Novel Materials for Low-Contrast Phantoms for Computed Tomography

Abstract

Acceptance testing and quality control of computed tomography (CT) scanners are of great importance. While most procedures and phantoms for testing other parameters are widely accepted, there is still discussion and uncertainty about low-contrast (LC) performance tests that measure the capability of a CT scanner to discriminate low-contrast objects. This work investigated the development of LC phantoms with available, low-cost polystyrene resin materials and some selected additives. We designed and tested phantoms with several different contrast steps by generating contrast in two different ways, one based on ‘physical density difference’ and the other on ‘atomic number difference’. Physical density difference was achieved by adding a small amount of glycerin to the polystyrene resin, both having similar low atomic-number elements but differing in the density of their atoms. Atomic number difference was achieved by adding a small amount of iodobenzene to the resin, both having approximately the same physical density (less than 1% variation in density) but different atomic (i.e., elemental) composition. Prototypes were evaluated using a Philips Tomoscan LX system and varying beam properties. The behavior and validity of the results are discussed.

1. INTRODUCTION

Since it was first introduced three decades ago, computed tomography (CT) has become an important investigative tool. In hospitals in the United States, CT accounts for 5–10% of all diagnostic radiology procedures performed [1]. Its domains have increased rapidly over the recent years with the developments in CT technology like helical and multi-slice CT, real-time cardiac CT, phase contrast CT, micro-CT, and novel detectors like variable resolution x-ray detectors that dramatically increase the image resolution [2,3,4]. The reliability of these radiographic methods must be quantitatively evaluated, and therefore acceptance testing and quality control of CT scanners are of great importance [5]. While most procedures and phantoms for testing image quality parameters are widely accepted, there is still discussion and uncertainty about the low-contrast (LC) performance test which determines the capability of a CT scanner to discriminate low-contrast objects [6]. Since much relevant soft tissue detail is low-contrast in nature, it is a clinically important test. The standardization of LC performance measurement techniques has yet to be reached and the lack of suitable phantoms is a major problem.

Phantoms are structures that contain one or more tissue substitutes and are used to simulate radiation interactions in the human body. There are two types of phantoms, anthropomorphic phantoms and mechanical phantoms. Anthropomorphic phantoms provide a simplified geometrical description of the human body and are designed to represent the human body's characteristics in terms of radiation attenuation, physical morphology, and geometry. Mechanical phantoms are designed to evaluate and standardize image quality parameters. They usually contain a variety of static and dynamic test targets for objective assessment of image quality. Robust mechanical phantoms can be manufactured easily at low cost which yield quantifiable, easily reproducible data and are superior for quality assurance and periodic long-term testing.

Measurement of low-contrast resolution requires special phantoms. The selection of suitable materials for the phantoms is important since contrast resolution depends on the intrinsic contrast, size and shape of the phantoms used, along with the physical characteristics of the CT scanner under study [7]. There are two ways of constructing low-contrast phantoms: using of the partial volume effect [8] or using materials with slightly different attenuation coefficients. In the first method, a tight tolerance of material composition is needed because most CT scanners can distinguish linear attenuation coefficients that differ by 0.5% or less. In addition, partial-volume techniques may not work well for thin slices. In the second method, the attenuation can be changed by altering physical density or effective atomic number or both.

With available, low-cost polyester resin materials and selected additives, we designed and tested a LC phantom with several different contrast steps. We generated contrast using two different methods of attenuation variation, one based on physical density difference and the other on atomic number difference. Physical density difference was achieved by altering the electron density with addition of glycerin to the polystyrene resin while maintaining approximately the same atomic composition or effective atomic number (Z). Atomic number difference was achieved by adding an organic material with a high Z component (iodobenzene) to the resin. It is useful to study physical density and atomic number effects separately because contrast resulting from both effects, and usually a combination of the effects, is seen in the clinical setting. For example, tumors/nodules are often similar to surrounding soft tissue and differ in density rather than composition, while calcified nodules exhibit contrast due to their composition. The new LC phantom constructed using the above methods proved to be very suitable for measuring a scanner's low contrast performance, one of the most important image quality parameters.

