1.

Introduction

2.

Methods

2.5.

Chest Wall with QC Metrics

The uniform chest wall region was designed to be customizable to include conventional QC test objects typically found in conventional, uniform phantoms. Prior to printing, a 35-mm piece of uniform “adipose” region was added to the chest wall of each breast slab, as shown in Fig. 1. Several hollow (air) features were created, including 3.35-mm diameter hollow columns for alignment and assembly of the slabs, and 2.5-mm hollow columns and 1.0-mm hollow spheres to act as air contrast fiducial markers. Solid, embedded features included a “glandular” sphere and rectangular prism for uniformity, geometric distortion, and signal-to-noise ratio, and a scale bar ranging from adipose to glandular in 20% increments to test the linearity of contrast.

Fig. 1

Photograph of two halves of hybrid breast phantom: (a) inferior and (b) superior, with uniform chest wall region shown by red box that can contain conventional QC test objects to supplement the anthropomorphic anatomy section.

In addition to 3-D-printed features, metal ink stickers were created as 2-D inserts using a commercial printing service (MorningPrint Inc., Irvine, California). The stickers were designed with line pair starburst patterns, circle and square test objects to evaluate modulation transfer function (MTF) or artifact spread function (ASF), and fiducial markers. The line pairs range from 1.5 to $10\mathrm{lp}/\mathrm{mm}$ and are intended for technologists to make quick, visual assessments. They also demonstrate the flexibility, robustness, and high resolution of the metal ink printing process. These metrics are already well-established in other phantoms, therefore, they are only qualitatively demonstrated in the context of this hybrid design. As an example, the line pairs were used to evaluate and compare the resolution of the Hologic Selenia Dimensions system in four modes: contact mammography, $1.8\times$ magnification view mammography, and DBT parallel or perpendicular to tube motion direction. The resolution of each mode was determined based on the consensus of three observers. Fiducial markers were added to allow for potential automation of this process.

3.

Results

All our phantoms use two materials to represent adipose versus glandular tissue, but there have been substantial improvements in the modeling of those tissue compartments, as shown in Fig. 2. In the first generation phantom, the volumetric data were binarized into adipose versus glandular surfaces, resulting in loss of small details and abrupt edges.¹⁵ The second generation phantom was segmented into four surfaces, which were printed using a commercially available option to create mixed, intermediate versions of the two extreme materials.¹⁶ Finally, the current study used the voxelized and dithered approach and also switched to the higher-resolution XCAT Cohort II data, providing the finest details and transitions between adipose and glandular regions.

Fig. 2

Three generations of breast phantoms show increasing detail and smoother transitions: (a) binary surfaces, (b) four surfaces, and (c) voxelized and dithered. Images are close-up photographs of $25\times 40\mathrm{mm}$, taken from a 10-mm-thick slab.

Jf Flexible resin was doped with zinc acetate or tungsten to increase the attenuation. Figure 3 depicts the photopolymer resin doping results. With 3 wt. % zinc acetate, printed samples achieved $\mu =0.68/\mathrm{cm}$, matching the attenuation of fibroglandular tissue. In subsequent repeat prints, however, the attenuation dropped to $0.58/\mathrm{cm}$, which suggested clogging or precipitation problems. Tungsten-doped Jf Flexible at 1.6 wt. % achieved $\mu =0.75/\mathrm{cm}$, which exceeded that of fibroglandular tissue. Although volumetric breast density is a fraction between 0% and 100%, this material can be thought of as 140% on that scale. Since materials available to represent adipose tissue are limited to $\sim 40\%$ volumetric breast density on the low end, this higher 140% density can qualitatively demonstrate the expected amount of visual contrast between adipose and glandular tissue. Another option to represent fibroglandular tissue is the new commercial material VeroPureWhite that was measured at $\mu =0.65/\mathrm{cm}$, 92% volumetric breast density.

Fig. 3

Plot of linear attenuation coefficient versus dopant concentration for Jf Flexible resin doped with zinc acetate or tungsten, imaged at W/Rh 28 kVp on the Hologic Selenia dimensions system. The horizontal dashed lines represent 0% and 100% fibroglandular tissue, $\mu =0.44$ and $\mu =0.67$, respectively.

