An integrated PET/CT simulation framework for virtual imaging trials and quantitative performance evaluation

Abstract

Purpose:

The purpose of this study was to demonstrate the use of integrated PET/CT virtual imaging trials (VITs) for evaluating PET/CT acquisition protocols and clinically relevant sources of PET quantification variability under controlled, repeatable conditions.

Methods:

An integrated PET/CT simulation framework was developed by combining SimSET for PET and DukeSim for CT, enabling concordant multi-modality simulations from a single human model input, including CT-derived attenuation correction. Geometric concordance was verified using an in-silico PET/CT coregistration test. Validation employed a NEMA IEC body phantom by comparing simulated PET/CT images with clinical scans using standard uptake value (SUV) and contrast recovery coefficient (CRC) metrics. Application studies assessed respiratory phase mismatch between PET and CT acquisitions, the impact of CT protocol selection on attenuation correction for free-breathing PET, and CT dose-reduction strategies for attenuation correction.

Results:

Geometric concordance was achieved with a maximum PET/CT centroid difference of 1.83 mm. Phantom validation demonstrated agreement between simulated and clinical PET quantification, with SUV measurements within 6% using CT-derived attenuation correction and within 12% using an ideal attenuation map. Application studies revealed that respiratory phase mismatch and CT protocol selection can produce clinically significant differences in PET quantification metrics. Dose-reduction studies showed minimal sensitivity of PET quantification to ultra-low-dose CT attenuation correction, with CRC deviations ≤ 0.23% across CT conditions for the same sphere size.

Conclusion:

The integrated PET/CT simulation framework enables virtual imaging trials for controlled and reproducible investigations of PET/CT protocols, supporting multi-modality imaging optimization and quantitative performance evaluation under clinically relevant conditions.

1. Introduction

Positron emission tomography-computed tomography (PET/CT) is a widely used multi-modality imaging technique. CT provides anatomical information based on x-ray attenuation, while PET records metabolic and physiological activity using radioactive tracers. CT enables co-localization and is used for attenuation and scatter correction of PET emission data[1,2]. The combined use of these modalities enhances lesion detection and tumor localization, which aids in staging, diagnosis, and treatment planning[3,4,5]. PET/CT is also essential in emerging fields such as theranostics, radiomics, and personalized therapy, playing a critical role in treatment monitoring and dosimetry calculations[6,7]. These advancements have driven significant progress in PET/CT technology. As PET/CT technology and its clinical applications continue to expand, objective evaluation is needed to ensure accurate and reliable imaging performance.

Objective evaluation of any imaging technology is ideally done through patient imaging trials. However, such trials are challenging due to costs, potential radiation exposure concerns, and limited knowledge of ground truth. As an alternative, physical phantom studies can mitigate some of these limitations, and further provide an opportunity for optimizing the technique. They, however, do not reflect the specifics of human patients. The limitations of human and phantom studies can be addressed using virtual imaging trials (VITs)[8] that provide patient-level specificity while also addressing the cost, patient burdens, and ground truth limitations of real trials. VITs in PET/CT require simulating both PET and CT imaging processes. Currently, these simulations are done independently[9,10,11,12]. This creates a challenge in that the outputs of the two cannot be readily integrated. Without integration, PET simulations would need to rely on reconstructing simulated PET emission data with idealized, pre-calculated, attenuation maps of materials under ideal conditions, rather than CT-derived attenuation correction coefficients that would be available from CT simulations[13]. Ideally, a virtual trial should replicate the condition under which real PET and CT data sets are integrated clinically, which further enable studying the best way of integration.

In this study, we developed an integrated PET/CT virtual imaging framework to enable the evaluation and optimization of PET/CT protocols. Using this framework, we demonstrate its utility for investigating clinically relevant sources of PET quantification variability, including the effects of PET/CT acquisition protocols that influence respiratory phase alignment between PET and CT acquisitions. Additionally, the framework was used to perform studies in a fully virtual setting that are analogous to prior investigations of PET/CT dose-reduction strategies, illustrating its potential as a valuable tool for protocol optimization and quantitative performance evaluation without the constraints of physical phantoms or patient studies.

2. Methods

This work demonstrates how virtual imaging trials can be used to support quantitative PET/CT performance evaluation and protocol optimization. To enable these studies, an integrated PET/CT simulation framework was developed and validated.