2. EXPERIMENTAL PROCEDURE

2.1. Phantom Construction

Plastics are important engineering materials. They are relatively low in cost and exhibit a wide range of properties. In this study, we used a two-component liquid plastic casting resin (Environmental Technology, Inc., Fields Landing, CA). It is a viscous, pale-colored unsaturated solution of polyester in styrene (35%) and a thermoset plastic that forms into a permanent shape and is ‘cured’ or set by a chemical reaction. The advantages of these thermoset plastics are many including high rigidity, high dimensional stability, and high thermal stability [9]. Catalyst is added to the resin shortly before use to initiate the polymerization reaction which creates heat to cure the resin. The reaction causes the resin to become more viscous until it reaches a state when it is no longer a liquid and has lost its ability to flow [10]. The curing time of the resin is approximately thirty minutes, but this depends on the amount of catalyst added to the resin and the curing temperature. The resin can be cast into flexible or rigid molds at room temperature. Once the resin is cured, it becomes a chemically-resistant, hard solid.

In CT, contrast is determined by both the differences in the physical densities and atomic numbers of the phantom materials. Estimated CT numbers for each of the resin mixtures were determined using XCOM: Photon Cross Sections Database (National Institute of Standards and Technology, Gaithersburg, MD). By definition, the CT number of water and air are assigned values of 0 HU and -1000 HU, respectively. Physical density difference was achieved by adding a small amount of glycerin to the polystyrene resin, both having nearly the same chemical distribution of elements but differing in the density of their atoms. Atomic number difference was achieved by adding a small amount of iodobenzene to the resin, both having approximately the same physical density (less than 1% variation in density) but different atomic composition.

In this study, low-contrast phantoms were constructed using two resin-casting sessions. The resin and additives were first measured to create the 5.5 cm diameter bases of the contrast phantoms. The catalyst was mixed with the resin according to the manufacturer’s instructions (room temperature, stirred for three minutes) to create 63.5 mm diameter cylindrical test phantoms. Each mixture contained the same amount of catalyst (1.48 ml catalyst per 60 ml of resin). The resin was poured into polyethylene molds and put in an oven at 65°C for 20 hours. A template was used on the cured bases to locate the phantom centers and then the bases were machined to create recessed holes which were filled with resin mixed with the selected amount of glycerin to create the 12.7 mm diameter ‘signal’ regions. This large diameter was used to verify easily the cylinder-to-background contrast level. The phantoms were then reheated (65°C for 20 hours) to finish the curing process. The iodobenzene phantoms were constructed in the same manner.

2.2. Measurement Procedure

A series of low-contrast and null (i.e., with no change of attenuation in the signal region) phantom images were obtained on a Tomoscan LX CT system (Philips Medical Systems, Madison, WI). This is a third-generation CT system for whole body scanning that uses a xenon gas detector. The details of the system parameters have been described previously [11]. The phantoms were scanned at various dose levels using different combinations of voltage (80–130 kVp) and current (50–300 mA). Ten images of each phantom were acquired at each of the dose settings within the same testing interval so that means and standard deviations could be calculated. An acrylic reference cylinder was also scanned with the phantoms for monitoring the variation of the mean CT numbers and measurement of the systematic error in the CT scanner.

Each low-contrast phantom consisted of a central cylindrical signal region made of resin and additives (glycerin and iodobenzene) and a surrounding background region made of resin only. Regions-of-interest were selected in the signal and background regions to calculate the mean CT numbers. The region-of-interest in the signal region was slightly smaller than the actual diameter to make sure that no contribution from the background region was included. The absolute contrast (C_abs) was calculated by measuring the mean CT numbers in the signal and surrounding background regions using Eq. (1).

$${\text{C}}_{\text{abs}}=\langle {\text{I}}_{\text{s}}\rangle -\langle {\text{I}}_{\text{b}}\rangle$$ (1)

In this equation, I_s and I_b are the pixel intensities in the signal and background regions respectively and the brackets indicate the mean value. The noise and contrast signal-to-noise ratio (SNR_C) were calculated from the measurement of absolute contrast and the standard deviation of the mean pixel intensities in the signal and background regions using Eq. (2).

