Virtual clinical trial for quantifying the effects of beam collimation and pitch on image quality in computed tomography

Abstract.

Purpose: To utilize a virtual clinical trial (VCT) construct to investigate the effects of beam collimation and pitch on image quality (IQ) in computed tomography (CT) under different respiratory and cardiac motion rates.

Approach: A computational human model [extended cardiac-torso (XCAT) phantom] with added lung lesions was used to simulate seven different rates of cardiac and respiratory motions. A validated CT simulator (DukeSim) was used in this study. A supplemental validation was done to ensure the accuracy of DukeSim across different pitches and beam collimations. Each XCAT phantom was imaged using the CT simulator at multiple pitches (0.5 to 1.5) and beam collimations (19.2 to 57.6 mm) at a constant dose level. The images were compared against the ground truth using three task-generic IQ metrics in the lungs. Additionally, the bias and variability in radiomics (morphological) feature measurements were quantified for task-specific lung lesion quantification across the studied imaging conditions.

Results: All task-generic metrics degraded by 1.6% to 13.3% with increasing pitch. When imaged with motion, increasing pitch reduced motion artifacts. The IQ slightly degraded (1.3%) with changes in the studied beam collimations. Patient motion exhibited negative effects (within 7%) on the IQ. Among all features across all imaging conditions studies, compactness2 and elongation showed the largest ($-26.5\%$, 7.8%) and smallest ($-0.8\%$, 2.7%) relative bias and variability. The radiomics results were robust across the motion profiles studied.

Conclusions: While high pitch and large beam collimations can negatively affect the quality of CT images, they are desirable for fast imaging. Further, our results showed no major adverse effects in morphology quantification of lung lesions with the increase in pitch or beam collimation. VCTs, such as the one demonstrated in this study, represent a viable methodology for experiments in CT.

1. Introduction

In computed tomography (CT), voluntary or involuntary movements of a patient cause motion artifacts resulting in degradation of image quality (IQ). These artifacts are more pronounced in applications such as cardiovascular CT, pulmonary CT, or when imaging trauma, uncooperative, or pediatric patients. Therefore, a major effort in CT research and development has been focused on the reduction of motion artifacts by accelerating the image acquisition.^1,2

In particular, the introduction of spiral^3–6 and multirow detector^7–9 technologies has significantly decreased the CT acquisition time. While desirable for fast acquisitions, increasing the beam collimation (i.e., increasing the number of detector rows or the cone angle) could negatively affect the IQ due to increased scatter and helical cone beam artifacts.^10,11 Similarly, increasing the pitch could also negatively affect the IQ due to undersampling.^2,12

To ascertain the appropriate use of collimation and pitch values, it is essential to quantify the effects of these two geometrical parameters on IQ under different patient motion attributes. Such a study requires (1) repetitive acquisitions of the same patients under different parameters (not ethically attainable in clinical images due to radiation concerns) and (2) the full knowledge of the patient’s dynamic anatomy (not known in clinical cases). Alternatively, realistic computer simulations [known as virtual clinical trials (VCTs) or in silico imaging] can be used to answer such clinically relevant questions.¹³

In a VCT framework, computational phantoms (modeling the human anatomy and physiology) are used as virtual patients. The phantoms are imaged using algorithms that accurately simulate the given imaging modality, instrumentation, and system configuration as well as the physics of the imaging process.¹³ In VCTs, the ground truth is known from the mathematically defined phantoms, repetitive studies at multiple imaging conditions are readily possible, and Institutional Review Board approvals are not needed. Therefore, VCTs are more flexible, faster, and cost-effective compared to conventional clinical trials.

In this work, we utilized a realistic VCT construct to investigate the effects of beam collimation and pitch on IQ in CT under different respiratory and cardiac motion rates. Our study included task-generic quality assessments in the lungs as well as bias and variability assessments in radiomics quantifications of lung lesions.

2. Materials and Methods

The overall framework of this VCT study is shown in Fig. 1. In summary, different cardiac and respiratory motions were applied to a realistic human model, which was then imaged using a scanner-specific CT simulator at multiple beam collimation and pitch values. The virtual reconstructed images were compared against the known ground truth in a task-generic and task-specific manner to evaluate the effects of collimation and pitch on IQ. The quantitative analysis was done in the lung regions, the organ of interest in this study.

