Scanner-specific validation of a CT simulator using a COPD-emulated anthropomorphic phantom

Abstract

Traditional methods of quantitative analysis of CT images typically involve working with patient data, which is often expensive and limited in terms of ground truth. To counter these restrictions, quantitative assessments can instead be made through Virtual Imaging Trials (VITs) which simulate the CT imaging process. This study sought to validate DukeSim (a scanner-specific CT simulator) utilizing clinically relevant biomarkers for a customized anthropomorphic chest phantom. The physical phantom was imaged utilizing two commercial CT scanners (Siemens Somatom Force and Definition Flash) with varying imaging parameters. A computational version of the phantom was simulated utilizing DukeSim for each corresponding real acquisition. Biomarkers were computed and compared between the real and virtually acquired CT images to assess the validity of DukeSim. The simulated images closely matched the real images both qualitatively and quantitatively, with the average biomarker percent difference of 3.84% (range 0.19% to 18.27%). Results showed that DukeSim is reasonably well validated across various patient imaging conditions and scanners, which indicates the utility of DukeSim for further VIT studies where real patient data may not be feasible.

1. Introduction

The utilization of Computed Tomography (CT) imaging is highly prevalent in modern medicine, with well over 70 million CT scans being performed annually [1]. Though CT technology aids physicians by providing a non-invasive form of diagnosis, quantitative analysis of such CT images traditionally involves working with real patient cases, which poses several limitations. Namely, working with real patient scans can be expensive, inefficient, and ground-truth [2].

An alternative approach is through use of Virtual Imaging Trials (VITs). VITs use virtual simulations of the image acquisition process. Computational patient phantoms along with computerized scanners are used to virtually conduct these trials. VIT simulations allow for quantitative assessment of the images in an efficient manner, compared to real patient images. This efficiency comes in part due to the nature of computer simulations being more cost effective and easily replicable, in addition to minimizing any radiation exposure. In addition, because the exact shape and composition of the virtual phantom is known, a ground truth is also known for the virtual patient, thereby alleviating some of the limitations seen with traditional methods [3].

The imaging processes can be modeled by DukeSim, a scanner-specific CT simulator platform. DukeSim outputs sinogram data for the input phantom through its primary ray tracing and Monte Carlo module to accommodate primary and secondary scatter [4–6]. Though DukeSim has been validated with cylindrical phantoms in the past, validation with more clinically relevant phantoms have not been performed [6, 7]. Hence, there is a need to validate DukeSim with an anthropomorphic phantom, as this may more realistically model the anatomies of patients seen in clinical settings.

Of particular interest is Chronic Obstructive Pulmonary Diseases (COPD) and other lower respiratory diseases, as this category of diseases is one of the leading causes of death in the United States [8]. This study seeks to validate DukeSim using measurements related to such diseases by using a chest phantom that models underlying patient anatomy.

2. Methods

2.1. Chest Phantom

An anthropomorphic chest phantom from the Kyoto Kagaku company was utilized (Multipurpose Chest Phantom N1 “LUNGMAN” (PH-1)) [9–11]. This phantom contained urethane models of a patient’s chest (including urethane lungs), along with synthetic bones (ribs, spine, scapulae, and sternum) made from an epoxy resin [11].

To enhance the utility of this phantom, three vertically spanning tubes were placed into the phantom as shown in Figure 1, each containing a set of experimental material inserts. One broad tube was placed in the patient model of the right lung, and two thinner tubes were placed in the patient model of the left lung. With this setup, it was possible to study multiple materials. Three different sets of experimental materials were used, labeled as Configuration 1, 2, and 3 in Figure 1. The materials studied were lung samples A, B, C, D, E, and F, which are derived from pig lung samples from an inflated and over-inflated pair of lungs. Additionally, standardized foams obtained from the National Institute of Standards (NIST) were placed into the phantom model [12].

Figure 1.

Diagram of the modified Kyoto Kagaku chest phantom and each of the three tube configurations

2.2. Real CT Image Data

To validate a virtual simulation of the Kyoto Kagaku chest phantom, a reference was needed. For this, the chest phantom was scanned utilizing two different commercial CT scanners: Siemens SOMATOM Definition Flash and Siemens SOMATOM Force. Each of these scans had their own set of CT imaging parameters, as outlined in Table 1.

Table 1.

