A novel dynamic cardiac phantom for comparing blood pool and myocardial perfusion tracers: initial evaluation with CZT-ERNA and PET-MPI

Abstract

Background

Nuclear imaging modalities can quantify cardiac volumes and left ventricular ejection fraction (LVEF), but systematic differences between techniques complicate direct comparison. Phantom-based evaluations comparing myocardial perfusion and blood-pool methods are scarce.

Methods

We used a dynamic two-compartment cardiac phantom (myocardium and left ventricular [LV] lumen) with adjustable stroke volume and heart rate. Static and gated acquisitions were performed on cadmium–zinc–telluride equilibrium radionuclilde angiocardiography (CZT-ERNA) and gated positron emission tomography myocardial perfusion imaging (PET-MPI) under identical volumes and heart-rate settings. Computed tomography (CT) served as the reference for end-diastolic volume (EDV) and end-systolic volume (ESV). LVEF was derived from these volumes for all modalities. We also assessed the impact of phantom position relative to detectors, as well as the effect of attenuation correction via a thorax phantom.

Results

CZT-ERNA overestimated static LV volumes relative to CT, whereas PET-MPI underestimated them. Both modalities underestimated LVEF compared with CT-derived values, with a larger bias for PET-MPI. On PET-MPI, attenuation correction produced similar volumes with and without the thorax phantom. CZT-ERNA volumes decreased as detector distance increased.

Conclusion

Under controlled phantom conditions, CZT-ERNA and PET-MPI show inverse biases for LV volumes and both underestimated LVEF versus CT. These findings point to the influence of reconstruction and contouring algorithms, attenuation correction, and camera positioning. The dynamic phantom enables direct comparison of perfusion- and blood-pool-based measurements and may support modality-specific calibration when translating absolute values across techniques.

Introduction

Cardiovascular imaging plays an essential role in assessing cardiac function, particularly left ventricular ejection fraction (LVEF), in patients at risk of declining pump function during cancer treatment.

According to the 2021 European Society of Cardiology guidelines for heart failure, 3D-echocardiography (3DE) or cardiac magnetic resonance (CMR) are recommended as first-line modalities. Multigated acquisition (MUGA), also known as equilibrium radionuclide angiocardiography (ERNA), serves as a third-line option when 3DE and CMR are unavailable or inconclusive [1].

Cadmium–zinc–telluride ERNA (CZT-ERNA), a type of 3D-MUGA using cadmium-zinc-telluride detectors, is an efficient tool for serial LVEF monitoring [2]. In patients, CZT-ERNA shows good correlation with CMR for assessing both LVEF and right ventricular ejection fraction (RVEF) [3, 4]. However, LVEF measured by CZT-ERNA correlates less strongly with 2D-echocardiography [5].

Cardiac phantom studies are valuable for evaluating and ensuring quality control in nuclear imaging techniques. For instance, myocardial perfusion imaging (MPI) with CZT-detectors or conventional Anger cameras has been tested for LVEF assessment. These studies highlight that CZT-MPI provides superior spatial resolution, sensitivity, and shorter acquisition times [6]. Yet, conflicting results include larger left ventricular (LV) volumes, especially end-systolic volume (ESV) [7], and decreased LVEF compared to Anger cameras [8]. To our knowledge, most cardiac phantoms estimate LVEF using MPI. Only one study has used blood-pool imaging with a static phantom to compare positron emission tomography-computed tomography (PET-CT), single-photon emission computed tomography-ERNA (SPECT-ERNA), and CMR [9]. No prior work has directly compared CZT-ERNA and PET-MPI.

The aim of this study is to address this gap by comparing CZT-ERNA and PET-MPI for the assessment of LVEF using both static and gated phantom acquisitions.

Background

Cardiac phantoms simulate the anatomy and/or function of the human heart and can be generally described as static or dynamic. Static cardiac phantoms are fixed models representing specific phases of the cardiac cycle, typically end-systole and end-diastole. They offer greater anatomical accuracy but lack the ability to assess the effect of motion. Dynamic cardiac phantoms replicate heart motion at varying volumes and heart rates to simulate the beating heart. Incorporating thorax phantoms to mimic soft tissue attenuation further enhances these studies. Cardiac phantoms are widely used in radiology to compare imaging modalities such as CMR, computed tomography (CT), and 3DE [10, 11].

