Abstract
Background
Calcium blooming artifacts, predominantly caused by the partial volume effect, may limit the accuracy of coronary stenosis assessment on coronary computed tomography angiography (CTA). Existing comparative tests or physical phantoms to evaluate de-blooming methods are costly, time-consuming, and limited in geometry and materials. This study developed a computational method to simulate CTA images incorporating calcified plaques from digital phantoms to help understand the characteristics of blooming artifacts.
Methods
Digital vascular phantoms with different plaque geometries and stenosis severities were created and imaged by computed tomography (CT) simulation. To assess the accuracy of simulation, CT images of three-dimensional (3D)-printed physical phantoms were acquired by CT scanners. To investigate the consistency of the impacts of tube voltage and reconstruction kernel on calcium blooming artifacts, both simulated and acquired CT images were quantitatively analyzed.
Results
The average structural similarity index measure (SSIM) of simulated and acquired images was 0.94±0.04. CT number, image noise and calcified area comparison between the simulated and corresponding acquired images showed good agreement. Normalized calcified area was 1.12±0.13, 1.05±0.12, 0.99±0.11 (P<0.001), for 80, 100 and 120 kV, respectively, across the simulated and acquired images. Normalized calcified area was 0.93±0.04, 0.96±0.06, 1.02±0.05, and 1.07±0.16, for reconstruction kernels from sharp to soft (P<0.001) for simulated images; result was 0.99±0.17 (sharp) vs. 1.09±0.13 (soft) for acquired images (P<0.001). Linear regression analysis indicated that measured calcified area increased as tube voltage decreased or reconstruction kernel became softer (r2>0.98) both for simulated and acquired images.
Conclusions
Simulated CT images from digital phantoms resembled the acquired images by CT scanners. Tube voltages and reconstruction kernels significantly affected the extent of calcium blooming artifacts both for simulated and acquired images. The proposed framework for calcium blooming artifacts may provide an in silico testbed for improving coronary CTA interpretation.
Keywords: Coronary artery disease, digital phantom and modeling, medical imaging simulation, vascular plaque characterization
Highlight box.
Key findings
• Simulated computed tomography (CT) images demonstrated good similarity in comparison with designed digital phantoms and acquired CT images. Tube voltages and reconstruction kernels significantly affected the extent of calcium blooming artifacts both for simulated and acquired images. In addition, the proposed framework may provide a clinically relevant in silico testbed for optimizing coronary computed tomography angiography (CTA) acquisition protocols in calcified lesions and for developing and validating de-blooming and stenosis-quantification tools before patient-based studies and regulatory evaluation.
What is known and what is new?
• Calcium blooming artifacts significantly impact the accuracy of coronary stenosis assessment in coronary CTA. However, comparative tests to evaluate the performance of de-blooming methods are time-consuming, expensive and the ground truth remains unknown.
• Computational method can simulate calcium blooming artifacts of coronary CTA imaging from digital phantoms with various plaque geometries and stenosis severities. Physical phantoms created using three-dimensional printing and scanned with actual CT can be used to assess the accuracy of simulation. Good similarity of simulated CT images with designed digital phantoms and actual scanned CT images were observed.
What is the implication, and what should change now?
• A validated in silico framework for calcium blooming artifacts could provide a clinically relevant in silico testbed for improving coronary CTA interpretation in heavily calcified lesions.
Introduction
Coronary computed tomography angiography (CTA) enables direct assessment of coronary artery stenosis and is considered the first-line test in acute and chronic coronary syndromes (1). However, the main challenge of CTA is image degradation in lesions with calcified plaques. The presence of calcified plaques introduces blooming artifacts which leads to deviation of computed tomography (CT) number of adjacent tissues, thus shadowing the visualization of adjacent tissues and exaggerating the stenosis (2,3). Various solutions have been proposed to address this issue, including protocol optimizations, hardware and post-processing improvements (4). Comparative tests with other imaging modalities or phantom studies have been performed to evaluate de-blooming methods. However, comparative tests are time-consuming, expensive and affected by reference standard (5); the geometries of physical phantoms are generally regular and the material compositions are limited (6); furthermore, the material may degrade with time, leading to inaccurate results. In contrast, digital phantoms (7) can produce anatomical shapes and materials rapidly and accurately, enabling fast and low-cost evaluation.
In this study, we aimed to develop a computational method to simulate calcium blooming artifacts of coronary CTA imaging from digital phantoms with various plaque geometries and stenosis severities. Physical phantoms were created using three-dimensional (3D) printing and scanned with actual CT to assess the accuracy of simulation. Clinically, a validated in silico framework for calcium blooming artifacts could provide a practical testbed for improving coronary CTA interpretation in heavily calcified lesions. Such a platform may enable systematic evaluation of how acquisition and reconstruction settings influence apparent stenosis severity, while also supporting the development and preclinical verification of de-blooming and stenosis-quantification algorithms before costly patient-based studies.