Scatter fractions for 4 cm of printed material were 7.2% for TangoPlus and 7.4% for VeroPureWhite, compared to 6.0% for adipose and 6.2% for glandular reference materials. By substituting adipose with TangoPlus, or glandular with VeroPureWhite, the difference in scatter intensities would amount to only 1.2% of the background pixel values, compared to 1.8% for the background noise of an ACR accreditation phantom imaged on the same system under AEC conditions.

A modular insert design was developed as shown in Fig. 4, which enabled for the incorporation of lesions without and with iodinated contrast into both the tungsten and VeroPureWhite phantom versions.

Fig. 4

Photographs of modular lesion insert design created with VeroPureWhite (white) or tungsten-doped (black) materials: (a) healthy breast anatomy, (b) breast anatomy with masses in contrast-detail patterns without contrast, and (c) iodinated inserts with colorless iodobutane masses on top and CIRS iodinated disks on bottom.

Figure 5 shows both the tungsten and VeroPureWhite phantom versions portraying normal anatomy imaged on the Hologic system in mammography and DBT modes. Note the higher visual contrast offered by the tungsten version.

Fig. 5

Breast phantoms with normal anatomy printed with (a) and (b) Jf Flexible to mimic adipose tissue and VeroPureWhite and (c) and (d) tungsten-doped Jf Flexible to mimic fibroglandular tissue. (a) and (c) A DBT reconstructed slice image of both breasts and (b) and (d) mammography image of both breasts.

Figure 6 displays the same phantoms but with the center 0.5-cm slab of normal anatomy replaced the lesion insert with addition of masses in four sets of contrast-detail patterns. Within each $4\times 4$ set of masses, each of the 16 masses has unique morphology that can be adjusted parametrically and vary in contrast parallel to the chest wall and size in the chest wall to nipple direction (left to right as shown in Fig. 6).

Fig. 6

Breasts imaged with the mass contrast-detail insert. Tungsten-doped breast on the top and VeroPureWhite breast on the bottom. (a) Simulated mammography image of the virtual phantom containing the lesion insert with exaggerated contrast to illustrate lesion location and details, (b) DBT reconstructed slice image of the phantoms, and (c) mammography image of the phantoms.

Figure 7 demonstrates the imaging of the phantom with lesion inserts featuring targets with iodine. As before, the mass contrast-detail pattern clearly shows the range of contrasts and sizes of the masses, and the CIRS iodinated cylinders also show differing levels of conspicuity. Since this clinical system was not optimized for contrast-enhanced imaging, these images were taken at the maximum energy allowed for mammography and DBT to demonstrate iodine signal qualitatively.

Fig. 7

Iodinated lesion insert: (a) mammography image of the mass insert by itself to accentuate lesion appearances, 39 kVp, (b) DBT reconstructed slice image of the mass insert sandwiched in middle depth of the whole breast phantom, 49 kVp, and (c) DBT reconstructed slice of iodinated discs sandwiched in whole breast, 49 kVp. Iodine concentrations are indicated in wt. %.

The calcification insert images in a uniform, tissue-equivalent background show the expected decrease in calcification detection with mammography and DBT imaging compared to the optical scan ground truth (Fig. 8). Based on that optical scan, the insert contained 276 measurable calcifications with major axis length ranging from 53 to $710\mu \mathrm{m}$ with an average size of $200\mu \mathrm{m}$. These sizes are clinically relevant to typical breast calcifications that range from 50 to $500\mu \mathrm{m}$.²² Based on this subjective experience, a breast imaging radiologist confirmed that the radiographic appearance realistically mimicked pleomorphic or coarse heterogeneous morphology.

Fig. 8

Calcification insert images, showing a cropped $6\times 6\mathrm{mm}$ region: (a) optical scan used as ground truth, (b) mammography, (c) magnification view, and (d) DBT reconstructed slice.