Advancements in 3D printing have facilitated the development of anthropomorphic cardiac phantoms that closely resemble human anatomy. For example, Kiss et al. [12] created two phantoms from patient-derived PET data: one depicting healthy myocardium and another with a perfusion defect. These designs are cost-effective, quick to produce, and customizable, including features such as 'hot' backgrounds and liver inserts. Similarly, Verrecchia-Ramos et al. [9] built static phantom pairs from CMR scans at end-diastole and end-systole to validate cardiac volumes and LVEF across multiple imaging modalities. Despite their anatomical precision, static models limit assessments of cardiac volumes and LVEF by failing to include effects of cardiac motion.

Dynamic cardiac phantoms such as the Amsterdam-gated cardiac phantom (AGATE) support validation of gated MPI SPECT. AGATE allows variable LV volumes with realistic ejection fractions but is specifically designed for MPI, rendering it unsuitable for gated blood pool imaging [13]. AGATE has been applied in comparisons of CZT cameras under dual-isotope conditions [7], evaluation of LVEF from CZT cameras against conventional gamma cameras [8], and multicenter studies examining variations in LV function [14].

De Bondt et al. [15, 16] designed dynamic cardiac phantoms explicitly for gated blood pool imaging, featuring biventricular and four-chamber models with valves to evaluate LV and right ventricular (RV) function, though they do not support MPI.

Currently, no phantom enables direct comparison between MPI and blood-pool imaging for cardiac volumes and LVEF. Our custom-built phantom fills this void. In this study, we showcase its utility by comparing volumes and LVEF from CZT-ERNA and gated PET-MPI, using CT as the reference standard.

Methods

Gated heart phantom

The left ventricle is simulated using two silicone balloons. The outer lumen represents the myocardium, while the inner lumen simulates the cavity of the left ventricle. Fluid is continually pumped in and out using a pneumatic system driven by pressurized air, supplied through compressed air outlets commonly found in hospitals.

The pneumatic pump consists of a piston housed within a cylindrical chamber. Pressurized air is alternately supplied to the top and bottom of the chamber of the pump, driving piston movement. Its velocity is regulated by two adjustable outlet valves positioned at either end of the chamber, emulating a cardiac cycle, characterized by a rapid systolic phase and a slower diastolic phase.

Two reed switches are mounted to the pump housing and are magnetically activated as the piston approaches them. The switches control an electromagnetic solenoid valve, switching air supply between the top or bottom of the pump. The spatial distance between the reed switches defines the piston stroke length, thereby determining the simulated cardiac stroke volume.

Piston displacement is monitored by a microswitch, which generates a standardized 5 V trigger signal, simulating an electrocardiogram (ECG) trigger. This trigger signal is transmitted to the imaging system to synchronize data acquisition with the mechanical cycle.

The air pump piston operates a modified 100 mL plastic syringe (Pic Solution). The syringe has ports specifically for filling the inner lumen of the system, while the flange has ports for filling the outer lumen. The syringe is connected via a rubber tube to the flange, on which the balloons are attached with O-rings. Activation of the system moves fluid back and forth from the innermost balloon to simulate the beating heart. The complete mobile phantom system is arranged on a rolling table (Fig. 1).

Fig. 1.

The phantom system is arranged on a rolling table. Top shelf: The pneumatic pump piston (right side) is supplied by two blue pressurized air tubes. The piston drives a modified 100 mL syringe connected via rubber tubing to a flange (front left). The brown circuit board provides the 5 V trigger signal based upon piston movement simulating an ECG. Middle shelf: Examples of the silicone balloons used to simulate the inner ventricular cavity and outer myocardial wall are placed here (not attached to the flange). Bottom shelf: The thorax phantom is positioned on the left, and the power supply converting 240 V AC to 5 V DC is located on the right

Thorax phantom

A thorax phantom mimicking the attenuation of human tissue for both 140 kV and CT energies was produced using the filament deposition modeling (FDM) print method, using a Prusa MK3s 3D printer and PBT + (Polybutylene Terephthalate) filament. The filament type was selected for having similar photon attenuation properties as human tissue.