Methods
Design of digital phantoms and generation of simulated CT images
Twenty digital phantoms modeling vessel segments with stenosis were designed in total and saved as stereolithographic files. Structures of these phantoms (Figure 1A) included coronary vessel wall, lumen, lipidic plaque, fibrous tissue, and calcified plaque (Figure 1B). The vessel segment was designed with dimensions of 15 mm in length, 1 mm in wall thickness, and 8 mm in cross-sectional outer diameter, respectively. Five models (Model 1–5) were defined to illustrate the variability of plaques (Figure 1C), representing eccentric plaque with basal calcification (Models 1 & 2), concentric plaque with central calcification (Models 3 & 4) and fully calcified plaque (Model 5). Four stenosis levels (mild, intermediate, advanced, and severe), corresponding to total plaque area percentages of 36.5%, 55.7%, 68.6%, and 84.3%, were defined to represent the progressive narrowing of the vessel lumen, with the calcified area within these plaques ranging from 4.95 to 23.82 mm2. The mild category reflects early, non-hemodynamically significant plaque; the intermediate category represents lesions that may warrant further functional investigation; the advanced category represents significant obstructive disease; and the severe category reflects high-risk lesions approaching total occlusion.
Figure 1.
Key structures of the digital phantom. Pink stands for vessel wall, yellow for lipidic plaques, green for fibrous tissues, white for calcified plaques, and grey for coronary lumen filled with contrast medium. (A) Volumetric rendering of the phantom. (B) Central axial section with key components. (C) Models with different shapes and stenosis severity. From left to right: Model 1–5, where Models 1 and 2 represent eccentric plaque with basal calcification, Models 3 and 4 represent concentric plaque with central calcification, and Model 5 represents fully calcified plaque; from top to bottom: mild, intermediate, advanced, and severe of area stenosis severity.
Material densities and compositions (Table S1) were defined by Geant4 Application for Tomographic Emission (GATE) (8) and matched the CT number of lipidic plaque [−42 to 47 Hounsfield unit (HU)], fibrous tissue (131 to 350 HU), calcified plaque (500 to 1,200 HU), and contrast medium (250 to 400 HU), assuming a 100 kV scanning protocol (9). The material definition of fibrous tissue, lipidic plaque and calcified plaque were modified from muscle, adipose and bone in the material database of GATE (GateMaterials.db). The material definition of contrast medium was defined from the molecular formula of iobitridol (XENETIX 350), which is C20H28I3N3O9.
A general schematic of simulated CT acquisition is presented in Figure S1. Simulated projections of digital phantoms were generated using GATE (version 9.1), a toolkit that enables precise CT simulation and models a comprehensive set of physical processes. The X-ray spectrum was estimated using SPEKTR (10) (version 2.2), with the tube voltage set to 80, 100 and 120 kV. Within GATE, the photon activity was set to 5×108 becquerel, defining the total number of emitted photons. Simulations were performed using supercomputing resources (Siyuan cluster core, Shanghai Jiao Tong University, Shanghai, China) for acceleration. Nodes utilized 2 Intel Xeon ICX Platinum 8358 processors, each with 32 cores and 192 GB RAM, without implementing GPU computing. Using parallel computing, 600 projections were generated per simulated scan.
Simulated projections were reconstructed using the FDK algorithm (11), implemented via the Tomographic Iterative GPU-based Reconstruction Toolbox (TIGRE) (12). As the reconstruction kernels used in clinical CT scanners are proprietary, four custom reconstruction kernels were implemented for simulated projections, including three soft-tissue kernels (Soft 1, Soft 2, Soft 3) and one sharp kernel (Sharp). Soft 1 was designed to match the soft kernels routinely used in clinical practice. Soft 2 and Soft 3 represented kernels with progressively lower spatial resolution and more smoothing. The sharp kernel represents edge-enhancing kernels used for detailed evaluation of calcified structures.
CT numbers were calculated according to Eq. [1], where µ denotes the linear attenuation coefficient of the material.
| [1] |
Calcium blooming is influenced by multiple factors (13-15). This study was performed under static conditions, investigating the effects of acquisition parameters (e.g., tube voltage, reconstruction kernel, detector pixel size and material compositions) on measured calcified area and CT number without the confounding influence of motion (Table S2).
Fabrication of physical phantoms and acquisition of CT images
Physical phantoms were fabricated from the digital phantoms using 3D printing. The lipidic plaque and fibrous tissue were made of red and white wax, respectively. The material composition of calcified plaque was developed by testing different formulations of a calcium carbonate (CaCO3) powder and epoxy adhesive mixture. A linear relationship (16) between the proportion of CaCO3 and the measured CT number was observed across different CT vendors. In all cases, Pearson correlation coefficients r2>0.99 were obtained (Figure S2). The average CT number for coronary calcium scoring was 672 HU in 120 kV scanning (17), thus 40% CaCO3 that corresponds to 720 HU at 120 kV was selected to simulate the calcified plaque. The contrast medium iobitridol (XENETIX 350) was diluted in distilled water at a volume ratio of 1:50 to obtain a solution with an expected CT number (250–400 HU).