Fig. 1.

The framework of the virtual clinical trial study.

2.1. Human Model

2.1.1. Textured extended cardiac-torso phantom

An extended cardiac-torso (XCAT) computational phantom¹⁴ was used as the human model. This phantom (adult male, 38 years old, 173 cm height, 66 kg weight) was created based on the segmentation of a patient CT dataset and included highly detailed organ anatomies. The phantom was further enhanced by modeling intraorgan structures of lung nonparenchyma, parenchyma, and trabecular bones using the algorithms developed in our previous studies.^15–17 The lung nonparenchyma tissues were arteries, veins, and airways. The arteries and veins were filled with blood, and airways were filled with air with airway walls modeled as muscles. The lung parenchyma tissues included secondary pulmonary lobules with a heterogeneous background inside them. Following the method by Petoussi-Henss et al.,¹⁸ the lung parenchyma tissue was modeled based on international commission on radiological protection definitions and binned into 14 different density values (0.15 to $1.00{\mathrm{g}/\mathrm{cm}}^3$) covering the densities typically observed in lung parenchyma. In this study, the phantom was modeled in a voxelized format with voxel size of 0.25 mm.

To study the effects of collimation and pitch on lesion quantification, four spiculated lesions were created based on the algorithm developed in Ref. 19 and inserted into the XCAT phantom at multiple locations within the lungs. The core of the lesions was created using a stochastic Gaussian random sphere model and the spicules were created using an iterative fractal branching algorithm.¹⁹ The volume of the four lesions was $470\pm 31{\mathrm{mm}}^3$. The lesions were deformed and moved due to the respiratory and cardiac motions. The lesions were modeled as homogenous solid lesions with a typical density of $1.06{\mathrm{g}/\mathrm{cm}}^3$. Figure 2 shows an example of how the lesions deformed during a full respiration with a heart rate of $90\text{beats}/\min$ and respiratory rate of $12\text{breaths}/\min$.

Fig. 2.

(a) Axial, (b) coronal, and (c) sagittal slices of the XCAT phantom defined at end-expiration (left) and end-inspiration (right) with the respiratory rate at 12 breaths/minute and the heart rate at $90\text{beats}/\min$. Color represents different tissue types.

2.1.2. Cardiac and respiratory models

To study the effects of motion, the phantom was rendered at multiple respiratory and heart rates. The respiratory motion model was derived from multiple sets of 4D respiratory-gated CT data of normal patients.²⁰ Using this model, one is able to simulate different motions with varying respiratory rate, diaphragm motion, chest expansion, and cardiac motion due to breathing.²⁰ The respiratory model was applied to the intraorgan heterogeneities using a B-spline smoothing process to convert the existing sparse motion vector fields (MVFs) into dense MVFs.²¹

The cardiac motion model was derived from sets of 4D cardiac-gated CT and tagged MRI data of normal patients.²⁰ Similar to the respiratory model, the cardiac motion model is parameterized with parameters, including heart rate, ejection fraction, longitudinal and radial contraction, and cardiac twist.

In this study, heart and respiratory rates were the only parameters that were varied. The phantom was setup to simulate seven combinations of heart and respiratory rates listed in Table 1, including no motion as well as a clinically realistic range of slow to rapid respiratory and heart rates, as observed in normal subjects.²⁰ For all the seven models, other motion parameters were chosen to be typical values observed in cardiac and respiratory-gated imaging data of normal subjects (ejection fraction: 60%, cardiac twist: 14 deg, longitudinal contraction: 13 mm, radial contraction: 20%, and diaphragm motion: 2 cm) suggested by Segars et al.²⁰ An example of the XCAT phantom is shown in Fig. 2 with the heart and respiratory rates, respectively, at $90\text{beats}/\min$ and $12\text{breaths}/\min$ at end-expiration and end-inspiration respiratory phases.

Table 1.

Heart and respiratory rates used in this study. The rates are in beats/breaths per minute.

