead-time factors (DTF), scatter fractions (SF), time-activity curves (TACs), areas-under-the-curves (AUCs) and flow values. Results On DMI-20, prompts of 12.8 Mcps, DTF of 2.06 and SF of 46.1% were measured with 1257 MBq of activity. On the D690, prompts of 6.85 Mcps, DTF of 1.57 and SF of 32.5% were measured with 1230 MBq of activity. AUC values were linear over all activities. Mean wash-in flow error was − 9% for both systems whereas wash-out flow error was − 5% and − 6% for DMI-20 and D690. With the highest activity, wash-out flow error was − 12% and − 7% for the DMI-20 and D690. Conclusion DMI-20 and D690 preserved accurate flow quantification over all injected activities, with maximum error of − 12%. In the future, flow quantification accuracy over the activities and count-rates evaluated in this study should be assessed. Supplementary Information The online version of this article (10.1007/s12350-021-02631-9) contains supplementary material, which is available to authorized users.


INTRODUCTION
Myocardial Perfusion Imaging (MPI) with dynamic Positron Emission Tomography (PET) allows quantitative measurements of absolute Myocardial Blood Flow (MBF). 1 MPI assessed using PET and short-lived tracers is highly reproducible. [2][3][4][5] However, physical performance of PET systems has been a key issue in achieving accurate flow quantification over a wide range of activities. [6][7][8][9] PET systems are exposed to high countrates and dead-time when activity is present in the injected vein and heart chamber, which may compromise image accuracy and quality. 10 Due to dead-time effects, time-activity curves (TACs) from left ventricle blood pool and myocardial tissue may be underestimated, resulting in MBF quantification inaccuracy. 2,[11][12][13] With the introduction of digital PET/Computed Tomography (PET/CT) systems, improvements in count-rate performance and image quality have been shown with National Electrical Manufacturers Association (NEMA) tests. 14,15 Therefore, there might be potential advantages in applying digital PET/CT systems in MPI quantification. Thus, the count-rate capabilities of digital systems need to be assessed with phantom studies at high count-rates representative to MPI measurements. Evaluation of flow quantification at various injected activities is of paramount importance to understand the effects of count-rate performance, which differs among PET systems. 6,7,9,16 Most importantly, flow quantification accuracy of digital PET/CT systems over various count-rates should be evaluated not only on static phantoms but with phantoms simulating dynamic imaging conditions. In 16 the authors investigated perfusion quantification accuracy at various activities by using a PET/MR compatible flow phantom. A similar phantom modelling realistic cardiac flow was presented and validated for MPI in PET in Reference 17 , allowing flow modelling and quantification against a ground truth value. Thus, PET system assessment using a flow phantom allows to evaluate both count-rate performance and flow quantification accuracy.
In this study, we aimed at assessing flow quantification accuracy of a novel digital PET/CT system in comparison to an analog PET/CT system. A state-of-theart flow phantom was selected to evaluate the accuracy of flow quantification for various activities up to 1260 MBq of [ 15 O]H 2 O at scan start time.

PET/CT Systems
The digital Discovery MI with four detector rings (DMI-20) 15 and analog Discovery 690 (D690) (GE Healthcare, US) 18 PET/CT systems were assessed. Performance characteristics are summarized in Supplemental Table 1.

Flow Phantom System
To investigate flow quantification accuracy, a flow phantom (DCE Dynamic Flow Phantom, Shelley Medical Imaging Technologies, Canada) was used. Phantom validation for MPI in PET is presented in Reference 17 .
The phantom set-up contains water container, peristaltic pump, injection port, phantom shell, flow constrictor valves and flow meters. The phantom schematic is shown in Supplemental Figure.  Flow inside the system is controlled with the peristaltic pump Qpump ð Þ. Flow inside the exchange cylinder (Qcyl) and perforated tube (Qtube) are controlled with flow constrictor valves and recorded with microturbine flow meters (Omega Engineering Inc., US). 24 Ideally, Qtube and Qcyl should be equal to Qpump. Reference flow (Qref ) values are related to Qcyl and derived from flow meter calibration using a lookup table. Further information from the flow values is given in Supplemental Figure. 1.
Kinetic Modeling of Flow Values The input and tissue TACs can be measured by a volume-ofinterest (VOI) analysis, defined for the input chamber and exchange cylinder, namely C inputVOI t ð Þ and C cylVOI t ð Þ. VOIs were selected as in Reference 5 . The TACs are used for image-based flow (Qin; Qout) modelling based on a two-compartment kinetic (one-tissue compartment) model as implemented in the phantom flow quantification software (QuantifyDCE 1.1, Shelley Medical Imaging Technologies, Canada). 17 The software models tracer concentration in the exchange cylinder as: where Ã represents discrete convolution operation, C cyl t ð Þ the time-dependent tracer concentration in the exchange cylinder, q in the tracer wash-in rate to the exchange cylinder (min -1 ), and q out the tracer wash-out rate from the exchange cylinder (min -1 ). C tube t ð Þ represents tracer concentration in the perforated tube, which can be estimated from the input chamber TAC by transport delay of C tube t ð Þ ¼ C inputVOI t À delay ð Þ ; when t [ delay: Moreover, as C cylVOI t ð Þ contains contributions from both C cyl t ð Þ and C tube t ð Þ, a signal mixing correction function is implemented as: Thus, complete kinetic model consists of following free parameters:q in ; q out , ISF and delay. Model fitting is performed with a weighted (by frame length) least squares algorithm.
The modeled parameters q in and q out can be considered analogous to rate parameters K 1 and k 2 in kinetic modelling of blood-tissue exchange. Final image-derived flow values Qin and Qout can be computed based on q in and q out , multiplied with the cylinder volume (V cyl ¼ 161mL), representing flow from the perforated tube to the exchange cylinder (Q in ¼ V cyl Â q in ) and flow out from the exchange cylinder (Q out ¼ V cyl Â q out ). Ideally, these flow values should be equal to the reference flow value, i.e. Qin ¼ Qout ¼ Qref (mL/min). 17