To assess the digital phantoms and simulation system, we performed actual CT acquisitions using scanners from three different vendors (United Imaging, Shanghai, China; Philips, Amsterdam, the Netherlands; and Siemens, Erlangen, Germany). Additionally, the exact geometry of the 3D-printed physical phantoms was determined via micro-CT scanning (uCT microPCCT, United Imaging Life Science Instrument, China) at 80 kV, 200 mA, using a soft kernel for reconstruction. The flowchart of this frame work is shown in Figure 2. Details of scan protocols are provided in Table 1.
Figure 2.
Flowchart of the proposed framework. The workflow includes digital phantom design, simulated CT generation, 3D-printed physical phantom fabrication, real CT scanning, and quantitative comparison using simulation performance evaluation metrics. 3D, three-dimensional; CT, computed tomography.
Table 1. Parameters for simulated CT and actual CT acquisition.
| Parameters | Simulation | Vendor 1 | Vendor 2 | Vendor 3 |
|---|---|---|---|---|
| Acquisition | ||||
| Model | – | uCT 780 | Brilliance eCT | SOMATOM Force |
| Tube voltage, kV | 80/100/120 | 80/100/120 | 80/100/120 | 80/100/120 |
| Tube current, mA | 5×108 beq | 496 | 493 | 282 |
| Focal spot size, mm | 0.80×0.80 | 1.00×1.00 | 0.60×0.60 | 0.80×0.80 |
| Detector pixel size, mm | 0.60×0.60 | 0.50×0.50 | 0.625×0.625 | 0.60×0.60 |
| Source-to-detector, mm | 1090.00 | 1060.20 | 1085.60 | 1113.00 |
| Source-to-isocenter, mm | 600.00 | 570.00 | 595.00 | 610.00 |
| Reconstruction | ||||
| Pixel size, mm | 0.50×0.50 | 0.50×0.50 | 0.55×0.55 | 0.39×0.39 |
| Columns and rows | 512×512 | 512×512 | 512×512 | 512×512 |
| Display field of view, mm | 256×256 | 256×256 | 281×281 | 200×200 |
| Kernel (soft) | Soft 1/Soft 2/Soft 3 | C_SOFT_BA | A | Bv44 |
| Kernel (sharp) | Sharp | B_SHARP_ZS | C | Bv59 |
Vendor 1 refers to United Imaging, Vendor 2 to Philips, and Vendor 3 to Siemens. A refers to soft kernel, and C to sharp kernel. CT, computed tomography.
Analysis method
Simulation performance evaluation metrics
The similarity between the simulated and acquired CT images was quantified using three metrics. Firstly, the structural similarity index measure (SSIM) was calculated using MATLAB (R2020b, The MathWorks Inc., Natick, MA, USA), where a higher value indicates better image similarity. Secondly, the mean CT number within a defined region of interest (ROI) was measured using ImageJ (version 1.8.0_172, National Institutes of Health, Bethesda, MD, USA) and compared between simulated and acquired images. Further, image noise was assessed by calculating the standard deviation (SD) of CT numbers within ROIs placed on the contrast medium in both the simulated and acquired images.
Impact on calcium blooming artifacts
Images were generated at varying tube voltages (80, 100 and 120 kV) and reconstructed with different kernels (soft and sharp) to evaluate their effects on calcium blooming artifacts, both for simulated and acquired images. The calcified area was measured using a standardized approach. First, the region was delineated by applying a kV-specific threshold. The area was then calculated as the product of the pixel size and the number of pixels within the delineated region using ImageJ. The designed calcified area varied across models (Model 1–5) and stenosis severities (mild to severe); for direct comparison, the measured calcified area was expressed using a measure-to-actual ratio (MTAR). The MTAR was defined in Eq. [2] as the delineated calcified area divided by its corresponding ground truth, which was the designed value for simulated images, and the geometry measured by micro-CT for images acquired from three vendors (Table S3).
| [2] |
Statistical analysis
For the quantitative analysis, a single central slice was selected from each phantom. All analyses were performed on the phantom level, with each phantom contributing one slice. Continuous variables were tested for normality using the Shapiro-Wilk test and are presented as mean ± SD. Linear regression analyses were performed to assess the following relationships: (I) CT number vs. CaCO3 proportion; (II) CT number vs. tube voltage; (III) measured calcified area vs. tube voltage; (IV) measured calcified area vs. reconstruction kernel; and (V) image noise vs. kernel sharpness. Independent samples t-tests were performed to compare: (I) the CT numbers of calcified plaques between simulated and acquired images; (II) the image noise between simulated and acquired images; (III) the measured calcified area between simulated and micro-CT images; and (IV) image noise between soft and sharp kernels. One sample t-test was used to analyze the calcified area (MTAR) of simulated images in comparison with the designed value (ground truth). One-way analysis of variance was conducted to investigate the impacts of tube voltages and reconstruction kernels on measured calcified area, followed by Bonferroni correction for multiple comparisons. The equality of SDs of the calcified area (MTAR) was assessed using F-tests for comparisons between (I) different stenosis levels and (II) actual scans and simulations, respectively.