The phantom consists of two types of tissue, lung tissue and ribcage tissue, the latter representing both bone and soft tissue. The shape of the tissue types was segmented on a single lung CT slice and extrapolated in a straight line along the height of the phantom.

A cavity for the dynamic heart phantom was then cut from the resulting phantom, allowing for uniform positioning of the heart within the thorax phantom, with a locking plug to fix the dynamic heart phantom in place.

To achieve similar attenuation characteristics, sample CT scans were used and an average HU value for lung (-642 HU) and ribcage tissue (10 HU) was calculated.

By modulating the ratio of PBT + plastic filament to air in the 3D printed phantom it was possible to achieve average attenuation characteristics of the two segments in the phantom that matched the average HU values of human tissue, using 31.5% infill for lung tissue and 92.0% infill for ribcage tissue. For data acquisition, the combined phantom setup is positioned inside the respective CZT and PET/CT scanners (Fig. 2).

Fig. 2.

Experimental setup of the dynamic cardiac phantom used for cross-modality imaging. a The thorax phantom with dynamic cardiac insert positioned inside a cardiac-dedicated CZT SPECT camera. Motion is driven by a pneumatic system placed externally on a support table. The black tubing at the lower left is the pressurized air intake supplying the actuation system. The small brown circuit board carries the gating electrode connections, where the trigger electrodes from the camera (red, green, and white) are attached. b The same phantom positioned inside a PET/CT scanner. The thorax phantom is centered in the bore, with the pneumatic actuation system extending outside. In the foreground, the external ECG-trigger device with electrodes attached is visible, used to deliver gating signals for static acquisitions

Volume estimation

The reference volume of the inner lumen was measured on CT-images using MIM Software (Cleveland, OH, USA) (Fig. 3). The outer lumen has a higher average HU than the inner due to the addition of iodized table-salt, this allowed us to contour the inner-lumen using the tool Region Grow. This tool allows the user to select an appropriate area that the software should contour depending on HU. By manually setting a boundary ROI at the edge of the flange, we can avoid the contour spilling into the feeding port of the inner balloon. The registered HU of the lumens varied slightly from scan to scan. To ensure correct contouring the HU thresholds for the Region Regrow were adjusted until a satisfactory contouring was achieved.

Fig. 3.

Volume estimation of the dynamic cardiac phantom using MIM software. A region-grow tool based on Hounsfield unit (HU) thresholds was applied to segment the three-dimensional volume of interest. The basal flange was used as the anatomical limit of the segmentation. The resulting contour is shown in pink. Within the phantom, three distinct layers can be identified: the inner lumen, the silicon balloon, and the outer lumen. A visible HU gradient is present between these compartments. The contrast between the inner and outer lumens arises from the addition of iodized table salt to the distilled water in the outer lumen, increasing its attenuation relative to the inner compartment. and providing a closer match between the difference in density between of blood (1.06 g/mL) and muscle (1.10 g/mL)

To validate CT as the reference, we first determined the absolute lumen volume by disassembling the phantom, weighing the components, and calculating the water-filled volume gravimetrically. This approach is limited by meniscus reading uncertainty, but the resulting fixed bias relative to CT was < 2 mL. Proportional bias was assessed post-hoc by comparing recorded syringe volume changes to corresponding CT-measured changes across all fills; no systematic bias was observed. The small residual error (< 2 mL) is well below the variability relevant for cardiac volume quantification.

PET/CT

Our study utilized a Siemens Biograph mCT Flow PET/CT system (Siemens Healthcare, Erlangen, Germany). The system consists of a PET detector with 4 rings, with each ring containing 48 detector blocks. Each block is equipped with lutetium oxyorthosilicate (LSO) crystals of 4 × 4 × 20 mm, assembled in a 13 × 13 array and coupled to a 2 × 2 PMT array in each detector block. The system has an axial field of view (FOV) of 20 cm, and a sensitivity of 10 cps/kBq. The detector ring diameter is 84.2 cm. The time coincidence window is 4.1 ns, with a system time resolution of 540 ps and an energy window of 435–650 keV. This model features a bore diameter of 78 cm, a tunnel length of 136 cm, and a table capacity of 227 kg.