PET and CT Data Acquisition
At the start of the measurement, the flow meters, peristaltic pump (MasterFlex L/S, 07522-20, US) and [ 15 O]H 2 O dispenser (Hidex Oy, Finland) were calibrated. Flow meter calibration and flow rates were fixed for all studies. Qpump was set to 200 mL/min, while Qcyl was adjusted to 60% of Qpump, following previous protocols. 5,17  A CT-based attenuation correction (CTAC) using a tube voltage of 120 kV and current of 67mAs was acquired. The flow phantom was imaged in the center of the PET FOV. Dynamic list-mode PET acquisition was started 50 s after the dose injection, with a duration of 4 min and 40 s, following the clinical protocol. 25 Measurements were repeated 9 times with individual doses. Injected doses were measured automatically by the [ 15 O]H 2 O dispenser (Hidex Oy, Finland), which is cross-calibrated to a Veenstra VDC-404 (Veenstra Instruments, Netherlands) dose calibrator. Activities at scan start times and flow rates are reported in Table 1.

Flow Quantification Accuracy
Count-rate performance was assessed for each measurement by extracting count-rate data from DICOM headers. Specifically, peak prompts (P), randoms (R), dead-time factors (DTF) and scatter fractions (SF) corresponding the peak prompts were extracted.
Scatter rates (S) and trues rates (T) were derived based on the extracted P, R and SF. The scatter rate was derived as: and the trues rate was derived as: We report the peak count-rates (P, T, R, S), DTF and SF as function of activity at scan start time.
Flow quantification accuracy was assessed in terms of shape of the input and tissue TACs and area-underthe-curve (AUC) of the TACs. Linearity of AUCs was determined with regression analysis of the AUC values versus activities at scan start time. We report AUCs and goodness-of-fits (R 2 ) of the regression slopes.
Finally, accuracy of the modeled flow values Q in and Q out with respect to the reference flow Q ref was calculated as relative error: where flow value represents either Q in or Q out for each measurement. We report the accuracy of flow values with activity at scan start time. All flow values are reported as mean ± standard deviation (SD). Figure 1 shows peak prompts, scatter, randoms and trues as a function of activity at scan start time on DMI-20 and D690. On both systems, count-rates increased in relation to activities, as expected. For DMI-20, peak prompts were 12.8 Mcps with 1257 MBq. On D690, peak prompts were 6.85 Mcps with 1230 MBq. Additional data of the DMI prompts and randoms for the measurement with the highest injected activity (1257 MBq) is given in Supplementary Data 1.

RESULTS
DTFs as a function of activity at scan start time are presented in Figure 2A. DTFs increased nearly linearly in relation to activities, reflecting the count-rates (Figure 1). DTFs were higher with DMI-20 compared to D690. The highest DTFs 2.06 and 1.57 were measured with 1257 MBq and 1230 MBq on DMI-20 and D690. Figure 2B shows SFs as a function of activity at scan start time. Due to larger axial FOV, SFs were higher on DMI-20 compared to D690. SFs increased up to the three highest activities (906 MBq, 1060 MBq, 1257 MBq at scan start time), after which SFs were nearly constant on DMI-20. On D690, SFs increased in relation to all activities. Figure 3 shows TACs derived from the input chamber and exchange cylinder of the flow phantom. No visible distortions in TAC shapes were seen. Twocompartmental model fitting R 2 values were 0.998 and 0.999 on both DMI-20 and D690.
AUCs as a function of activity at scan start time with regression analysis are presented in Figure 4. R 2 values for the input and tissue AUCs were 0.995 and 0.997 for DMI-20 and 0.998 and 0.998 for D690. AUCs increased linearly in relation to all activities on both systems. Tissue AUCs followed the line-of-identity closely whereas the input AUCs were above the identity line on both systems.
Prompts, DTFs and SFs with the accuracy of the modeled flow values are presented in Supplemental Table 2. Errors of the modeled flow values with respect to Q ref are presented as a function of activity at scan start time in Figure 5. A slight increase in Q out error was seen above activities of 906 MBq and 995 MBq at scan start time on DMI-20 and D690. In general, the error in flow quantification accuracy was small over all count-rates and activities. Also, the error magnitude of Q in and Q out can be considered small (below 5%) between DMI-20 and D690.