Motion models	Heart rate (beats/min)	Respiratory rate (breaths/min)
1	0	0
2	60	0
3	90	0
4	120	0
5	90	8
6	90	12
7	90	16

2.2. Imaging Model

2.2.1. CT simulator and a validation study

The phantoms were imaged using a validated, scanner-specific CT simulator,^22,23 called DukeSim. DukeSim is a hybrid simulator that generates CT projection images of any voxel-based computational phantom considering the geometry and physics of a specific scanner. It accounts for the primary (using ray-tracing) and scatter (using Monte Carlo) signals, polyenergetic source spectra, bowtie filter, focal spot wobbling (also known as flying focal spot), axial/helical imaging, antiscatter grids, polyenergetic detector response, quantum and electronic noise, and CT preprocessing techniques prior to reconstruction (i.e., water calibration and beam hardening corrections). For the scatter estimations, we used a GPU-based Monte Carlo module, originally developed by the Food and Drug Administration,²⁴ with scanner-specific features implemented in a previous study.²⁵ The Monte Carlo module used the same phantom that was used in the ray-tracing module but with a lower resolution (1.0 mm voxel size versus 0.25 mm used for ray-tracing).

To image the phantoms with motion, DukeSim assigned time tags to each projection image as a function of rotation time and the number of projections per rotation. To simulate each projection image, the simulator loaded the phantom at the corresponding calculated time.

DukeSim has been previously validated in terms of IQ using a fixed pitch and beam collimation. Since the current study was focused on imaging at different pitch values and beam collimations, a supplemental validation study was performed under these conditions. To do so, a water phantom and a cylindrical phantom with iodine insert ($8.5\mathrm{mgI}/\mathrm{mL}$) were imaged. The water phantom was a cylinder with a diameter of 20 cm. The cylindrical phantom was 23 cm in diameter with inserts (diameter of 2.54 cm) that were 4.2 cm away from the center. The acquisitions were done using a commercial CT scanner (Siemens Definition Flash, Forchheim, Germany) varying beam collimation and pitch values while other parameters were constant. Similar to the actual image acquisition, a counterpart, computational water phantom, and a cylindrical phantom with iodine were imaged using DukeSim, where DukeSim was set up to model the geometry and physics of the same scanner.

The pitch was varied from 0.5 to 1.5 within increments of 0.25 with the beam collimation being set at 38.4 mm and tube voltage at 120 kV. The beam collimation was set to 6.0, 12.0, and 38.4 mm with the pitch being set at 1.0 and tube voltage at 120 kV. These were the only available collimation values on the actual scanner with other parameters being fixed. All real and simulated acquired sinograms were reconstructed using the manufacturer reconstruction software (ReconCT, Siemens Healthcare). The reconstruction algorithm was filtered backprojection algorithm with the “B31f” kernel, an in-plane pixel size of 0.48 mm, and a slice thickness of 0.6 mm. The CT numbers for water and iodine as well as image noise were calculated for both real and simulated images. The image noise was calculated in the uniform regions of the water phantom by putting circular ROIs (radius of 30 mm with a 50-mm offset from center, located at the top, bottom, right, and left) and measuring the standard deviations across 20 center slices. The real and simulated measurements were compared against each other.

2.2.2. Image acquisitions for the virtual clinical trial

The XCAT phantoms were imaged based on the geometry and detector physics of the same scanner (Definition Flash) that validations were done. To study the effects of pitch, beam collimation was kept constant at 38.4 mm with pitch varying from 0.50 to 1.50 in increments of 0.25. To study the effects of beam collimation, a constant pitch of 1.0 was used with collimation varying from 19.2 mm to 57.6 mm (at the isocenter) in increments of 9.6 mm. It should be noted that the beam collimations wider than 38.4 mm are hypothetical values and are not available in the actual scanner.

The images were acquired at a constant mAs value of 200 per slice (also known as effective mAs), a tube voltage of 120 kV, and a rotation time of 0.5 s. Similar to the validation study, the projection images were reconstructed using the manufacturer ReconCT. All images were reconstructed with a filtered backprojection algorithm with the “B31f” kernel, an in-plane pixel size of 0.48 mm, and a slice thickness of 0.6 mm.

2.3. Quality Assessment

2.3.1. Task-generic assessment

The quality of the XCAT CT images was evaluated by measuring three IQ metrics of root mean squared error (RMSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM).²⁶ These metrics quantify the quality degradation or information loss of a 2D or 3D dataset (in our case the CT images at different imaging conditions) by comparing it against a reference dataset (in our case, the ground truth images). All the metrics were calculated in the lung regions, the organ of interest in this study. To do so, the IQ values were calculated in each voxel. The final IQ value was set as the average value across all the voxels inside the lung which were identified using the ground truth masks.