2.1. Integrated PET/CT simulation framework development

The integrated PET/CT simulation framework developed in this work used PET and CT simulators that have been independently validated in previous studies. The PET simulator was SimSET, a Monte Carlo-based simulation software that models physical effects and acquisition geometry for emission imaging. SimSET performs decay-by-decay particle tracking, in which radioactive decays and photon histories are sampled and transported through the phantom and scanner system. This approach enables modeling of stochastic and physical effects relevant to clinical PET imaging, including positron range, non-collinearity, scatter, and random events. A key differentiator of SimSET is that is only tracks interactions in the energy range of nuclear medicine simulators, making it many orders of magnitude faster than general purpose simulators[14]. The utilized CT simulator was Duke-Sim, which uses a hybrid approach to estimate x-ray interactions by combining ray tracing and photon tracking methods. For computational efficiency, SimSET was executed via SimPET[15], which facilitates parallel sub-processing. Building upon our previous work[16], SimSET [17] and DukeSim[18] were integrated to ensure geometric and input concordance, allowing a single anthropomorphic model to be propagated consistently through both PET and CT simulations and to generate co-registered images suitable for CT-based attenuation correction. An overview of the integrated framework is shown in Fig. 1. SimSET and DukeSim were developed independently, meaning their integration required standardizing inputs and geometrical definitions, followed by verification and validations. These processes were systematically undertaken as detailed in the following sections.

Fig. 1.

Integrated PET/CT simulation framework overview.

2.1.1. Input concordance

DukeSim (CT) and SimSET (PET) both use voxelized computational phantoms with label-based material assignments but differ in their requirements. Both simulators require energy-dependent attenuation properties assigned to each voxel label. PET simulations additionally require radioisotope concentrations assigned to each voxel. Simulating unique or patient-specific materials in both modalities requires the ability to define new materials with custom physical properties. This functionality is directly supported within the current DukeSim simulation package. While it is also possible in SimSET, the process is tedious and generally discouraged without extensive programming experience. To address this and enable consistent, flexible multi-modality simulations, we developed an automated material mapping method between both simulators. This method uses a unified material table implemented as a spreadsheet input file, with each row corresponding to a phantom label and columns defining physical properties for each phantom label, including elemental composition and density (for attenuation modeling in both CT and PET) and activity concentration (for PET), and is structured for compatibility with both simulators. This table is used directly as the DukeSim material definition input, while a wrapper script reads the same table and populates the SimSET-specific material definition and activity input files required for PET simulation.

2.1.2. Geometric concordance

Geometric concordance in multi-modality imaging simulations ensures proper acquisition geometry definitions, sinogram formation, and image reconstruction. Geometric concordance was established for the integrated PET/CT simulation framework by standardizing the coordinate systems and maintaining a consistent reference frame (range across both the PET and CT simulation modules), enabling the use of a singular input simulation object for the entire integrated framework. An example of the unified coordinate system for a four-dimensional male human model (extended cardiac-torso (XCAT) phantom[19]) is shown in Fig. 2. In this example, a sagittal view of the PET and CT acquisition geometries is shown, where both are defined relative to the XCAT phantom.

Fig. 2.

XCAT coordinate system for PET/CT integrated simulation framework. Both simulators (PET and CT) are defined in terms of input computational phantom.

2.1.3. Performing a PET/CT simulation with the integrated framework

To run a simulation, the framework begins by performing a CT simulation of the input phantom with a given acquisition protocol, generating CT sinogram data. This CT sinogram data is then reconstructed using filtered back projection, with a smooth Hamming reconstruction filter applied to approximate a soft-tissue CT reconstruction kernel, through an open-source reconstruction toolkit (Multi-Channel Reconstruction Toolkit[20]). This produced CT images in Hounsfield units (HU) corresponding to an effective energy of 40–70 keV, depending on the scanner model and x-ray tube power selection. The HU values are subsequently scaled to 511 keV, which is the relevant energy of annihilation photons produced in PET imaging[1,5]. The scaled CT image is then downsampled and smoothed to PET resolution to generate the attenuation map. The attenuation map is forward projected and used for attenuation correction of the simulated PET emission data, which is generated by simulating a PET acquisition based on user defined scanner parameters, activity distributions, and acquisition protocols. The PET emission sinogram is then reconstructed using either an open-source reconstruction toolkit[21] or the GE Duetto toolbox (GE Healthcare, United States) producing PET images in activity concentration units (Bq/ml).

2.2. Framework verification and validation

To ensure the accuracy and reliability of the integrated simulation framework, verification and validation tests were performed. Verification tests were employed to confirm that framework inputs and outputs functioned as expected, while validation experiments assessed whether the simulated data aligned with real-world physical measurements.