Outline of the CT scan imaging parameters from Siemens Force and Flash

	Number of Images	CTDIvol Range (mGy)	kV	Reconstruction Algorithm	Reconstruction Kernel	Pixel Spacing (mm) × (mm)	Slice Thickness (mm)
Siemens SOMATOM Definition Flash	15	3.71 – 7.42	120	SAFIRE Strengths 3 and 5 or wFBP	Q30f and B35f	0.51 × 0.51	0.75
Siemens SOMATOM Force	13	1.15 – 7.35	120	ADMIRE Strengths 3 and 5 and wFBP	Qr40d and Bf40d	0.51× 0.51 and 0.98× 0.98	0.75

2.3. Virtual Imaging Trial (VIT)

A corresponding VIT was conducted utilizing the same imaging conditions for each of the 28 real images in Table 1. To generate a computational (computerized) version of the physical Kyoto Kagaku phantom, an open-source software (Seg3D [13]) was utilized to manually segment masks of each material in the chest phantom given a real CT image input. Each individual mask corresponding to a particular material in the phantom was combined using MATLAB to generate a final virtual phantom, with 0.25 mm voxel size. Cross-sectional views of this phantom are shown in Figure 2, in which each intensity corresponds to a different material of the phantom.

Figure 2.

Cross-sectional views of the computational version of the Kyoto Kagaku Chest Phantom, seen from an axial (left), sagittal (middle), and coronal (right) plane.

The linear attenuation profiles of each material of the Kyoto Kagaku phantom were then computed using an X-ray library database [14], in addition to other vendor-specific material information. Note that the low contrast material was approximated as a liver tissue, since the elemental composition of low contrast was unknown. The virtual phantom in Figure 3, along with the linear attenuation profiles, imaging parameters in Table 1, and scanner-specific information were then inputted into DukeSim for each of the 28 simulations. The resulting sinogram projection data from each simulation was then reconstructed with the corresponding reconstruction parameters outlined in Table 1, using ReconCT, version 15.0.53098.0, a vendor-specific reconstruction software.

Figure 3.

Axial cross-sectional views of the real (left column) versus simulated (right column) images for both the Siemens Flash (top row) and Force (bottom row) scanners. Level: −437.5, Width: 1173.

2.4. Biomarker Validation Analysis

In order to evaluate the accuracy of DukeSim, a set of pulmonary imaging biomarkers (outlined in Table 2) were computed and compared between the real CT image data and the corresponding simulated image generated from reconstructing DukeSim’s projection data.

Table 2.

Pulmonary imaging biomarkers computed for both real and simulated CT image data, adapted from [15].

Biomarker	Definition
LAA −950 (%)	% Lung Voxels with HU ≤ −950 (measure of airflow obstruction)
LAA −856 (%)	% Lung voxels with HU ≤ −856 (measure of gas trapping)
Perc 15 (HU)	Lung HU value at 15th percentile (measure of emphysema progression)
Lung Mass (g)	$\sum _{\mathit{\text{All Lung Voxels}}}\frac{HU+1024}{1024}\left(\mathit{\text{Voxel Size}}\right)$
Lung Volume (cm3)	# Lung Voxels × Voxel Size
Lung Voxel Distribution (HU)	Mean ± Std
Experimental Insert Voxel Distribution (HU)	Mean ± Std

3. Results

Visual results of the simulated and real results across both the Siemens Flash and Force CT scanners are seen in Figure 3, in which the real and simulated reconstructions seem qualitatively similar.

Raw comparisons between the biomarkers computed from Table 2 between the real and simulated data were expressed as a percent difference, assuming the biomarker obtained from the real CT image to be the ground truth. These relative differences were computed to be relatively small, with the average percent difference across all biomarkers being only 3.84% (range 0.19% to 18.27%).

The raw measurements between real and simulated data for each scanner can be seen in Figure 4, with each subplot corresponding to a different unit of biomarker (HU, %, g, or cm³). Each bar represents the mean biomarker measurement across all relevant images, with the corresponding uncertainty being derived from the standard deviation of the measurements. A portion of the error bars tend to overlap in Figure 4, generally indicating a closer match between real and simulated data. Some scanner-specific results are obtained as well. For example, the HU measurements across images acquired with the Force scanner tended to have a higher magnitude than those for the Flash scanner. Some of the inserts shown in Figure 1 were not considered (Water and Lung Samples A, B, C, D, E, and F) due to their unknown textures.

Figure 4.

Visual comparisons of each biomarker for each scanner, split by biomarker unit: HU (top), % (bottom left), and cm³/g (bottom right). Note that RT = Right tube, LPT = Left Posterior Tube, LAT = Left Anterior Tube.

4. Discussion

In this work, DukeSim was validated with respect to clinically relevant biomarkers using an anthropomorphic COPD phantom across a range of different imaging parameters and scanners both qualitatively and quantitatively (mean biomarker difference of 3.84%).

In terms of discrepancy between real and simulated images, one potential reason could be due the differences in material compositions and density for a particular material of the real phantom compared to its theoretical model under DukeSim. There may also be differences in the scanner-specific modeling and calibration of DukeSim compared to the real Siemens CT scanners.

5. Conclusion

This study highlighted the utility of VITs in clinically meaningful contexts across a range of different imaging and reconstruction parameters. Not only do VITs provide efficient and robust means of obtaining clinical information, but they also avoid the limitations with traditional imaging trials associated with higher levels of radiation exposure. More VIT studies can be conducted in the future to better understand other topics in clinical medicine.