$${\text{SNR}}_{\text{C}}=\frac{\langle {\text{C}}_{\text{abs}}\rangle }{\sigma \left(\langle {\text{C}}_{\text{abs}}\rangle \right)}=\frac{\langle {\text{I}}_{\text{s}}\rangle -\langle {\text{I}}_{\text{b}}\rangle }{\sqrt{\frac{{\sigma }^2\langle {\text{I}}_{\text{s}}\rangle }{{\text{N}}_{\text{s}}}+\frac{{\sigma }^2\langle {\text{I}}_{\text{b}}\rangle }{{\text{N}}_{\text{b}}}}}$$ (2)

In this equation, N_s and N_b are the number of pixels in the signal and background regions respectively and σ indicates standard deviation. The mean pixel intensities in the signal and background regions of a phantom were measured using ten images acquired at each set of parameters. From these ten measurements, the mean contrast and the standard deviation in the mean were calculated for each set of acquisition conditions. The noise measured with this equation is the ‘uncertainty’ in the mean contrast. As noise increases, the error in the mean and uncertainty in the contrast increases and thus, the SNR_C decreases. This is a significant measure, because it has been shown that once the SNR_C falls below a certain value (~5), one can no longer discriminate contrast [12].

3. RESULTS

The presented results are for a given set of imaging parameters: 100 kVp, 100 mA, 1 cm slice, 3.8 s scan time, and 30 cm field-of-view, but the behavior of the phantoms was similar for each dose level. The phantoms demonstrated temporal stability, linear contrast and SNR_C response as expected.

Contrast differences observed in images of resin-only phantoms are independent of the amount of catalyst added to the resin. Varying the amount of catalyst does not noticeably change the density or atomic number of the material since the test phantoms made with different amounts of catalyst all have the same mean CT number (Fig. 1), but it does affect the curing time. Consistent and selectable contrast levels were generated by changing the amount of glycerin or iodobenzene added to the resin. Temporal stability of both of the glycerin and iodobenzene phantoms was studied and the absolute contrast remained essentially constant within errors over a year for each set of phantoms as shown in Fig. 2 and Fig. 3. The scatter in the data for both glycerin and iodobenzene is the result of statistical and experimental error. The error bars show the standard deviation of the means, but the changes in absolute contrast indicate a systematic error mostly likely due to the CT scanner itself. This is further seen by the apparent decrease in the absolute contrast of the acrylic reference provided by the CT manufacturer. Since the resin system follows the same trend, we assume that LC phantoms are not changing with time. The linear fits for all contrast levels and for the overall mean are virtually flat over the course of a year within statistical and systematic limits. The amount of additive can be varied to create any desired contrast level and the actual contrast response for the glycerin and iodobenzene phantoms is shown in Fig. 4 and Fig. 5, respectively. The additive amounts were taken as powers of two (e.g., 1/8, 1/4, 1/2, 1, 2, and 4 ml of glycerin) to give more weight to the important (lowest) contrast values. The measured contrast is linear with added material within statistical error. The Χ² probabilities, 36% for glycerin and 25% for iodobenzene, are acceptable.

Fig. 1.

CT number stability of the low-contrastphantoms (μ = 125.92 HU, Χ²= 4.91, p=0.30).

Fig. 2.

Temporal stability of physical density (glycerin) contrast phantoms (‘overall mean’ Χ²=18.98, p=0.02) where ● = 0 ml (noadditive), ○ = 0.1 ml, ■ = 0.25 ml, □ = 0.5 ml, ▲ = 1 ml, △ = 2 ml, ◆ = 4 ml, ◇ = Acrylic (reference), = overall mean.

Fig. 3.

Temporal stability of atomic number contrast phantoms with iodobenzene (‘overall mean’ Χ² =11.3, p=0.26).

Fig. 4.

Contrast response of phantoms with varying amounts of glycerin (Χ² = 6.54, p=0.36)

Fig. 5.