2.2.1. Verification

Verification was performed through simulating a PET/CT coregistration test, following the Image Quality (IQ) phantom NEMA NU 2–2018 guidelines[22]. This test verifies that the reconstructed PET/CT images generated with the integrated simulation framework are properly aligned. The procedure was adapted from Pan et al.[23], who performed a physical co-registration test on a clinical PET/CT scanner; in this work, the procedure was implemented in-silico, using simulated PET and CT acquisitions. Five point sources, modeled as 1 mm × 1 mm capillary tubes, were simulated at the center of the axial field of view. Three sources were positioned laterally at vertical offsets of 1 cm, 10 cm, and 20 cm, while two were positioned vertically with horizontal offsets of 10 cm and 20 cm. The point sources were simulated as iodine-based CT contrast with an F-18 activity concentration of 185 MBq/ml. For co-registration analysis, localized volume of interests (VOIs) were defined using the FWHM of the reconstructed point-source profiles. The FWHM was determined from the distance between the left and right half-maximum positions for PET and CT profiles in each direction, and the largest FWHM across sources, directions, and modalities was used to define the VOI width as eight times the maximum FWHM. PET and CT centroid locations were then calculated within each VOI using an intensity-weighted centroid calculation, where each coordinate was weighted by the corresponding voxel intensity. The Euclidean distance between centroids was measured to assess alignment accuracy.

2.2.2. Validation

Validation was based on comparisons between simulated and experimental measurements of a NEMA IQ hot sphere phantom with a 10:1 sphere-to-background activity ratio. A physical phantom was used for experimental measurements. For simulation, a computational phantom was generated with the same dimensions, material definitions, and activity distributions.

Both the background and hot sphere regions of the physical phantom were filled with water. The background regions had an activity concentration of 2 kBq/ml and hot spheres had an activity concentration of 20 kBq/ml. The physical phantom was scanned using a clinical PET/CT scanner (Discovery MI 5-ring, GE Healthcare). The CT acquisition was acquired in helical mode with tube voltage of 120 kV and tube current of 50 mAs. The PET acquisition was acquired over a 300 s duration and was reconstructed using the ordered subsets expectation–maximization algorithm (OSEM)[24], 34 subsets and 6 iterations with a 2.73 × 2.73 × 2.80 mm voxel size. Reconstructed images were collected with no post reconstruction filter and with a post reconstruction 6 mm Gaussian filter.

A PET/CT acquisition of the computational phantom was simulated using the integrated framework. This simulation modeled the same scanner geometry, acquisition parameters, activity concentrations, and reconstruction parameters as the experimental measurement. Attenuation correction was applied to the reconstructed PET images using both simulated CT attenuation maps from DukeSim and ideal attenuation maps. The ideal attenuation maps were generated using the linear attenuation coefficient for water at 511 keV (0.096 cm⁻¹)[25], with the phantom modeled as a uniform water-equivalent volume.

To demonstrate the impact of using a simulated CT compared to an ideal uniform water-equivalent attenuation map at 511 keV, which does not model any CT acquisition or scaling effects, on the realism of the simulated PET/CT images, the simulated images were compared with the corresponding experimental images in terms of relevant quantitative metrics, including activity concentrations, contrast recovery coefficient (CRC), and standard uptake value (SUV).

CRC was measured following the NEMA NU-2 standard as:

$$CR{\text{C}}_{\text{hot}}=\frac{\frac{{\text{C}}_{\text{hot}}}{{\text{C}}_{\text{bkgd}}}-1}{\frac{{\text{a}}_{\text{hot}}}{{\text{a}}_{\text{bkgd}}}-1},$$ (1)

where $C_{hot}$ and $C_{bkgd}$ are hot sphere and background counts for each sphere size, $a_{hot}$ is the known hot sphere activity concentration, and $a_{bkgd}$ is the background activity concentration.

SUV (g/ml) was measured as:

$$SUV=\frac{\text{A}}{\left(d^{\prime }/w\right)}.$$ (2)

where $A$ is the radioactivity concentration (Bq/ml) from a voxel or region of interest (ROI) in the image, $d^{\prime }$ is the decay corrected injection dose to match the time the activity $\left(A\right)$ is measured by the PET scanner, and $w$ is the weight of the patient (g). For these simulations, the decay-corrected injected activity was equal to the value of the initial injected activity as the simulation modeled a single time point.

2.3. Framework application studies

The integrated framework was used to evaluate the impact of PET/CT acquisition techniques on PET quantification metrics, focusing on respiratory phase mismatch and CT dose reduction for attenuation correction.

2.3.1. Pilot VIT 1: CTAC alignment accuracy

The first VIT examined the effects of misaligned CT based attenuation correction (CTAC) maps on PET image quantification. In clinical PET/CT imaging, changes in respiratory phase between the two acquisitions can cause a misalignment of the applied CTAC map due to anatomical shifts between inspiration and expiration respiratory states. In inspiration, the liver is pushed downward as the lung volume increases, while in expiration the liver rises as the lung volume decreases. The uptake of different anatomical sites in PET imaging is often normalized to the liver or blood pool[26]. Therefore, misalignment of the normalization structures in the applied CTAC map due to respiratory phase differences impacts SUV measurements observed in PET images.

