Development of a method to create uniform phantoms for task‐based assessment of CT image quality

Abstract Purpose To develop a customized method to produce uniform phantoms for task‐based assessment of CT image quality. Methods Contrasts between polymethyl methacrylate (PMMA) and fructose solutions of different concentrations (240, 250, 260, 280, 290, 300, 310, 320, 330, and 340 mg/mL) were calculated. A phantom was produced by laser cutting PMMA slabs to the shape of a patient’s neck. An opening of 10 mm diameter was cut into the left parapharyngeal space. An angioplasty balloon was inserted and filled with the fructose solutions to simulate low‐contrast lesions. The phantom was scanned with six tube currents. Images were reconstructed with filtered back projection (FBP) and adaptive iterative dose reduction 3D (AIDR 3D). Calculated and measured contrasts were compared. The phantom was evaluated in a detectability experiment using images with 4 and 20 HU lesion contrast. Results Low‐contrast lesions of 4, 9, 11, 13, 18, 20, 24, 30, 35, and 37 HU contrast were simulated. Calculated and measured contrasts correlated excellently (r = 0.998; 95% confidence interval: 0.991 to 1). The mean ± SD difference was 0.41 ± 2.32 HU (P < 0.0001). Detection accuracy and reader confidence were 62.9 ± 18.2% and 1.58 ± 0.68 for 4 HU lesion contrast and 99.6 ± 1.3% and 4.27 ± 0.92 for 20 HU lesion contrast (P < 0.0001), confirming that the method produced lesions at the threshold of detectability. Conclusion A cost‐effective and flexible approach was developed to create uniform phantoms with low‐contrast signals. The method should facilitate access to customized phantoms for task‐based image quality assessment.

accreditation phantom (Gammex, Middleton, WI) contains cylinders of 2 to 25 mm diameter and 6 HU contrast. The phantom can be expanded with the Advanced iqModule, which has cylinders of 1.5 to 25 mm diameter and 3, 6, or 10 HU contrast. The Catphan phantom (Phantom Laboratories, Salem, NY) has a module with a series of cylindrical rods of 2 to 15 mm diameter and 0.3 to 1% contrast. The same company also provides the MITA IQ low-contrast phantom, which has rods of 3 to 15 mm diameter and 3 to 14 HU contrast.
As these phantoms differ in their arrangement of signals, signal diameters, and contrasts, they can offer advantages for different experimental designs, for example with regard to the choice of signals at the interface between detectable and undetectable. However, most institutions do not have all of these phantoms at their disposal.
Furthermore, some requirements are not fulfilled by any of the commercially available phantoms, for example, large signal spacing and multiple signals at the threshold of detectability. 5 Previous work used customized LCD phantoms that were tailored to a specific experimental setup. 6 However, such phantoms often have to be ordered from specialized manufacturers, which can delay studies and cause significant costs. A fast, cost-effective, and flexible method would be desirable to facilitate access to LCD phantoms which could ideally be tailored to particular study designs.
To create phantoms with low-contrast signals, at least two different materials must be combined. The contrast results from the difference of these materials' linear attenuation coefficients and can be calculated, if the chemical composition and the physical density of the materials are known. Polymethyl methacrylate (PMMA) is a uniform material frequently used for phantom bodies. 7,8 A second material to generate a signal can be a homogeneous fructose solution, which has the advantage that the contrast can be adjusted by the concentration of the solution. The present study explored these materials for the construction of LCD phantoms. The hypothesis was that calculated contrasts between two materials can be used as a basis to produce a uniform phantom with corresponding low-contrast signals. As part of ongoing work to evaluate a CT system for neck imaging, a neck-shaped phantom was created as proof of principle to illustrate the approach. The aim was to develop a customized method to produce uniform phantoms for task-based assessment of CT image quality.

| MATERIALS AND METHODS
The institutional ethics committee approved the study and waived informed consent.

2.A. | Calculation of Hounsfield units and contrast values
Mass attenuation coefficients of PMMA, water, and aqueous fructose solutions (240, 250, 260, 280, 290, 300, 310, 320, 330, and 340 mg/mL) were calculated as weighted sums of the mass attenuation coefficients of their atomic constituents [Eq. (1)] as provided by the National Institute of Standards and Technology (NIST) database. 9 The calculations were performed in the same way for PMMA, water, and fructose solutions.
Equation (1), where μ ρ is the mass attenuation coefficient (cm 2 /g) and w i is the mass fraction of element i.
Mass attenuation coefficients were used at 72 keV based on preliminary experiments to approximate the mean photon energy at 120 kVp CT imaging. The physical density of PMMA was provided by the manufacturer (1.18 g/cm 3 ). Densities of fructose solutions at 20°C were obtained from interpolation of published data. 10 Linear attenuation coefficients were calculated by multiplying the mass attenuation coefficients of PMMA, water, and the fructose solutions with their respective physical density. Hounsfield units (HU) were calculated, and contrast values were calculated as difference between PMMA HU and fructose HU.