All statistical analyses were performed using MedCalc Statistical Software (version 22.001, MedCalc Software Ltd., Ostend, Belgium). A two-sided P value of <0.05 was considered statistically significant.
Results
Baseline characteristics
For digital phantoms, projections were simulated at three tube voltages (80, 100 and 120 kV) using GATE. For physical phantoms, corresponding CT images were acquired at these voltages by three different vendors. All projections were reconstructed using both soft and sharp kernels. Consequently, for each modality (simulated and acquired), a complete image set encompassing all reconstruction and voltage combinations was obtained for the twenty phantoms.
Simulation performance
Calcified area comparison between simulated and ground truth
Calcified area measurement of simulated images (80 kV and kernel Soft 1) was analyzed to assess the consistency with ground truth (Table S3). A one-sample t-test indicated that the average calcified area (MTAR) from simulated images did not significantly differ from the designed value [0.99±0.05 vs. 1, P=0.54; 95% confidence interval (CI): 0.97–1.02]. An independent samples t-test showed no significant difference in calcified area (MTAR) between the simulated images and micro-CT measurements (mean 0.99±0.05 vs. 1.00±0.14, P=0.77).
SSIM between simulated and acquired CT images
The SSIM value between the simulated images and the acquired images from Vendor 1, Vendor 2, and Vendor 3 was 0.95±0.02, 0.95±0.05 and 0.92±0.01 (mean 0.94±0.04), across all tube voltages and reconstruction kernels, indicating a high similarity between the simulated and acquired images (Figure 3).
Figure 3.
Model 5 (fully calcified plaque) at different stenosis and tube voltages (kernel Soft 1). (A-D) Area stenosis levels from mild to severe, where (A) represents mild, (B) represents intermediate, (C) represents advanced, and (D) represents severe. (A-D) From top to bottom: 80, 100 and 120 kV; from left to right: acquired images from three vendors and simulated images, where Vendor 1 refers to United Imaging, Vendor 2 to Philips, and Vendor 3 to Siemens. The window level and window width are set at 200 and 600 HU, same for all images.
CT number comparison between simulated and acquired images
The average CT number of calcified plaques for simulated and acquired images was 1005±65 vs. 1096±94 HU (P=0.33), 918±59 vs. 854±109 HU (P=0.07), and 885±104 vs. 818±108 HU (P=0.09), respectively for 80, 100 and 120 kV (Table S4). Linear regression analysis indicated an inverse relationship between tube voltage and CT number (r2=0.98). This relationship was also observed for contrast medium and fibrous tissue. However, CT number for lipids increased with tube voltage, showing an opposite trend (Figure S3).
Image noise comparison between simulated and acquired CT images
Image noise between the simulated and acquired images showed no significant difference (mean 7.13±1.35 vs. 6.79±2.47 HU, P=0.83 with a soft kernel; mean 23.06±1.30 vs. 18.83±6.31 HU, P=0.29 with a sharp kernel). Details are shown in Table S5.
Impact on calcium blooming artifacts
Impact of tube voltages
For simulated images generated at 80, 100 and 120 kV with a soft reconstruction kernel, calcified area (expressed using MTAR as the delineated calcified area divided by the designed value) was 1.00±0.05, 0.95±0.05, 0.92±0.05 (adjusted P<0.05); for acquired images, calcified area (expressed using MTAR as the delineated calcified area divided by micro-CT measurements) was 1.16±0.13, 1.09±0.12, 1.02±0.11 (adjusted P<0.01). Across all simulated and acquired images, calcified area (MTAR) was 1.12±0.13, 1.05±0.12, 0.99±0.11 (adjusted P<0.01), for 80, 100 and 120 kV with a soft reconstruction kernel (Table 2). Linear regression analysis indicated a negative correlation between tube voltages and calcified area (r2=0.99), showing that higher tube voltages led to a decreased measured calcified area, both for simulated and acquired images. The impact of tube voltages on calcium measurement is shown in Figure 4; the box-and-whisker graphs of calcified area against tube voltages are displayed in Figure S4.
Table 2. Statistics of calcified area (MTAR) across different tube voltages.
| Tube voltage | Kernel | Acquired images | Simulation | Total | |||
|---|---|---|---|---|---|---|---|
| Vendor 1 | Vendor 2 | Vendor 3 | All vendors | ||||
| 80 kV | Soft | 1.21±0.14 | 1.15±0.09 | 1.12±0.13 | 1.16±0.13 | 1.00±0.05 | 1.12±0.13 |
| Sharp | 1.18±0.16 | 1.13±0.16 | 1.10±0.11 | 1.13±0.15 | 0.96±0.04 | 1.08±0.15 | |
| 100 kV | Soft | 1.17±0.13 | 1.05±0.07 | 1.05±0.11 | 1.09±0.12 | 0.95±0.05 | 1.05±0.12 |
| Sharp | 1.01±0.13 | 0.90±0.11 | 1.01±0.14 | 0.97±0.15 | 0.93±0.03 | 0.96±0.13 | |
| 120 kV | Soft | 1.09±0.11 | 0.98±0.08 | 0.98±0.10 | 1.02±0.11 | 0.92±0.05 | 0.99±0.11 |
| Sharp | 0.92±0.13 | 0.83±0.08 | 0.90±0.12 | 0.89±0.12 | 0.90±0.03 | 0.89±0.10 | |
Data are presented as mean ± standard deviation. Vendor 1 refers to United Imaging, Vendor 2 to Philips, and Vendor 3 to Siemens. MTAR, measure-to-actual ratio.