DISCUSSION
We assessed the accuracy of flow quantification on digital (DMI-20) and analog (D690) PET/CT systems over various injected activities with a flow phantom, for the first time.
DMI-20 showed improved performance with higher prompts compared to D690 over similar activities (Figure 1), in line with 9 and NEMA performance studies. 15 DMI-20 limits the amount of data passed to the acquisition sorter after reaching a throttle limit, due to its sensitivity. The throttle is specific for different ring configurations on DMI systems, limiting the amount of data transmitted for histogramming. The number of prompts in the DICOM header will be also capped at the scanner-specific limit due to the throttle. The DMI-20 specific 12.8 Mcps limit was reached at activity of 1257 MBq. Therefore, a sudden notch in scatter and trues rate in Figure 1A and B is seen, similarly as in Reference 16  However, during reconstruction, a scaling factor based on the prompts seen by the coincidence processor and prompts transmitted to histogramming will be applied to preserve quantification despite the throttle. Moreover, the actual RFS estimate will be slightly lower than the value for randoms in the DICOM header above throttle, as no count losses are taken into account for the randoms value saved in the DICOM header. The interested reader is referred to Supplementary Data 1 for more information.
Higher DTFs and SFs ( Figure 2) were seen on DMI-20 over all activities. SFs increased in relation to activities, stabilizing at the highest activities (Figure 2B). This might be caused by biased scatter scaling at high activities by counts outside the body contour or by increased detector scatter, which is usually seen in NEMA testing. 15 However, we detected no CTAC-PET misalignment that could affect scatter scaling. In general, no artifacts were seen resulting from scatter overcorrection on either system.
For flow quantification, TACs showed no visible artifacts (Figure 3). Similarly, the input and tissue AUCs showed high linearity (R 2 close to 0.99) and followed the line-of-identity well (Figure 4). Saturation of TACs  13 have been shown previously. Saturation or non-linearity of AUCs were not seen on either system, contrary to 16 and. 13 However, the input AUCs were slightly higher compared to tissue AUCs, which can contribute to the difference seen between Q in and Q out values ( Figure 5), as we have previously discussed in. 5 In line with the AUC linearity, errors in the modeled flow values were similar over all activities and systems ( Figure 5). Mean Q in error was -9% and mean Q out error was -5% and -6% on DMI-20 and D690. The small difference between systems is contributed by different reconstruction parameters, which were selected based on clinical protocols on each system. Q out error increased when activity increased, up to -12% on DMI-20 at the highest activity, and up to 7% on D690. Our finding is similar to 13 measured with [ 82 Rb], although the maximum error in Q out (-12%) was smaller compared to the reported error of 22%. Similarly to Reference 5 , there was a systematic offset between in Q in vs Q out (Supplementary Table 1) and offset of 10 mL/min in Qcyl and Qtube (Table 1), which results in systematic negative bias irrespective of count-rate performance or the PET systems.
As a limitation, flow quantification was assessed only up to 1257 MBq at scan start time due to reaching the maximum limit of our [ 15  O can be achieved if PET is acquired under the activities and count-rates specified above, as in References 6 , 7 , 9 , and 16 . The advantage of using a flow phantom is two-fold: the reference flow is always known, and CT-PET data can be collected free of any motion artefacts or misalignment. Thus, the phantom protocol used in this study could be applied for MPI harmonization studies for several PET/CT systems, according to their count-rate performance and reconstruction methods. error was -9% for wash-in flow and -5% and -6% for wash-out flow on DMI-20 and D690, indicating similar flow quantification errors on both systems. As flow quantification error was -12% at 12.8 Mcps prompts on the DMI and -7% at 6.85 Mcps prompts on the D690, we assume DMI-20 system might offer potential benefits in MPI in PET.

Disclosures
The authors declare no conflict of interest.

Funding
Open access funding provided by University of Turku (UTU) including Turku University Central Hospital.
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 http://creativecommo ns.org/licenses/by/4.0/.