2.B. | Phantom construction
A neck CT image of a patient was used as template to create a phantom with the patient's neck shape. The dimensions were 15.4 cm (length) × 10.6 cm (width). A circular area of 10 mm diameter in the left parapharyngeal space was selected to insert lowcontrast signals. Figure 1 shows the CT image of the patient and the F I G . 1. Neck computed tomography image of a patient that was used as a template for the phantom shape. The region of interest in the left parapharyngeal space indicates the position for insertion of low-contrast signals. signal position. Nineteen PMMA slabs of 5 mm thickness were laser cut to the patient's shape. Circular openings of 10 mm diameter were inserted into four slabs and openings of 14 mm diameter were inserted into 11 slabs in the left parapharyngeal space. The remaining four slabs did not contain any openings. Next, the metal markers were removed from a Mustang angioplasty balloon of 12 mm diameter and 4 cm length (Boston Scientific, Marlborough, MA) to avoid artifacts that could interfere with HU measurement and detection tasks: First, the balloon was opened at the top and the markers were removed. Second, the balloon was closed with thin thread and sealed with cyanoacrylate. After these preparations, the PMMA slabs were stacked so that slabs with 10 and 14 mm openings were stacked alternately. The slabs were compressed between two wood panels, the angioplasty balloon was inserted into the opening and  imaging on our CT system. The tube voltage was 120 kVp. Six tube currents were used: 10, 20, 30, 40, 100, and 120 mA. CTDIvol values were 0.5, 0.9, 1.4, 1.9, 4.7, and 5.6 mGy, respectively. Images were reconstructed with filtered back projection (FBP) and adaptive iterative dose reduction 3D (AIDR 3D). The standard soft tissue kernel (FC08) and 0.5 mm thin slice reconstruction as in our clinical neck protocol were used. Five repeated acquisitions were performed per tube current and fructose concentration. A total of 600 data sets were generated (10 fructose concentrations × 6 tube currents × 2 reconstruction methods × 5 repetitions).

2.D. | HU and contrast analysis
Nine images per data set were used to measure contrasts between low-contrast lesions simulated with the fructose solutions and the PMMA background. On every image, one circular region of interest (ROI) of 8 mm diameter was placed into the low-contrast lesion and six circular ROIs of 25 mm diameter were placed into the PMMA background surrounding the lesion. The contrast per image was calculated as the difference between mean PMMA HU and fructose HU. For comparison with calculated contrasts, the measured contrast values were averaged over all data sets to reduce bias resulting from effects of acquisition and reconstruction parameters on measured values. 11,12 In addition to the CT scans of the neck phantom, we also analyzed HU in CT images of a 16-cm CTDI phantom, which, like the background of the neck phantom developed in our study, consists of PMMA. CT images of the CTDI phantom were acquired in the same manner as for the neck phantom (6 tube currents × 2 reconstructions × 5 repetitions), and HU measurements were also analyzed in the same way for six circular ROIs of 25 mm diameter in nine images per acquisition.

2.F. | Data analysis
Measured HU values are presented as mean AE standard deviation (SD) and as median and range. Correlation analysis was performed using Pearson correlation. Estimates are given as correlation coefficient r and

3.B. | HU and contrast analysis
Measured contrast values correlated excellently with calculated values (Fig. 4) Table 3 provides separate detection accuracy results per tube current and image reconstruction. Table 4 summarizes corresponding confidence results. At 4 HU lesion contrast and 30 mA tube current (CTDIvol 1.4 mGy), detection accuracy was significantly higher for AIDR 3D-than for FBP-reconstructed images (P = 0.008), which was not the case at 120 mA tube current  decreased between 120 and 30 mA tube current for FBP-(P = 0.0495), but not for AIDR 3D-reconstructed images (P = 0.992).

| DISCUSSION
A frequent approach to task-based image quality assessment is to test how well an image enables an observer to detect a signal, which requires to examine appropriate phantoms with embedded signals.
The availability of such phantoms is limited and they cannot easily be tailored to specific study needs. The present study aimed at developing a method to produce customized phantoms for taskbased image quality assessment. The method presented here produced uniform signals, the con-    The detectability experiment we performed ruled out signals or artifacts distorting the low-contrast signals. Thus, our results show that the phantom we designed is suitable for its intended purpose.
While detection accuracy and reader confidence were high at 20 HU lesion contrast, readers were unconfident in selecting the image containing the lesion, and their detection accuracy was low at 4 HU contrast. Results for 4 HU contrast were similar to those of a previous study investigating signals of 6 mm diameter and 5 HU contrast at similar dose levels. 17  The limitations of this study include that only one lesion size was investigated and that long-term stability and reproducibility were not investigated. HU values were only measured with one CT system, and results may slightly vary with different systems and energy spectra. 18,19 However, this limitation applies to all LCD phantoms, and such variations can be expected to be small, as only materials with low atomic numbers (and thus relatively low energy dependence) were used. Yet, if necessary, the calculation model could also be adapted to account for different energy spectra.
Effects of dose and reconstruction methods were beyond the scope of this work and therefore not analyzed in detail. It should also be noted that a uniform phantom background texture was produced and that more complex textures were previously shown to affect detection results. 20  In conclusion, the present work shows that contrast values between two uniform materials can be calculated and used to produce phantoms for task-based image quality assessment. The method we propose facilitates access to such phantoms and provides flexibility in creating phantoms tailored to specific study designs involving detection tasks of low-contrast signals for assessment of image quality.

ACKNOWLEDGMENTS
We thank our colleagues from the department of radiology and neuroradiology for participating in the detectability experiment and Bettina Herwig for assistance with preparation of the article.

CONF LICT OF I NTEREST
This study has received funding by the Bundesministerium für