Figure 4.
Impact of tube voltages on calcified area (kernel Soft 3). (A,B) Model 2 (eccentric plaque with basal calcification) at mild and intermediate stenosis severity. (C,D) Model 4 at advanced and severe stenosis severity. MTAR, measure-to-actual ratio.
Impact of reconstruction kernels
For simulated images, calcified area (MTAR) was 0.93±0.04, 0.96±0.06, 1.02±0.05, and 1.07±0.16, for Sharp and Soft 1/2/3 kernel, respectively (adjusted P<0.01), see Table 3. In comparison, the result was 0.99±0.17 (sharp), and 1.09±0.13 (soft) for acquired images (P<0.001), see Table 2. Linear regression analysis indicated a negative correlation between reconstruction kernels and calcified area, both for simulated and acquired images (r2=0.98), showing that calcified area approached ground truth as the kernel became sharper. Simulated images with different kernels are shown in Figure 5; the impact of kernels on calcium measurement is shown in Figure S5; the box-and-whisker graphs of calcified area against kernels are displayed in Figure 6.
Table 3. Statistics of calcified area (MTAR) across different reconstruction kernels in simulated images.
| Kernel | 80 kV | 100 kV | 120 kV |
|---|---|---|---|
| Sharp | 0.96±0.04 | 0.93±0.03 | 0.90±0.03 |
| Soft 1 | 0.99±0.05 | 0.95±0.05 | 0.92±0.05 |
| Soft 2 | 1.05±0.06 | 1.02±0.05 | 1.00±0.03 |
| Soft 3 | 1.12±0.08 | 1.08±0.08 | 1.01±0.24 |
Data are presented as mean ± standard deviation. MTAR, measure-to-actual ratio.
Figure 5.
Simulated images of models at different stenosis severities and reconstruction kernels (100 kV). (A-D) Area stenosis levels from mild to severe, where (A) represents mild, (B) represents intermediate, (C) represents advanced, and (D) represents severe. (A-D) From top to bottom row: GT and simulated images with kernel Soft 3/2/1 and sharp; for GT, the calcified plaque is colored in white, accompanied by lipidic plaque in dark gray, fibrous tissue in medium gray, and contrast medium in light gray; from left to right: Model 1–5, where Models 1 and 2 represent eccentric plaque with basal calcification, Models 3 and 4 represent concentric plaque with central calcification, and Model 5 represents fully calcified plaque. The window level and width are set at 200 and 600 HU, same for all images. GT, ground truth.
Figure 6.
The MTAR of calcified area against reconstruction kernels for simulated images. The tube voltage is set as 80, 100 and 120 kV for (A-C), respectively. MTAR, measure-to-actual ratio.
Discussion
In this study, a novel method was developed to simulate CT images with calcium blooming artifacts from a series of digital vascular phantoms and was evaluated using acquired images from different CT vendors. The consistency of the impact of X-ray tube voltage and reconstruction kernel on calcium blooming severity was also investigated between the simulated and acquired CT images. The following main findings were observed: (I) the simulated CT images (without blooming artifacts) closely resembled the designed digital phantoms. Under matched scanning and reconstruction conditions, the simulated images also closely matched the acquired CT images; (II) both simulated and acquired images showed similar responses to changes in acquisition parameters: higher X-ray tube voltages and sharper kernels reduce calcium blooming artifacts; (III) simulation of calcium blooming artifacts from digital phantom demonstrated good consistency with images acquired by different CT vendors.
The diagnostic accuracy of coronary CTA is influenced by calcium blooming artifacts, and various de-blooming methods have been developed (4). Validation depends on physical phantoms or comparative tests with other imaging modalities such as optical coherence tomography (5). However, physical phantoms face challenges in simulating complex anatomy structures (6). In comparison, digital phantoms are feasible to produce complex and diverse anatomy structures and can be extended by adding surrounding tissue. Validation with other modalities is time-consuming, affected by selection of reference standard, and would introduce systemic error associated with cross-modality misalignment, larger acquisition spacing and acquisition process (5). In contrast, the ground truth of digital phantom is precisely known, making it more conducive for fair comparison and verification of plaque quantification. By simulating various and even extreme clinical situations, digital phantom provides controlled investigation. Thus, computational modeling and simulation based on digital phantoms have been an alternative selection to optimize or validate coronary CTA imaging and analysis.