An integrated 64-slice CT is used for attenuation correction of PET data. PET images were reconstructed using a Point Spread Function and Time-of-Flight (PSF + TOF) algorithm with 3 iterations and 21 subsets. A Gaussian post-reconstruction filter of 10.0 mm was applied. The images were reconstructed into a 256 × 256 matrix with a pixel spacing of 1.59 mm and a slice thickness of 2.0 mm. Standard corrections included CT-based attenuation correction, model-based scatter correction, and delayed window randoms subtraction. All images were decay-corrected to the start of the scan. The acquired PET images were analyzed with Corridor4DM (INVIA, Ann Arbor, MI, USA).

CZT-ERNA

The CZT-ERNA acquisitions were performed on the GE Discovery NM530c (GE Healthcare, Milwaukee, WI) equipped with a stationary multi-pinhole collimator. The acquisition protocol used 16 bins per R-R interval, set to acquire a predefined number of accepted beats (200), and a 20% energy window centered on 140 keV. Images were reconstructed using the manufacturer’s dedicated 3D iterative Bayesian reconstruction algorithm (50 iterations, 2 subsets) into a 70 × 70 matrix with a voxel size of 4.0 × 4.0 × 4.0 mm with a post-reconstruction Butterworth filter (cutoff 0.40 cycles/cm, power 7) as per the manufacturer’s standard clinical protocol. For the analysis of the acquired gated images we used the Xeleris 3 Imaging workstation’s reorientation software (GE Healthcare, Milwaukee, WI, version no. 3.0562) and Cedars-Sinai QBS processing software (Cedars-Sinai, Los Angeles, CA, revision 2009.0).

For static imaging, an ECG signal generator was connected to the scanner to simulate a trigger signal, as our acquisition protocol is designed for gated acquisitions. This produced end-systolic and end-diastolic volumes in the QBS software, and the mean of these values was reported for static acquisitions.

Study protocol

The study consisted of three series, each conducted over two days. Data acquisition was performed in two phases. Series 1 and 2 served as an initial evaluation of the phantom setup. Based on findings from these series, specifically the impact of compartment contact, the phantom configuration was optimized for the final acquisitions in Series 3.

Preparation for each series included filling the inner balloon with distilled water and a 3.5% iodized salt solution for the outer lumen. This setup ensured a clear difference in Hounsfield Units (HU) between the two compartments. After filling, the phantom was set to pump, allowing the balloons to reach equilibrium.

On the first day of each series, a PET-isotope was injected into the outer lumen of the phantom. Stepwise static acquisitions were then performed by adjusting the volume in the feeding syringe. After completing the stepwise acquisitions, the phantom was set to pump, and gated sequences were acquired. A CT scan was performed at each step including the end points of the pumping sequences.

After the PET acquisitions, the phantom was placed in a safe location for tracer decay. On the following day, Tc99m was administered to the inner lumen for CZT-ERNA acquisitions, and the process was repeated with the corresponding static and gated acquisitions. As our CZT-camera is not equipped with a CT-scanner the subsequent CT scans were performed on the same scanner used for the PET acquisitions to ensure consistent volume estimation.

Several static CZT-ERNA acquisitions were performed with increasing detector-to-patient distances to simulate progressively suboptimal camera positioning. The effect of scan length on measured volumes was also assessed for both CZT-ERNA and PET.

Results

For Series 3, we measured a mean activity concentration for PET-MPI of 196 kBq/ml during static acquisitions and 257 kBq/ml during gated acquisitions. For CZT-ERNA, we measured a range between 584,102 and 1,101,736 counts from the smallest to largest static measurement, while the gated acquisitions on average recorded 1,200,000 counts.

Optimal camera positioning resulted in the largest CZT-ERNA-derived volumes, while a displacement of more than 9 cm anteriorly or laterally led to significant underestimation (Table 1).

Table 1.