When the line pair stickers were also imaged in a uniform, tissue-equivalent background, there was a similar, expected decrease in resolution from mammography and magnification view at $6.5\mathrm{lp}/\mathrm{mm}$ to DBT images at $4\mathrm{lp}/\mathrm{mm}$ when oriented either parallel or orthogonal to the tube motion direction (Fig. 9). The larger objects were designed to evaluate quantitative metrics such as MTF and ASF, but that is beyond the scope of this work.

Fig. 9

Metal sticker test objects used to measure standard QC metrics: (a) DBT reconstructed slice image with line pairs oriented parallel to tube motion direction ($4\mathrm{lp}/\mathrm{mm}$), (b) DBT reconstructed slice with line pairs oriented perpendicular to tube motion direction ($4\mathrm{lp}/\mathrm{mm}$), (c) magnification view ($6.5\mathrm{lp}/\mathrm{mm}$), and (d) mammography image ($6.5\mathrm{lp}/\mathrm{mm}$). Numeric line pair indicators are superimposed on (d) for display purposes only. The square and circle objects were not used in this study.

4.

Discussion

For decades, assessment of mammographic image quality relied upon subjective scoring of intentionally simplistic, uniform phantoms, namely the ACR¹ phantom in the US and CDMAM² in the EU. To assess breast imaging technologies such as DBT and contrast-enhanced mammography, there is a need for realistic, 3-D phantoms and new evaluation procedures. To help address this need, this study developed a breast phantom with three key components: a principal section with realistic breast anatomy, anthropomorphic lesions for task-based assessment, and a uniform region for conventional QC purposes. This phantom provided several major advancements over previous generations. This was the first physical rendition of the virtual phantoms from the Duke XCAT Breast Cohort II with higher resolution. In terms of materials, this study achieved targeted x-ray attenuation of fibroglandular tissue using new commercial and custom-doped photopolymer ink. A process of voxelized and dithered photopolymer inkjet printing preserved fine details and subtle tissue gradations. Furthermore, this study introduced several new lesion or task features, including multimaterial anthropomorphic masses, iodine-doped photopolymer printing, calcifications matched to optical scan ground truth, and metal ink sticker QC objects.

There are several broad implications arising from this study. First, these “patient-based” phantoms inherently mimic anatomy because they are based on high-resolution breast CT scans of human subjects. Although this study was based on a single virtual phantom, the process can be applied to other virtual subjects to create phantom “families” to represent anatomical variation. In a separate study, this was demonstrated by assessing lesion detectability in three different phantoms with thickness ranging from 5 to 8 cm and volumetric breast density from 10% to 39%.²³ Such realism and diversity may play an important role when evaluating the clinical performance of FFDM and DBT systems. Second, the addition of masses and calcifications represent a step toward task-based assessment within a real patient scenario. In a departure from the status quo of subjective scoring, automated QC methods such as model observers may be used to quantify detectability for a relatively small number of lesion tasks and do so in the context of different anatomical backgrounds. Finally, the uniform region was designed for various traditional QC evaluation tasks, including quantitative metrics such as MTF, noise power spectrum, or ASF, as well as qualitative metrics such as resolution line pairs. However, more work is needed to integrate those metrologies to account for the small size of the uniform region and any residual, nonhomogeneous texture caused by the printing process. By acquiring both traditional and new measures of image quality in the same image, one can study their relationships, which should facilitate adoption of these new phantoms and new procedures.

In addition to our work, many other physical breast phantoms have been proposed in recent years. An evaluation of the CIRS BR3D Model 020 phantom, ACR FFDM accreditation phantom, Penn phantom, and Quart mam/digiEPQC phantom found these phantoms to be unacceptable to evaluate DBT systems.⁶ Most of these commercial DBT phantoms are fundamentally limited by their uniform background, which cannot fully evaluate 3-D imaging performance. In addition to the ACR FFDM and quart, other uniform DBT phantoms include the CIRS Model 021, Gammex Modular DBT, Artinis EU DBT Test Set, and Phantom Laboratory Tomophan Phantom. To address the uniform background limitation, several newer phantoms introduced heterogeneous texture, such as the CIRS BR3D, Penn phantom, Leeds Voxmam, and some research phantoms;^5,8,9,24 however, they simplify texture as binary, random, and/or procedurally generated patterns that are still readily distinguishable from the real breast. Moreover, only a few research phantoms contain realistic tasks for breast cancer detection, such as Ref. 24 that included calcifications from the Leeds Voxmam and 3-D printed anthropomorphic masses, and Ref. 8 that added hand-crushed calcifications and inkjet-printed masses.