However, computational modeling and simulation based on digital phantoms exhibit shortcomings such as simplified geometries or patient-derived models without modification (17) and assessing multiple image quality factors (18,19). This study designed more dedicated digital vascular phantoms and demonstrated consistent blooming severity compared with acquired CT images. Moreover, the open-source Monte Carlo simulation platform (GATE) provides transparent, reproducible and accurate modeling. Verified against acquired images from physical phantoms by multiple metrics, this framework could help investigate the impact of scanning protocol on calcium blooming artifacts, thus has the potential on simulating various clinical situations and facilitating de-blooming related investigations.
Simulation performance
Extreme values of calcified area appeared at acquired images rather than simulated images (Figure S4). Of note, SD of calcified area (MTAR) decreased with stenosis level increased (SD =0.18 for mild vs. SD =0.13 for other three stenosis levels, F=1.99, P=0.047), indicating that smaller size plaques were challenging to reconstruct and delineate ROIs (Table 2 and Table S3). An F-test showed that the calcified area (MTAR) from actual scans (SD =0.15) were significantly more variable than those from simulation (SD =0.05), F=7.49, P<0.001 (Table 2 and Table S3). Observations above demonstrates that the image quality of digital phantom and simulation is more stable compared with physical phantom and actual scans. In clinical practice, inconsistent calcium classification and scanning protocols were observed (20), which could affect quantification and introduce inter-observer variability. By avoiding modeling errors including material degradation and geometry inaccuracy, and acquisition errors such as aberration, improper positioning, and motion artifacts, the proposed method has the potential to further help develop a consistent measurement and decrease the variability.
The impact factors of blooming artifacts
The blooming artifacts decreased with the increase in tube voltage and sharpness of reconstruction kernel both for simulated and acquired images, which is consistent with previous studies (14,21). The use of a sharper reconstruction kernel enhances high-frequency components, thereby increasing the effective spatial resolution and improving both the edge clarity of calcified plaques and the detectability of small calcified plaques. The SD of calcified area (MTAR) was observed to decrease as the reconstruction kernel became sharper (Table 3), demonstrating a more concentrated measurement. Image noise differed significantly between soft and sharp kernels (P<0.05 across the acquired and simulated images), and increased linearly with kernel sharpness (r2>0.99), see Table S5. Since the finite detector resolution could lead to calcium blooming, the simulated images were observed to be blurring greatly as the detector pixel size becomes larger (Figure S6). In clinical practice, higher tube voltage, sharper kernel and kV-specific delineation threshold can help better observe the calcified lesions.
Previous study (22) has demonstrated that bone and iodine exhibited similar linear attenuation coefficient at approximately 100 kV and discriminated at other tube voltages, although their material densities differ a lot. This study also observed that different combinations of material compositions achieved similar CT number at 120 kV, but differed at 80 and 100 kV (Figure S7), showing that the same attenuation value can be obtained from two different materials at a given energy. This is consistent with the idea that dual-energy CT can help determine atherosclerotic plaque composition by using different energies.
The development of digital phantoms in this study is the first step toward creating models that accurately represent clinical population. The digital phantom framework established in this study lays the foundation for future clinical investigation in several ways. Firstly, the ground truth of data obtained from digital phantoms are precisely known (e.g., calcification size and morphology), thus they can be utilized for the development (23), optimization and verification (24) of stenosis quantification algorithms and related medical products. By producing preliminary evidence prior to clinical patient data, this framework is a promising in silico-based approach at a low cost. Secondly, by generating large-scale and diverse training datasets with known ground truth, digital phantoms can help improve the generalizability of deep learning-based artifact reduction algorithms in real patient scans. Thirdly, by collaborating with clinical experts to evaluate the simulated images with blooming artifacts generated by different acquisition parameters, this framework quantifies how these parameters influence stenosis severity assessment and can help optimize scan protocols in clinical practice. Furthermore, the gap between this phantom study and clinical studies quantifies the impact of patient-specific factors (e.g., cardiac motion and calcification morphology) on blooming artifact.
Moreover, this study has several practical clinical implications. Firstly, because the geometry and composition of the calcified plaque are known a priori, this framework provides a controlled reference standard for testing how calcium blooming alters apparent lumen size and stenosis severity on coronary CTA. This is clinically relevant because overestimation of stenosis in heavily calcified lesions may affect downstream decisions regarding additional functional testing, invasive angiography, or revascularization.
Secondly, the present results show that tube voltage and reconstruction kernel systematically influence the magnitude of blooming. Therefore, this platform could be used to optimize acquisition and reconstruction strategies for calcified coronary disease in a reproducible manner, potentially improving consistency of CTA interpretation across scanners and centers.
Thirdly, the proposed digital-phantom framework may serve as an in silico development and benchmarking environment for de-blooming, plaque characterization, and stenosis-quantification algorithms, including deep learning approaches. By allowing large numbers of labeled cases with known ground truth to be generated at low cost, it can help accelerate technical development before clinical validation in patient cohorts.
Although this proof-of-concept study does not yet establish direct patient-level benefit, it provides an important translational bridge between basic imaging physics and clinically deployable tools aimed at improving the assessment of calcified coronary lesions on CTA.