For this VIT, a four-dimensional male human model XCAT phantom was used with an ¹⁸F-FDG radionuclide distribution based on clinically published average biodistributions[27,28,29], and a body mass index (BMI) of 24.38 (representing 50th percentile, world population)[30]. The published biodistributions provided reference organ SUV values, which were used to define organ activity concentrations for the XCAT model. Specifically, organ activity concentrations were calculated to approximate the published reference SUV values for the injected activity and body weight of the XCAT phantom. These activity concentrations were integrated with the phantom geometry through the activity input table that assigned a radionuclide concentration to each XCAT phantom label, modeling organ specific uptake values. The computational XCAT phantom was imaged using the developed PET/CT simulation framework.

A static, expiration phase PET acquisition of the XCAT model was simulated, and the corresponding emission data was reconstructed with an inspiration CTAC map that was misaligned due to the anatomical shifts between expiration and inspiration, and an aligned expiration CTAC map. The expiration phase was chosen for the PET acquisition as about 2/3 of the respiratory cycle for an average patient with normal breathing is spent in the expiration state[31]. While these static phase simulations are not reflective of an active breathing pattern that a patient would exhibit, they provide a reference understanding for how the alignment of CTAC maps influence observed uptake levels in anatomical sites. SUVmean and SUVmax measurements were taken for the liver, myocardium, and blood pool for the simulated PET images reconstructed with the misaligned and aligned CTAC maps.

2.3.2. Pilot VIT 2: Impact of CT protocol on PET SUV

The second VIT aimed to assess the impact of different CT protocols on PET image quantification for free breathing PET acquisitions. For this study, a respiratory motion XCAT phantom was generated using the default XCAT parameters for normal tidal breathing, sampled over a five second respiratory cycle initiated at the end-expiration phase. The activity distribution and body composition of the phantom was the same as the phantom used in the static CTAC alignment study described above.

The simulated PET acquisition of the motion phantom was acquired over a 300 s period and reconstructed with the OSEM algorithm and 2.73 × 2.73 × 2.80 mm voxel sizes. The same simulated PET data was reconstructed with four different CTAC maps generated from simulated CT acquisitions. The simulated CT acquisitions were all acquired at 120 kVp, 50 mAs, and 1.0 pitch, and scaled to PET resolution and energy levels.

The simulated CT acquisitions were acquired for breath hold and free breathing protocols to model common clinical scenarios for CT acquisitions. More specifically, simulated CT images were acquired at 1) inspiration breath-hold, 2) expiration breath-hold, 3) free-breathing with six respiratory phase bins, and 4) free-breathing with sixty respiratory phase bins. For the free-breathing simulations, the phase bins were interpolated across the respiratory trace to model continuous motion. The two free breathing simulated CT acquisitions were performed with a different number of respiratory phase bins to further assess the effect of the temporal resolution of the simulated phantom used in the CT acquisition on PET image quantification.

SUVmean and SUVmax measurements were taken for the liver, myocardium, and blood pool for the simulated PET images reconstructed with each of the different CTAC maps. The activity concentrations in each of these sites were measured on the simulated PET images and compared to the input, or ground truth, activity concentrations defined for the liver, myocardium, and blood pool. As the simulation domain was limited to the thoracic region, SUV calculations were normalized using the corresponding computational phantom mass and the ground truth activity defined for that region, representing a realistic approximation of post-injection tracer distribution.

2.3.3. Pilot VIT 3: PET/CT dose reduction evaluation

The third VIT evaluated the impact of CT dose reduction on CT-based attenuation correction and PET quantification. This study was designed to be analogous to part of a prior investigation by Mostafapour et al. on ultra-low-dose CT protocols for PET/CT[32], but in a fully virtual setting.

Two studies were performed. A NEMA IEC image quality phantom configuration was simulated with a 10:1 hot sphere-to-background activity ratio. PET emission data was reconstructed using CTAC maps generated from CT acquisitions performed with different tube potentials and tube current–time products (100 and 120 kV; 10, 25, and 50 mAs). A baseline CTAC condition of 120 kV and 50 mAs was used for comparison. PET differences between CTAC conditions were assessed using voxel-wise activity concentration differences and contrast recovery coefficient as a function of sphere size.