To make the ground truth images, each voxel in the phantom was set to be the attenuation coefficient value of its corresponding material at the effective energy of the source spectrum (64 keV in this study). To calculate the quality metrics, the simulated CT and the ground truth images should be in the same resolution. Therefore, the CT images were linearly upsampled to a 0.25-mm isotropic resolution, matching the ground truth images voxel size. Since each phantom was four-dimensional, the ground truth was chosen to be at the central time that each CT slice was imaged.

2.3.2. Task-specific assessment

The bias and variability of radiomics quantifications of the lesions were assessed for the XCAT phantom with different motion rates across all imaging conditions. For all simulated CT and ground truth phantom images, a research-prototype radiomics software (CT Radiomics v1.2.2, Siemens Healthcare) was used to semiautomatically segment the lesions (single click on the center slice of the lesion) and calculate 10 morphological radiomics features (flatness, volume, surface area, maximum 3D diameter, spherical disproportion, compactness1, compactness2, sphericity, elongation, and surface to volume ratio), as defined by Zwanenburg et al.²⁷ and based on work by Aerts et al.²⁸ Since the modeled lesions were solid without tissue heterogeneity, textural features were not studied in this study. The morphological features were also measured in the ground truth images similar to the simulated images. Average relative bias and its variability (standard deviation) were calculated for each feature across all of the motion models and imaging conditions. Here, bias means the difference between a radiomics feature extracted from the simulated image and extracted from the ground truth image. The variability also refers to the variation between the bias measurements across the four lesions in each XCAT.

3. Results

3.1. Validation Results

Figure 3 shows the CT numbers for water and iodine as well as noise magnitude as a function of pitch and beam collimations. For all conditions studied, the simulation results were close to the real measurements, with errors smaller than 3 HU. For the pitch values studied, the CT numbers and noise magnitude were constant, with some fluctuations ($<3\mathrm{HU}$) in the iodine CT numbers. For the beam collimations, the “iodine” CT number decreased ($\sim 14\mathrm{HU}$) as a function of beam collimation. This decreasing trend can be attributed to the increased presence of scattered photons at larger collimations.²⁹ The decreasing trend did not happen in the “water” measurements due to “water calibration” postprocessing that is done per tube voltage and beam collimation. (For the real data, water calibration happens at the scanner; for DukeSim, it happens during the beam hardening correction.) Noise magnitude was also found to be constant as a function of beam collimation with a slight increase (1 HU) in the real images.

Fig. 3.

HU values in (a), (d) water, (b), (e) iodine, and (c), (f) image noise at multiple beam collimations and pitch values. Red is for real measurements and blue is for simulations.

In Fig. 3, both the first row (fixed collimation varying pitch) and second row (fixed pitch varying collimation) were done using similar kV, effective mAs, and reconstruction settings. However, due to scanner limitations, the measurements in the first row (fixed collimation varying pitch) were done using a “chest” bowtie filter protocol, while the second row (fixed pitch varying collimation) measurements were done under a “head” protocol. In the scanner that we used in this study, “head” was the only protocol that has multiple collimation options, and “chest” was the only protocol that has this range of pitch values. Therefore, there is a slight difference between the results for the pitch of 1 and the collimation of 38.4 mm in the top and bottom rows.

3.2. Task-Generic Assessment

The runtime for each simulation was between 6 and $12\min /\text{rotation}$ (depending on the collimation and rates of motion) using a machine with Intel^® Xeon^® CPU 2.10 GHz, 64 GB memory, and four NVIDIA GeForce GTX Titan GPUs. The runtime for the image reconstruction was between 3 and 5 min for each dataset.

Simulated CT images of an XCAT with corresponding ground truth are shown in Fig. 4, qualitatively demonstrating how the imaging systems degrade the ground truth. When the XCAT phantom was imaged without cardiac or respiratory motions, helical artifacts were observed around high contrast materials, as the pitch increased (Fig. 4), due to the helical cone-beam artifacts that occur at high pitch values. When cardiac and respiratory motions were added, the helical artifacts exhibited a similar trend, but motion artifacts were suppressed (Fig. 5). This is likely because the time to acquire the data range to reconstruct an image slice decreases with increasing pitch while beam collimation and rotation time are constant.⁶

Fig. 4.