Contrast response of phantoms with varying amounts of iodobenzene (Χ²= 5.35, p=0.25)

As seen in Eq. (2), the SNR_C is dependent on both the contrast and noise in the image. Since the contrast generated increases with the increasing amount of additive and the noise is substantially constant for a given set of imaging parameters, the SNR_C increases linearly with the amount of additives as shown in Fig. 6 and Fig. 7. These plots are also useful in allowing one to make a phantom with targets set at (or near) the threshold SNR_C for a particular set of x-ray technique factors (kVp and mAs). By changing the imaging parameters, the photon noise in the image is changed, and it decreases with increasing tube current (I) (proportional to the number of photons) at a constant tube voltage. The observed noise (σ_obs) is dependent on the photon noise (σ_p) as well as a noise floor (σ₀) generated by the inherent physical limitations of the system and the image reconstruction process as shown by the following equations [8, 9]:

Fig. 6.

Contrast signal-to-noise ratio vs. amount of glycerin (Χ²= 10.07, p=0.25)

Fig. 7.

Contrast signal-to-noise ratio vs. amount of iodobenzene (Χ²=4.40, p=0.35)

$${{\sigma }_{\text{obs}}}^2={{\sigma }_{\text{p}}}^2+{{\sigma }_{\text{0}}}^2$$ (3)

$${\sigma }_{\text{p}}\alpha \frac{1}{\sqrt{\text{I}}}$$ (4)

$${\text{SNR}}_{\text{C}}=\frac{\langle {\text{C}}_{\text{abs}}\rangle }{\sigma \left(\langle {\text{C}}_{\text{abs}}\rangle \right)}=\frac{\langle {\text{C}}_{\text{abs}}\rangle }{\sqrt{{{\sigma }_{\text{p}}}^2+{{\sigma }_0}^2}}=\frac{\langle {\text{C}}_{\text{abs}}\rangle }{\sqrt{\frac{\text{k}}{\text{I}}+{{\sigma }_0}^2}}(\text{where\hspace{0.17em}}\hspace{0.17em}\text{k}=\text{constant})$$ (5)

The noise response with varying tube current for selected glycerin and iodobenzene phantoms is shown in Fig. 8 and Fig. 9. The model fits the data well when σ₀ = 0.016 ± 0.001HU. The Χ² probability for these results is larger (i.e., >99%) than expected from normal statistics. This is a reflection of the slight nonuniformity of the noise fields in small objects generated by the reconstruction algorithm of the CT scanner. This assertion is supported by the observation that even though the measured standard deviation in the noise (in each image) is larger than expected due to a slight shading in the images, the actual mean noise at each tube current value shows relatively little fluctuation.

Fig. 8.

Measured noise vs. x-ray tube current for phantoms with glycerin (Χ² = 0.20, p>0.99)

Fig. 9.

Measured noise vs. x-ray tube current for phantoms with iodobenzene (Χ² = 0.30, p>0.99)

4. DISCUSSION

Mechanical phantoms provide an easy, cost-effective method to assess low-contrast performance of CT scanners. The lack of a definitive standard gives researchers many possible choices for designing suitable (size, cost, material, etc.) phantoms. Since the contrast in the clinical setting is generated by a combination of both the differences in physical density and atomic number, the best evaluation method of contrast should include both concepts. This design generates contrast based on both principles and is better than contrast phantoms based on the partial volume effect where the contrast is achieved by adjusting the layer thickness in the image slice and for which the contrast is slice thickness dependent. With the proposed method, any contrast level can be easily achieved by adjusting the proportion of additive without affecting parameters for image acquisition.

5. CONCLUSIONS

This paper sought to provide information and a method to design and fabricate precision low-contrast phantoms with smaller degrees of contrast than were previously available. Using the chosen materials, durable and stable LC phantoms were constructed inexpensively with a simple procedure. The phantoms were made of long-term stable solids which offer different contrast steps. This work established that any desired contrast level can be achieved. Phantoms with a 0.1% contrast or even less can be created using the equations relating the amount of required additive and the resultant absolute contrast. If desired, larger contrast values may be obtained easily as well. The phantoms have the advantages of low cost, extreme versatility, an easy manufacturing process and CT number stability. Because the new low-contrast phantom has been demonstrated to allow very precise levels of contrast for different target sizes, it should prove to be very suitable for 1) measuring a CT scanner’s low-contrast sensitivity, one of the most important image quality parameters, and 2) monitoring the low contrast discrimination of clinical and research scanners. In addition, the phantoms allow one to investigate the atomic number and physical density contrast separately, which should prove of value to the research and quality control communities.