Series description

Series	Tracer	Comments
Initial evaluation (Series 1)	175 MBq NaF 40 MBq Tc-99m	Close proximity of inner and outer lumen
Initial evaluation (Series 2)	70 MBq NaF/25 MBq Tc-99m	Collision of inner and outer lumen
Optimized protocol (Series 3)	115 MBq NaF/25 MBq Tc-99m	No collision

Discussion

This study compared CZT-ERNA and PET-MPI with CT-derived volumes using a dynamic cardiac phantom. The main findings were as follows. PET volumes tracked below CT with increasing size (Fig. 4). CZT-ERNA volumes drifted above CT (Fig. 5). Both modalities underestimated LVEF, with a larger bias for PET (Fig. 6; Table 2). Attenuation correction stabilized PET-derived volumes and reduced dependence on the thorax phantom (Fig. 7). CZT-ERNA was sensitive to thorax phantom presence and camera positioning (Fig. 8; Table 3).

Fig. 4.

Relationship between CT-derived and PET-derived volumes for static PET acquisitions. Data are shown for three acquisition series: Series 1 (green circles), Series 2 (blue–green squares), and Series 3 (purple triangles). Symbols are filled with series-specific colors and outlined in black; solid regression lines are shown for each series, with the dashed line indicating unity. Across all series, PET volumes demonstrated a consistent downward drift relative to CT volumes with increasing ventricular size

Fig. 5.

Relationship between CZT-ERNA static measurements and CT-derived volumes. Data are shown for three acquisition series: Series 1 (green circles), Series 2 (blue–green squares), and Series 3 (purple triangles). Symbols are filled with series-specific colors and outlined in black; solid regression lines are shown for each series, with the dashed line indicating unity. All three series showed an upward drift in CZT-ERNA volumes relative to CT volumes with increasing phantom volume

Fig. 6.

Paired line plots comparing CT-derived and modality-derived LVEF values for PET (left) and CZT-ERNA (right) acquisitions. Data are shown for three acquisition series: Series 1 (green circles), Series 2 (blue–green squares), and Series 3 (purple triangles). All gated PET measurements underestimated LVEF compared to CT-derived values, with the steepest declines between CT and PET observed in Series 1 and 2. A similar pattern was seen for CZT-ERNA, though the underestimation was generally less pronounced, and in some cases values were close to CT-derived LVEF, as indicated by shallower connecting lines

Table 2.

Quantitative comparison of modalities versus CT reference

Acquisition	N	Regression (Slope, Int, R2)	Mean bias	95% LoA
PET-MPI
Static Volumes, Series 3	6	0.72, 10.46, 1.00	− 7.0 mL	± 8.4 mL
Static Volumes, Series 1–2	14	0.69, 10.69, 0.99	− 6.8 mL	± 7.7 mL
Gated LVEF, Series 3	3	0.86, − 5.75, 0.99	− 11.4%	± 2.7%
Gated LVEF, Series 1–2	7	0.60, 2.64, 0.91	− 18.2%	± 5.9%
CZT− ERNA
Static Volumes, Series 3	5	1.21, − 14.63, 1.00	− 2.4 mL	± 6.4 mL
Static Volumes, Series 1–2	14	1.23, − 6.60, 0.99	+ 6.1 mL	± 7.1 mL
Gated LVEF, Series 3	4	0.92, − 0.51, 0.94	− 3.9%	± 3.4%
Gated LVEF, Series 1–2	7	0.54, 15.43, 0.96	− 5.2%	± 8.9%

Fig. 7.

Relationship between CT-derived and PET-derived volumes for static PET acquisitions performed with (green circles) and without (purple squares) the thorax phantom. Separate panels show non-attenuation corrected (left) and attenuation corrected (right) reconstructions. While following the same pattern observed for static PET in general, attenuation-corrected volumes were similar regardless of phantom presence. In the absence of attenuation correction, volumes measured without the thorax phantom were larger than those measured with the phantom

Fig. 8.

Relationship between CZT-ERNA static measurements and CT-derived volumes for acquisitions performed with (green circles) and without (purple squares) the thorax phantom. A marked upward shift in measured volumes was observed when the thorax phantom was absent compared to when it was present

Table 3.