(a)–(c) Magnified regions of the ground truth and simulated CT images of the XCAT at the lowest and the highest pitch levels, showing the helical cone-beam artifacts at higher pitch values.

Fig. 5.

(a)–(e) Magnified regions of the ground truth and simulated CT images of the XCAT (with $16\text{breaths}/\min$ respiratory and $90\text{beats}/\min$ heart rates) imaged at the lowest and the highest pitch and beam collimations. Images show that motion artifacts are reduced at higher pitch values, while remaining unchanged at wider collimations.

Increasing the beam collimation yielded no major visual differences. Figure 6 shows a difference image between the images acquired at the largest and smallest beam collimations. To demonstrate only the effects of beam collimation, noise was not simulated for these two images. The RMSE between the largest and smallest collimations was 33, 47, 22, and 23 HU in the lung, bones, muscle, and fat regions, demonstrating that differences were larger in the highly attenuating bone structures.

Fig. 6.

A subtraction image between the CT images acquired at the smallest and largest beam collimations showing the main difference existed in the high attenuation and periphery regions of the image.

Moreover, motion artifacts were not improved at the wider beam collimations (Fig. 5) while keeping the pitch constant. This is likely because the time to acquire the data range to reconstruct an image slice remains constant with increasing beam collimation when pitch and rotation time are constant.⁶

The quantitative IQ assessment results are summarized in Figs. 7–9. Overall, the IQ metrics were sensitive to the amounts of cardiac and respiratory motions with lower quality in XCAT images with more motions. The RMSE, PSNR, and SSIM (all measured in the lungs) exhibited a degradation trend as the pitch increased due to the helical cone-beam artifacts at higher pitches. The IQ metrics were found to be sensitive to the amounts of cardiac and respiratory motion modeled in this study. When the XCAT was imaged with motion, due to the trade-off between spatial degradation and temporal resolution, the IQ showed a combination of degradation (due to helical artifact) and improvement (due to reduced motion artifact) with increasing pitch. In other words, the IQ metrics degradations were 13.3%, 4.9%, and 4.0% (in terms of RMSE, PSNR, and SSIM) when imaged with no motion, whereas 6.6%, 2.5%, and 1.6% when imaged with the highest motion ($16\text{breaths}/\min$ respiratory and $90\text{beats}/\min$ heart rate, Figs. 7–9).

Fig. 7.

RMSE of the simulated CT images (in the lungs) compared with the ground truth. The plots are for (a)–(c) different pitch values and (d)–(f) beam collimations. Panels (a) and (d) show the results when XCAT was imaged without any motion. In panels (b) and (e), XCAT was imaged with different heart rate without respiratory motion. In panels (c) and (f), the results were compared when XCAT was imaged at different magnitudes of respiratory motions.

Fig. 8.

PSNR of the simulated CT images (in the lungs) compared with the ground truth. The plots are for different (a)–(c) pitch values and (d)–(f) beam collimations. Panels (a) and (d) show the results when XCAT was imaged without any motion. In panels (b) and (e), XCAT was imaged with different heart rate without respiratory motion. In panels (c) and (f), the results were compared when XCAT was imaged at different magnitudes of respiratory motions.

Fig. 9.

SSIM index of the simulated CT images (in the lungs) compared with the ground truth. The plots are for different (a)–(c) pitch values and (d)–(f) beam collimations. Panels (a) and (d) show the results when XCAT was imaged without any motion. In panels (b) and (e), XCAT was imaged with different heart rate without respiratory motion. In panels (c) and (f), the results were compared when XCAT was imaged at different magnitudes of respiratory motions.

The RMSE, PSNR, and SSIM showed slight degradation ($<1.3\%$) with the increase in collimation values (Figs. 7–9). This is mainly because the effects of beam collimations are minor at low-attenuating regions (for this study, lungs) and central parts of the images.

3.3. Task-Specific Assessment

The relative bias and variability for the radiomics features, the pitch values, and beam collimations are shown in Fig. 10. Each cell in this figure is the average result across all four lesions and phantom motion conditions.

Fig. 10.