The study was then repeated using a chest XCAT phantom to evaluate CT dose reduction effects for realistic anatomy. In addition to voxel-wise activity concentration differences, relative differences in measured activity concentration were quantified for anatomical structures of interest (liver, heart blood pool, and spinal cord) relative to the baseline CTAC condition (120 kV, 50 mAs), calculated as[32]:

$$RD=\frac{A_{ULD}-A_{LD}}{A_{LD}}$$ (3)

where $A_{LD}$ represents the baseline CTAC and $A_{ULD}$ represents the subsequent ultra-low-dose CTAC maps. RDmean represents the percent difference in the average activity concentration within a defined region of interest, while RDmax reflects the percentage difference in the single highest voxel value within that ROI, calculated similarly using the maximum voxel intensity.

3. Results

3.1. Integrated framework concordance

Fig. 3 provides a qualitative demonstration of geometric concordance within the integrated PET/CT simulation framework. Reconstructed PET images are shown with aligned CT based attenuation correction and misaligned CT based attenuation correction. The aligned CT based attenuation correction condition represents the expected operation of the integrated framework, in which the PET activity distribution and CT derived attenuation map are registered in a common coordinate system. In contrast, the misaligned CT based attenuation correction condition demonstrates the image errors that can arise when geometric concordance between the PET and CT components is not maintained. The left column contains the original PET reconstructions, while the right column displays difference images, highlighting deviations from the ground truth activity distribution used in the simulation. These results demonstrate the importance of geometric concordance across the PET and CT simulation modules.

Fig. 3.

Simulated PET images (left column) & difference between the simulated PET image and ground truth activity distribution (right column).

3.2. Framework verification and validation

3.2.1. Verification

Fig. 4 shows the line profile of PET signal profiles alongside CT centroid locations for the three point sources positioned along the vertical axis. A PET/CT fusion image of all five point sources is also included. Table 1 provides the measured PET and CT centroid locations and the Euclidean distance between centroids, quantifying the spatial alignment between the two modalities for the integrated framework. Small deviations between the nominal and measured source positions reflect finite source size, voxel sampling, and stochastic effects modeled in the PET simulation, including positron range, non-collinearity, scatter, and random coincidences. The centroids are reported with sub-voxel numerical precision because the calculation uses intensity-weighted information across multiple voxels rather than assigning each centroid to a single voxel center. The presentation format, including the line profiles and fusion image, was adapted from an experimentally validated approach in prior work [23].

Fig. 4.

Point source PET signal profiles and CT centroid locations for the PET/CT coregistration test. PET/CT coregistration test fusion image is shown to the right of the plot.

Table 1.

PET and CT point source centroid locations and differences by Euclidian distance (Δ).

Distance (cm)	PET Centroid (x, y, z) (cm)	CT Centroid (x, y, z) (cm)	Δ (cm)
1 cm Vertical	(0.060, 1.026, 0.051)	(−0.016, 1.085, 0.053)	0.10
10 cm Vertical	(−0.005, 9.981, 0.041)	(0.000, 10.022, 0.037)	0.04
20 cm Vertical	(−0.065, 20.003, 0.034)	(0.009, 20.002, 0.001)	0.08
10 cm Horizontal	(9.955, 0.114, 0.051)	(9.983, 0.038, 0.044)	0.08
20 cm Horizontal	(19.876, 0.193, 0.012)	(19.940, 0.022, 0.0184)	0.18

3.2.2. Validation

Fig. 5 presents the reconstructed PET images from the NEMA hot sphere PET simulation with simulated CT based attenuation correction and clinical scanner data.

Fig. 5.

NEMA hot sphere PET emission images. DukeSim CT attenuation map used for attenuation correction (top row) and clinical scan data (bottom row).

Fig. 6 shows standard uptake values (SUVs) between clinical PET images, simulated PET images reconstructed with a DukeSim CT attenuation map, and simulated PET images reconstructed with an ideal attenuation map. Nonfiltered data is represented in darker shades, while data with a post-reconstruction smoothing filter appears in lighter shades.

Fig. 6.

NEMA IEC hot sphere SUV versus sphere size for clinical PET and simulated PET (DukeSim CTAC or ideal attenuation). Dark bars indicate no filter; light bars indicate a 6 mm Gaussian filter. Error bars show standard deviation across noise realizations.

CRC measurements for the clinical data, simulated data reconstructed with a DukeSim CT attenuation map, and simulated data reconstructed with an idealized attenuation map demonstrated the same trend and relative relationships as the SUV measurements shown in Fig. 6.

3.3. Framework application studies

3.3.1. Pilot VIT 1: CTAC alignment accuracy

Fig. 7 presents the reconstructed images from the pilot virtual imaging trial, which investigated the effect of misaligned CT based attenuation correction (CTAC) maps on SUV quantification. The first row displays images from an expiration-phase PET acquisition reconstructed using an inspiration-phase CT attenuation map, while the second row presents images reconstructed using a matched expiration-phase CT attenuation map. Axial, coronal, and sagittal views are provided for each case, with consistent window level settings applied across all images. This provides an anatomical reference for the effect of misaligned CTAC maps on reconstructed PET images.