Relation between cardiac phantom displacement and CZT-ERNA volumes

Series	Positional Shift (mm) anterior-lateral	CZT-ERNA volume (mL)
Series 1 CT volume 56 mL	5–12	64
	30–38	62
	60–65	61
	92–99	53
Series 2 CT volume 70mL	5–25	77
	45–65	71
	75–95	72
	105–125	48

A key explanation for the LVEF discrepancy lies in the software. The PET package (Corridor4DM) allowed basal plane motion to be fixed at 0 mm to match the phantom, whereas the ERNA package (QBS) modeled basal displacement by default. This yields larger EDV, smaller ESV, and thus higher LVEF in QBS than in Corridor, highlighting that algorithmic assumptions can drive inter-modality differences as much as detector physics.

Volumes were not interchangeable across modalities because systematic scale biases persisted (Figs. 4, 5). LVEF behaved more robustly than volumes, as the EDV/ESV ratio reduces scale error. In this phantom, QBS’s modeled basal motion favored CZT-ERNA for LVEF. In vivo, where basal motion is present, this software-driven difference may narrow, and attenuation-related effects might dominate any remaining discrepancies.

Phantom design also influenced results. In our initial evaluation series (1 and 2), close contact between the inner and outer balloons created an apparent perfusion defect, leading to underestimated PET volumes and LVEF (Fig. 9). In series 3 we mitigated this issue, yielding values closer to CT, which is supported by the lower mean bias and narrower limits of agreement for PET shown in Table 2. Thus, volumes may be further underestimated if the edge-detection algorithm fails to constrain itself in myocardial regions with low uptake. Alongside tracer doses, collisions observed during series are reported in Table 1.

Fig. 9.

Collision of inner and outer lumens in the dynamic cardiac phantom. a CT image used for volume estimation shows direct contact between the inner balloon and the surrounding lumen. b When analyzed with myocardial perfusion imaging (MPI) software, this resulted in an apparent perfusion defect in the septal region

CT served as the reference standard, offering clear boundary definition and excellent reproducibility. While CT contouring is simplified by sharp lumen contrast and may not fully replicate the blurred borders of real myocardium, it provided a consistent benchmark. By comparison, nuclear software is optimized for noisy, less well-defined endocardial edges, which could partly explain the systematic underestimation of LVEF.

Several limitations must be noted. The phantom reproduces circular contraction of the myocardium but lacks longitudinal motion, particularly basal plane displacement. This simplification likely contributes to the systematic underestimation of LVEF in the nuclear methods compared with CT. Importantly, inter-modality agreement between CZT-ERNA and PET-MPI was closer than either modality’s agreement with CT, suggesting that a more physiologically realistic phantom, including basal movement, might improve absolute agreement with CT. Lastly, extracardiac tracer activity was not simulated.

Conclusion

This study evaluated the accuracy of cardiac PET and CZT-ERNA for volume measurements using a novel dynamic cardiac phantom. PET progressively underestimates volumes in the static series, while CZT-ERNA progressively overestimates them. Both modalities underestimate LVEF, with a greater discrepancy in PET. Attenuation correction functions as expected in PET but is absent in CZT-ERNA, which may lead to patient size-dependent errors.

Overall, in this phantom, volumes were not interchangeable across modalities. LVEF behaved more robustly than volumes yet remained modality- and software-dependent. In clinical settings where basal motion is present, software-driven differences may diminish, and attenuation correction may dominate any residual discrepancies. Future work should aim to improve longitudinal/basal motion which would refine cross-modality comparisons.

Author contributions

MØN and BZ designed the study. MØN performed phantom acquisitions, analysed data, and drafted the manuscript. RW contributed to drafting the manuscript. TCH, LLD, JDH, and SH contributed to phantom design, construction, and technical implementation. LTJ and BZ supervised the study and provided critical revisions. All authors read and approved the final manuscript.

Funding

Open access funding provided by Copenhagen University. This research was funded by the Department of Nuclear Medicine, Copenhagen University Hospital – Herlev and Gentofte.

Data availability

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

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.