Limitations
This work has several limitations. Firstly, the simulated images do not perfectly match actual CT in detector pixel size, geometry distance and reconstruction kernel. Further optimization for a precise simulation platform is necessary.
Secondly, this study was performed in homogeneous surrounding environments without the confounding effects of complex anatomy. In clinical coronary CTA, surrounding structures such as myocardium, epicardial fat, and the chest wall contribute additional attenuation, scatter, and image noise, which may alter the appearance and severity of blooming artifacts. Therefore, although the present model is suitable for controlled assessment of the directional effects of acquisition and reconstruction parameters, it does not fully reproduce the in vivo imaging environment, and the magnitude of the observed effects should not be assumed to translate directly to patient scans (20).
Thirdly, this study did not account for cardiac or respiratory motion. In clinical practice, motion can blur plaque boundaries, interact with partial-volume effects, and further increase variability in stenosis assessment. As a result, the relationships observed here between tube voltage, reconstruction kernel, and blooming artifact are likely more stable than those encountered in vivo. The present findings should therefore be interpreted as mechanistic and proof-of-concept rather than as a complete representation of clinical coronary CTA performance.
Moreover, the assessment performed in this study was not fully independent. The 3D-printed physical phantoms were derived from the same digital models used for simulation. Therefore, the observed agreement between simulated and acquired images primarily demonstrates internal consistency of the overall modeling, printing, and imaging pipeline rather than external validation against an independent reference standard. Patient-specific factors, including motion, vessel curvature, heterogeneous surrounding tissues, and variable attenuation, may attenuate the strength of these correlations in clinical datasets. Future studies incorporating more realistic anatomy and motion models, followed by validation in patient-based coronary CTA cohorts, are needed to establish the full clinical applicability of this framework.
Conclusions
Simulated CT images demonstrated good similarity in comparison with designed digital phantoms and acquired CT images. Tube voltages and reconstruction kernels significantly affected the extent of calcium blooming artifacts both for simulated and acquired images. In addition, the proposed framework may provide a clinically relevant in silico testbed for optimizing coronary CTA acquisition protocols in calcified lesions and for developing and validating de-blooming and stenosis-quantification tools before patient-based studies and regulatory evaluation.
Supplementary
The article’s supplementary files as
Acknowledgments
None.
Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
Footnotes
Funding: This study was supported by the National Key Research and Development Program of China (No. 2023YFC2506503) and the National Natural Science Foundation of China (Nos. 81871460 and 82020108015).
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://cdt.amegroups.com/article/view/10.21037/cdt-2026-0156/coif). J.H. reports funding from the Young Scientists Fund of the National Natural Science Foundation of China (No. 32501154) and the China Postdoctoral Science Foundation (No. 2025M772896). S.T. reports funding from the National Key Research and Development Program of China (No. 2023YFC2506503) and National Natural Science Foundation of China (Nos. 81871460 and 82020108015). He is also a co-founder of Pulse Medical, from which he has received research grants and for which he has served as a consultant. The other authors have no conflicts of interest to declare.
Data Sharing Statement
Available at https://cdt.amegroups.com/article/view/10.21037/cdt-2026-0156/dss
References
- 1.Cury RC, Leipsic J, Abbara S, et al. CAD-RADS™ 2.0 - 2022 Coronary Artery Disease - Reporting and Data System An Expert Consensus Document of the Society of Cardiovascular Computed Tomography (SCCT), the American College of Cardiology (ACC), the American College of Radiology (ACR) and the North America Society of Cardiovascular Imaging (NASCI). Radiol Cardiothorac Imaging 2022;4:e220183 . 10.1148/ryct.220183 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Kalisz K, Buethe J, Saboo SS, et al. Artifacts at Cardiac CT: Physics and Solutions. Radiographics 2016;36:2064-83. 10.1148/rg.2016160079 [DOI] [PubMed] [Google Scholar]
- 3.Pack JD, Xu M, Wang G, et al. Cardiac CT blooming artifacts: clinical significance, root causes and potential solutions. Vis Comput Ind Biomed Art 2022;5:29 . 10.1186/s42492-022-00125-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Weir-McCall JR, Wang R, Halankar J, et al. Effect of a calcium deblooming algorithm on accuracy of coronary computed tomography angiography. J Cardiovasc Comput Tomogr 2020;14:131-6. 10.1016/j.jcct.2019.07.007 [DOI] [PubMed] [Google Scholar]