Fig. 7.

Simulated PET images acquired in expiration, reconstructed with inspiration and expiration CTAC maps. Mismatch liver dome artifacts indicated by red arrows. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Table 2 quantifies these results, showing SUV measurements in the liver, myocardium, and blood pool for the two simulations. The table reports mean and maximum SUV values for each region, measured using identical regions of interest (ROIs) across both conditions. This table quantifies the effects of CTAC alignment for reconstructed PET images.

Table 2.

SUV metrics for expiration PET simulation reconstructed with inspiration and expiration simulated CT based attenuation maps.

Region of Interest	Inspiration CTAC	Expiration CTAC
SUVmax Liver	2.40	3.30
SUVmean Liver	1.61	2.42
SUVmax Myocardium	4.62	5.17
SUVmean Myocardium	3.60	3.99
SUVmax Blood Pool	2.06	2.20
SUVmean Blood Pool	1.11	1.17

3.3.2. Pilot VIT 2: Impact of CT protocol on PET SUV

Fig. 8 presents simulated PET/CT images for the free-breathing PET simulation reconstructed with different simulated CT based attenuation correction maps. The figure PET image reconstructions with an inspiration-phase CT based attenuation correction, an expiration-phase CT based attenuation correction, a free-breathing CT based attenuation correction with six motion bins, and free-breathing CT based attenuation correction with sixty motion bins. All images are displayed using the same axial slice and window level settings to enable direct visual comparison. Differences in SUVs for the liver, myocardium, and blood pool across the four reconstructions are presented in Table 3 to demonstrate the effect of CT acquisition protocols on PET quantification. Table 4 shows percent difference between the mean recorded uptake values in the liver, myocardium and blood pool compared to the ground truth activity distribution used as an input for the simulations.

Fig. 8.

Simulated PET/CT fusion images for free breathing PET acquisition reconstructed with different simulated CTAC maps.

Table 3.

SUV measurements for liver, myocardium, and blood pool for free breathing PET simulation reconstructed with different simulated CTAC maps.

Region of Interest	Inspiration CTAC	Expiration CTAC	Respiratory Motion CTAC, 6 bins	Respiratory Motion CTAC, 60 bins
SUVmax Liver	3.79	4.82	4.32	4.74
SUVmean Liver	2.26	3.35	2.83	3.30
SUVmax Myocardium	6.87	7.18	6.95	7.16
SUVmean Myocardium	4.50	4.73	4.57	4.70
SUVmax Blood Pool	2.03	2.14	2.07	2.12
SUVmean Blood Pool	1.39	1.44	1.40	1.43

Table 4.

Percent difference in mean activity uptake for liver, myocardium, and blood pool between simulated data and ground truth activity map.

Region of Interest	Inspiration CTAC	Expiration CTAC	Respiratory Motion CTAC, 6 bins	Respiratory Motion CTAC, 60 bins
Liver	−29.48	4.70	−11.48	2.94
Myocardium	−15.97	−11.74	−14.81	−12.35
Blood Pool	2.88	2.21	2.92	2.46

3.3.3. Pilot VIT 3: PET/CT dose reduction evaluation

Results are presented in a format consistent with the physical phantom PET/CT dose-reduction evaluation[32] to facilitate comparison between the published experimental work and this virtual study. Fig. 9 presents reconstructed PET images for the NEMA IEC image quality phantom using CTAC derived from baseline low-dose CT (120 kV, 50 mAs) and ultra-low-dose CT (100 kV, 10 mAs). Difference images and activity concentration line profiles are shown to visualize voxel-wise PET activity changes attributable to CTAC dose differences. Fig. 10 shows contrast recovery coefficient (CRC) as a function of sphere size across CTAC dose. For each sphere size, CRC values were closely aligned across CTAC dose levels, with a maximum deviation of approximately 0.23%.

Fig. 9.

NEMA IEC phantom PET images reconstructed using baseline low-dose (120 kV, 50 mAs) and ultra-low-dose CTAC (100 kV, 10 mAs), with corresponding PET difference images and activity concentration line profiles.

Fig. 10.

CRC versus sphere size for the NEMA IEC phantom across CT acquisition conditions (100/120 kV; 10/25/50 mAs) used to generate CTAC maps.

Fig. 11 presents PET images reconstructed with baseline low-dose and ultra-low-dose CTAC for the chest XCAT simulation, with corresponding activity concentration line profiles and voxel-wise difference profiles. Table 5 reports relative differences in measured activity concentration metrics for anatomical structures of interest (liver, heart blood pool, spinal cord) across CTAC dose levels relative to the baseline CTAC map (120 kV, 50 mAs).