(a) Relative bias and (b) variability, across multiple pitch and beam collimations, averaged across different XCAT models, each with four lesions. “p” refers to pitch and “c” refers to beam collimation. For the relative bias plot, red means overestimation of the feature and blue means the opposite.

Size-related features such as surface area, spherical disproportion, and volume were overestimated. This is due to overestimate of the lesion segmentation, which is likely caused by the inherent blur of the lesion representation after being imaged. Other features like compactness and sphericity were underestimated.

Among all features across all imaging conditions studies, compactness2 and elongation showed the largest ($-26.5\%$, 7.8%) and smallest ($-0.8\%$, 2.7%) relative bias and variability, respectively (negative means underestimation). As demonstrated in Fig. 10, the relative bias and variability were generally consistent across the imaging conditions due to the robustness of the segmentation algorithm across all these conditions. Table 2 includes the relative bias and variability results when averaged across all radiomics features, lesions, and phantom motion conditions studied. The results show that the pitch of 0.50 with beam collimation of 38.4 mm had the smallest relative bias and the pitch of 1.25 had the largest average relative bias. The values in Table 2 demonstrate that the relative bias and variability were generally consistent across all imaging conditions studied.

Table 2.

Average relative bias and variability across all the phantoms for different pitch values and beam collimations.

	$p=0.5$	$p=0.75$	$p=1.00$	$p=1.25$	$p=1.50$	$c=19.2$	$c=28.0$	$c=38.4$	$c=48.0$	$c=57.6$
Relative bias	$8.3\pm 5.7$	$11.0\pm 6.7$	$10.8\pm 7.3$	$13.45\pm 8.5$	$13.0\pm 8.7$	$8.6\pm 6.1$	$10.4\pm 7.0$	$10.8\pm 7.3$	$10.9\pm 7.3$	$11.2\pm 7.2$
Variability	$4.3\pm 1.6$	$5.3\pm 2.1$	$4.6\pm 1.7$	$4.1\pm 2.0$	$4.5\pm 1.9$	$4.7\pm 2.4$	$4.7\pm 1.8$	$4.6\pm 1.7$	$4.2\pm 1.6$	$4.0\pm 1.7$

Note: $p$, pitch; $c$, beam collimation.

Figure 11 shows the radiomics of the relative bias and variability results for XCAT with no motion and the highest motion rates as well as its absolute difference. The absolute differences between no motion and highest motion were [$-7.4\%$,4.3%] and [$-4.6\%$,4.1%] (positive means that no motion had a better performance) in terms of relative bias and variability, respectively.

Fig. 11.

(a), (c), (e) Relative bias and (b), (d), (f) variability for XCAT with no motion (a), (b) and with $16\text{breaths}/\min$ of respiratory and $90\text{beats}/\min$ heart rates (c), (d). “$p$” refers to pitch and “$c$” refers to beam collimation. For the relative bias plots, red means overestimation of the feature and blue means the opposite. (e), (f) The absolute error difference between the two conditions is shown. In the bottom row, red means that XCAT with no motion had a better performance and blue means the opposite.

While the differences were small, the XCAT with motion had less bias for some radiomics features compared with the XCAT without motion [Fig. 11(e)]. As shown in Fig. 10, size-related features are overestimated across all conditions likely due to the inherent blur of the lesion after being imaged. With motion added, the size, blurring, and appearance of the lesion will be affected in a nonlinear complex way. These effects combined with the segmentation algorithm will determine the magnitude of the overestimation. For example, Fig. 12 shows XCAT without motion and with highest respiratory motions, both simulated at pitch of 0.5 and beam collimation of 38.4 mm. Beyond the motion artifact that is present in the phantom with motion, this figure shows that the lesion looks slightly smaller for the moving phantom, resulting in less overestimation of the size-related radiomics features. Overall, the results did not show any significant difference between the performances of XCAT and different motion magnitudes, demonstrating that both relative bias and variability results being generally consistent across different motion profiles.

Fig. 12.

XCAT (a) without motion and (b) with $16\text{breaths}/\min$ of respiratory and $90\text{beats}/\min$ heart rates, both simulated at pitch of 0.5 and beam collimation of 38.4 mm. The images highlight the effects of motion on lesion appearance.

4. Discussion

Large beam collimations and high pitch values can accelerate the acquisition speeds in CT. In this study, we constructed and performed a VCT to assess the effects of large beam collimations and high pitch values on IQ.