- 5.Faris O, Shuren J. An FDA Viewpoint on Unique Considerations for Medical-Device Clinical Trials. N Engl J Med 2017;376:1350-7. 10.1056/NEJMra1512592 [DOI] [PubMed] [Google Scholar]
- 6.Horiguchi J, Fujioka C, Kiguchi M, et al. In vitro measurement of CT density and estimation of stenosis related to coronary soft plaque at 100 kV and 120 kV on ECG-triggered scan. Eur J Radiol 2011;77:294-8. 10.1016/j.ejrad.2009.08.002 [DOI] [PubMed] [Google Scholar]
- 7.Sauer TJ, Buckler AJ, Abadi E, et al. Development of physiologically-informed computational coronary artery plaques for use in virtual imaging trials. Med Phys 2024;51:1583-96. 10.1002/mp.16959 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Jan S, Benoit D, Becheva E, et al. GATE V6: a major enhancement of the GATE simulation platform enabling modelling of CT and radiotherapy. Phys Med Biol 2011;56:881-901. 10.1088/0031-9155/56/4/001 [DOI] [PubMed] [Google Scholar]
- 9.Papadopoulou SL, Neefjes LA, Garcia-Garcia HM, et al. Natural history of coronary atherosclerosis by multislice computed tomography. JACC Cardiovasc Imaging 2012;5:S28-37. 10.1016/j.jcmg.2012.01.009 [DOI] [PubMed] [Google Scholar]
- 10.Siewerdsen JH, Waese AM, Moseley DJ, et al. Spektr: a computational tool for x-ray spectral analysis and imaging system optimization. Med Phys 2004;31:3057-67. 10.1118/1.1758350 [DOI] [PubMed] [Google Scholar]
- 11.Feldkamp L, Davis L, Kress J. Practical Cone-Beam Algorithm. J Opt Soc Am 1984;1:612-619. [Google Scholar]
- 12.Biguri A, Dosanjh M, Hancock S. TIGRE: a MATLAB-GPU toolbox for CBCT image reconstruction. Biomed Phys Eng Express 2016;2:055010. [Google Scholar]
- 13.Cademartiri F, La Grutta L, Runza G, et al. Influence of convolution filtering on coronary plaque attenuation values: observations in an ex vivo model of multislice computed tomography coronary angiography. Eur Radiol 2007;17:1842-9. 10.1007/s00330-006-0548-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Achenbach S, Boehmer K, Pflederer T, et al. Influence of slice thickness and reconstruction kernel on the computed tomographic attenuation of coronary atherosclerotic plaque. J Cardiovasc Comput Tomogr 2010;4:110-5. 10.1016/j.jcct.2010.01.013 [DOI] [PubMed] [Google Scholar]
- 15.Fan Y, Pack J, De Man B. A virtual imaging trial framework to study cardiac CT blooming artifacts. Proc. SPIE 12304, 7th International Conference on Image Formation in X-Ray Computed Tomography, 1230415 (17 October 2022). Available online: https://doi.org/ 10.1117/12.2646407 [DOI] [Google Scholar]
- 16.Hoy CF, Naguib HE, Paul N. Fabrication and control of CT number through polymeric composites based on coronary plaque CT phantom applications. J Med Imaging (Bellingham) 2016;3:016001 . 10.1117/1.JMI.3.1.016001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Bischoff B, Kantert C, Meyer T, et al. Cardiovascular risk assessment based on the quantification of coronary calcium in contrast-enhanced coronary computed tomography angiography. Eur Heart J Cardiovasc Imaging 2012;13:468-75. 10.1093/ejechocard/jer261 [DOI] [PubMed] [Google Scholar]
- 18.Samei E, Richards T, Segars WP, et al. Task-dependent estimability index to assess the quality of cardiac computed tomography angiography for quantifying coronary stenosis. J Med Imaging (Bellingham) 2021;8:013501 . 10.1117/1.JMI.8.1.013501 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Zarei M, Abadi E, Segars WP, et al. Coronary stenosis quantification in cardiac computed tomography angiography: multi-factorial optimization of image quality and radiation dose. J Med Imaging (Bellingham) 2023;10:063502 . 10.1117/1.JMI.10.6.063502 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Williams MC, Earls JP, Hecht H. Quantitative assessment of atherosclerotic plaque, recent progress and current limitations. J Cardiovasc Comput Tomogr 2022;16:124-37. 10.1016/j.jcct.2021.07.001 [DOI] [PubMed] [Google Scholar]
- 21.Takagi H, Leipsic JA, Indraratna P, et al. Association of Tube Voltage With Plaque Composition on Coronary CT Angiography: Results From PARADIGM Registry. JACC Cardiovasc Imaging 2021;14:2429-40. 10.1016/j.jcmg.2021.07.011 [DOI] [PubMed] [Google Scholar]
- 22.Tsurusaki M, Sofue K, Hori M, et al. Dual-Energy Computed Tomography of the Liver: Uses in Clinical Practices and Applications. Diagnostics (Basel) 2021;11:161 . 10.3390/diagnostics11020161 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Pederzani G, Czechowicz K, Ghorab N, et al. The Use of Digital Coronary Phantoms for the Validation of Arterial Geometry Reconstruction and Computation of Virtual FFR. Fluids 2022;7:201. [Google Scholar]
- 24.Li P, Xu L, Yang L, et al. Blooming Artifact Reduction in Coronary Artery Calcification by A New De-blooming Algorithm: Initial Study. Sci Rep 2018;8:6945 . 10.1038/s41598-018-25352-5 [DOI] [PMC free article] [PubMed] [Google Scholar]