Fig. 11.

Chest XCAT PET images reconstructed using baseline low-dose (120 kV, 50 mAs) and ultra-low-dose CTAC (100 kV, 10 mAs), with activity concentration line profiles and PET difference profiles.

Table 5.

Relative differences in activity concentration metrics for structures of interest across ultra-low-dose ct acquisitions relative to low-dose baseline (120 kV, 50 mAs).

100 kV, 10 mAs
	Liver	Heart Blood Pool	Spinal Cord
RDMean (%)	0.32	0.41	0.39
RDMax (%)	0.372	0.36	−0.05
100 kV, 25 mAs
	Liver	Heart Blood Pool	Spinal Cord
RDMean (%)	−0.07	−0.12	−0.17
RDMax (%)	−0.25	−0.28	−0.39
100 kV, 50 mAs
	Liver	Heart Blood Pool	Spinal Cord
RDMean (%)	−0.38	−0.39	−0.51
RDMax (%)	−0.45	−0.44	−0.78
120 kV, 10 mAs
	Liver	Heart Blood Pool	Spinal Cord
RDMean (%)	0.77	0.89	0.87
RDMax (%)	0.76	0.69	0.65
120 kV, 25 mAs
	Liver	Heart Blood Pool	Spinal Cord
RDMean (%)	0.37	0.37	0.31
RDMax (%)	−0.10	0.12	0.19

4. Discussion

This study demonstrates how virtual imaging trials (VITs) can be used to evaluate PET/CT acquisition protocols and clinically relevant sources of PET quantification variability. To enable these studies, a concordant PET/CT simulation framework was developed and validated that utilizes a single computational phantom as input, ensuring spatial and input consistency across both modalities.

The concordance test, as shown by the PET/CT simulation of the XCAT phantom in Fig. 3 provides an anatomical reference that demonstrates the importance of accurately aligned CTAC maps for PET image quantification relative to the ground truth activity distribution used in the simulation framework.

The PET/CT coregistration tests, shown in Fig. 4 and Table 1 verified the geometric concordance of the framework, with a maximum centroid difference between PET and CT of 1.83 mm, which is within the expected 2–4 mm clinical PET voxel size. The measured centroid positions are image-derived estimates and are influenced by finite source size, voxel sampling, reconstruction effects, and physical processes modeled in the PET simulation. For example, the mean positron range of F-18 in soft tissue has been reported as approximately 0.27 mm[33], which is below the reconstructed PET voxel size but can still contribute to sub-voxel differences in reconstructed source localization. These results show that the PET/CT simulation framework can achieve geometric concordance within acceptable clinical limits[34,35], which is an essential requirement for multi-modality imaging research and important for accurate quantification. The NEMA IQ phantom hot sphere simulations, shown in Fig. 5, provide quantitative validations for using a simulated CT compared to an ideal attenuation map for correction of PET emission data.

As demonstrated in Fig. 6, PET image quantification measurements were within 6% of the clinical data comparison when using the simulated CT for attenuation correction, and within 12% of the clinical comparison when using an ideal attenuation map for attenuation correction. On average, the PET emission data that was reconstructed with the simulated CT based attenuation correction map, rather than the ideal attenuation map, resulted in PET image metrics that more closely resembled clinical data, further reinforcing the clinical relevance of the proposed framework for multimodality imaging studies. This study suggests that simulated CT based attenuation correction leads to more accurate quantification of PET image, exemplifying the benefit of incorporating and modeling CT acquisition processes into this integrated framework.

The integrated PET/CT simulation framework provides a valuable tool for evaluating clinical imaging protocols in a controlled and reproducible manner. Its utility was demonstrated through two virtual imaging trials designed to study the impact of respiratory motion during CT acquisitions on PET SUV quantification. Specifically, the integrated simulation platform enables the study of CT respiratory motion effects on PET image quantification, which is critical for accurate SUV measurements in clinical and research settings. Blood pool and liver SUV values are frequently used as metrics to guide patient treatment[26], with blood pool SUV baselines typically established through clinical trials and the liver serving as a reference for individual patient scans [36,37]. This VIT investigated the effect of respiratory phase differences between CT and PET acquisitions on PET quantification, as shown in Fig. 7. As shown in Table 2, this study confirmed that using a CTAC map acquired at inspiration for an expiration-phase PET acquisition, results in quantifiable SUV differences.