The results indicate that under constant exposure conditions, images acquired at higher clinical pitches (up to 1.5) showed inferior IQ compared to the lower pitch images. However, when the patient moved during the acquisition, the higher pitch images had less motion artifacts. This finding is consistent with other studies, such as the recommendation of Raman et al.,³⁰ on imaging patients at lower pitch values, especially in tasks, where motion artifacts are not a concern. For example, an observer study by Lança et al.,³¹ indicated that lower pitch axial CT images have superior IQ in the liver and pancreas. However, in pediatric or cardiac imaging, where motion artifact is a common challenge, high pitch values have been shown great values in lowering the motion artifacts³² and having more accurate quantifications.^32,33

Moreover, the IQ results showed that the images acquired using larger beam collimations (up to 57.6 mm at the isocenter) had slight inferior IQ compared to the lower collimations in the lung regions. The RMSE between the largest and smallest collimations was 22–47 HU in different regions, suggesting the use of narrow collimations for imaging tasks, where HU quantification accuracy is paramount, e.g., lung density quantification in COPD imaging.³⁴

Our results indicate a robust morphological radiomics feature quantification across the different pitches and beam collimations. This finding, integrated with early investigations,^35–38 can lead to better understanding of the sources of bias and variability in radiomics quantifications in CT. In particular, the segmentation algorithms play a crucial role in accurate and consistent performance of radiomics quantifications.³⁹ VCT studies help to quantify the sensitivity of the segmentation algorithms across different imaging conditions.

In this study, we demonstrated how a realistic VCT construct, consisting of a detailed dynamic human phantom and a scanner-specific simulator, can be utilized to investigate clinically relevant questions in radiology. The VCT platform enabled us to understand the effects of two key imaging parameters on the CT images while the other factors (both patient and scanner related) were kept constant: a crucial condition that is not possible in conventional clinical trials. Moreover, the ground truth knowledge of the dynamic human models enables highly accurate comparisons of the images across different imaging conditions.

This study has some limitations. First, a limited number of patient models and imaging parameters were utilized. We used one male anatomy with seven common combinations of cardiac ($60-120\text{beats}/\min$) and respiratory motion ($0-16\text{breaths}/\min$) profiles. This study focused on the rates of respiratory and cardiac motions, while their magnitudes were constant. Further investigations are envisioned for normal and diseased patients with variable motion magnitudes. This study was focused on one specific CT scanner with clinically relevant pitch and collimations. Future work will evaluate other scanner models and reconstruction algorithms.

The simulator used in this study has been validated against real data under various imaging conditions (dose, pitch, and collimation) in terms of HU accuracy, image noise, noise texture, and spatial resolution. These validations were done using simplistic uniform phantoms. Future studies will include more rigorous validations using heterogeneous phantoms. Further, we have not quantitatively validated the simulations in terms of motion artifacts. While the motion models were based on patient data and the appearance of the motion artifacts in the images looked visually realistic, we envision conducting a study to evaluate our simulations in terms of motion artifacts using a dynamic physical phantom with inserted lesions and a known motion profile, not currently available to us.

5. Conclusion

This study demonstrates the utility of VCTs in comprehensively exploring fundamental questions in CT. Computational phantoms and imaging simulation methods allow one to conduct studies that would be otherwise impossible with live subjects. Here, we used the virtual tools to investigate the effects of beam collimation and pitch on CT IQ under different rates of respiratory and cardiac motion. High pitch and large beam collimations can accelerate the image acquisition time; however, they can also negatively affect IQ. Our findings suggest that increasing beam collimations (from 19.2 to 57.6 mm) affects the CT numbers by 22 to 47 HU, depending on the materials, due to the increased presence of scatter and helical cone-beam artifacts. While increasing the pitch increased the RMSE, this increased rate decreased as a function of respiratory and cardiac motion (from 20 to 10 HU) due to the improved motion artifacts. The results indicate that pitch and beam collimations do not significantly affect the accuracy and variability of morphological radiomics quantifications.

Biography

Biographies of the authors are not available.

Disclosure

E.A., W.P.S., B.H., S.S., and A.K. have no conflicts of interest to disclose. Unrelated to this study, E.S. has active research grants with Siemens and GE and advisory board member of medInt Holdings, LLC.