The second VIT, which simulated a free-breathing PET acquisition reconstructed with different CT acquisition protocols, demonstrated the impact of intrascan CT respiratory motion on PET image quantification from a clinical perspective, while also highlighting how, from a simulation standpoint, the temporal resolution of computational phantoms influences quantification accuracy. As shown in Fig. 8 and quantified in Table 3, it showed that different CT acquisition protocols lead to quantifiable variations in SUV measurements. This further highlighted the sensitivity of PET quantification to intrascan CT respiratory motion. Clinically, these results suggest that using an expiration breath-hold CT acquisition protocol for free-breathing PET acquisitions results in more accurate image quantification that is closer to the ground truth activity distribution, demonstrated in Table 4. From a simulation perspective, this further implies that incorporating more phase bins in CT motion phantoms used for PET attenuation correction enhances the realism and accuracy of simulations.

The third VIT evaluated PET quantification sensitivity to CT dose reduction for attenuation correction in a fully virtual setting. This study demonstrated that ultra-low-dose CT acquisitions used for CTAC resulted in minimal changes in reconstructed PET activity and quantification metrics compared to the baseline CTAC condition, suggesting that CT dose reduction does not significantly impact PET quantification for the evaluated cases. Notably, these findings are consistent with prior physical phantom investigations of PET/CT dose-reduction strategies [32], demonstrating that similar conclusions can be reached using integrated PET/CT virtual imaging trials without the constraints of physical phantom scanning or patient studies.

This work underscores the importance of optimizing CT acquisition protocols for PET/CT, as even small differences in SUV in regions such as the blood pool can markedly impact patient management[38,39,40,41]. The integrated framework provides a unique capability to simulate the same patient or simulation object multiple times with slight variations in acquisition protocols, allowing for systematic evaluation of PET/CT protocol variations and their effects on PET image quality metrics. These findings underscore the framework’s value as an essential tool for optimizing imaging strategies and improving quantitative accuracy.

This study had some limitations. First, the validations included a single scanner model, although these can be expanded to additional PET/CT scanner models and conditions. Further, the pilot VIT was performed with an XCAT phantom and activity distribution that modeled a standard male patient. For the pilot VIT studies, repeated noise realizations were not performed to estimate uncertainty from stochastic effects in the PET simulations. Future work could characterize uncertainty in simulated PET quantification through repeated simulations. Additional pilot imaging studies that model a variety of body compositions and radionuclide distributions will provide an enhanced understanding of the effect of respiratory motion and CT acquisition protocols on PET image quantification for a larger patient population. To perform these larger scale virtual imaging trials in a computationally efficient manner, the framework can be further expanded to include an analytical PET simulator option[42]. Future work will focus on expanding the verification and validation dataset, harmonizing PET/CT image quality across different scanner systems and utilizing the framework to study theranostics through virtual clinical trials. These advancements will further establish the framework’s clinical relevance and extend its applications to a broader range of PET/CT research areas.

5. Conclusion

This study demonstrates how integrated PET/CT virtual imaging trials can be used to support evaluation and optimization of PET/CT imaging techniques. To enable these studies, this work addressed a key limitation of existing PET simulation platforms by developing a fully integrated PET/CT simulation framework for multi-modality virtual imaging trials. The framework provides co-registered PET and CT images from a single human model input, enabling systematic, patient-specific evaluations of PET/CT imaging techniques. Using this integrated approach, protocol variations and acquisition effects, including respiratory phase mismatch and CT dose reduction for attenuation correction, can be evaluated in a controlled and repeatable environment. Further, the integrated framework improves the realism of simulated images by incorporating CT-derived attenuation correction. Overall, this framework provides a valuable tool for multi-modality virtual imaging trials (VITs), which are simulation-based methods to address limitations associated with real trials and can be used to support quantitative performance evaluation and optimization of PET/CT technologies and protocols for improved image quality and quantification accuracy.

Code and Data Availability

The simulation and reconstruction workflow used in this study incorporated SimSET, DukeSim v1.2, the Multi-Channel Reconstruction Toolkit (MCR Toolkit), STIR, and the GE Duetto toolbox. SimSET, the public version of the MCR Toolkit, and STIR are publicly available through their respective project repositories or websites: SimSET (depts.washington.edu/simset/downloads), MCR Toolkit (gitlab.oit.duke.edu/dpc18/mcr-toolkit-public), and STIR (github.com/UCL/STIR). DukeSim v1.2 is available for research use by request through the Center for Virtual Imaging Trials at Duke University (cvit.duke.edu/resource/dukesim-v1-2). The GE Duetto toolbox used for selected pilot virtual imaging trials is proprietary software and cannot be redistributed. Wrapper scripts for the integrated framework are specific to the computing environment, file-system structure, and licensed software configuration used in this study. Additional information regarding wrapper script implementation may be provided by the corresponding author upon reasonable request.