The physical phantoms previously presented by our group also did not include any lesion targets aside from 2-D inserts.^8,17 We now introduce several types of custom lesions and tasks to allow for nuanced study of how they interplay with varying local volumetric breast density and texture. First, microlobulated masses were arranged in contrast-detail patterns, but the masses are readily customizable including their shape, size, contrast, number, and spatial location,²⁵ all while retaining voxel-level ground truth because of the integrated, multimaterial printing process. Second, we used the same process with custom ink to create iodinated lesions, the first ever created using 3-D printing. This is a major advancement over the current commercially available iodinated targets, which are limited to uniform cylinders of resins or just liquid. Third, we reintroduced the use of manually crushed calcifications¹⁵ but added registration with high-resolution optical scanning to provide ground truth for calcification morphology and location. Finally, metal ink stickers were used to evaluate system resolution as they have a finer resolution and higher contrast than achievable using photopolymer 3-D printing. The metal ink is equivalent to 1 to 2 half value layers of attenuation at mammographic energies, so in addition to line pairs, this process may also be used to create high-contrast objects such as for measuring MTF or ASF.

Another problem of our earlier phantom versions was the limited contrast range of 36% to 64% volumetric breast density using the commercial photopolymer inks TangoPlus and VeroWhitePlus.²⁶ We raised the upper end to full glandular density with the new commercial ink VeroPureWhite and went beyond with custom, tungsten-doped ink. The doping agents investigated to increase the radiographic density were chosen because they had $k$-edges well outside the mammography energy range (W = 70 and Zn = 10 keV), were readily available, and were miscible in the photopolymer ink. For contrast-enhanced imaging, iodobutane was chosen for similar reasons.

There are limitations to our current phantom design. Our current inks can achieve or exceed the attenuation of 100% fibroglandular tissue because the addition of substances is limited only by solubility or miscibility. However, we cannot yet achieve volumetric breast density equivalent below 36%. We have tried to achieve low attenuation by adding the surfactant Span 60 and sonicating to create bubbles, without success.²⁷ Our group and others have tried printing only one compartment such as the glandular tissue, then backfilling the adipose portions manually with more tissue equivalent materials such as wax, but this requires greatly simplifying the anatomy and may still result in unacceptable levels of bubbles or artifacts.^5,15 In an analysis of 3-D printed materials for breast phantoms, acrylonitrile butadiene styrene used in fused deposition modeling printing was shown to maintain an absorption close to that of adipose tissue throughout the mammographic energy range;²⁸ however, there remains considerable work to develop adipose equivalent materials that can comply with the inkjet and curing process. For simplicity in this study, phantom materials were evaluated with a typical radiographic technique (W/Rh at 28 kVp). However, previous work validated that these photopolymers maintained their relative attenuation compared to tissue-equivalent reference materials over the full mammographic energy range.¹⁵ In addition, the phantoms in this study focus on a single-subject anatomy. In other studies, however, three other models have been fabricated based on subjects with different breast size, thickness, and density.²³ Finally, although various lesion tasks are presented, they may need to be refined for clinical application. Currently, lesions are restricted to the middle depth, but the phantom can be made in thinner cross sections such that lesions may be sandwiched at other depths. Similarly, calcifications are confined to a 2-D plane just like conventional phantoms, but potentially multiple layers can be stacked with spacer slabs to depict a 3-D cluster. Finally, protocols would need to be developed for task-based assessment in anthropomorphic physical phantoms, but that is beyond the scope of this study. The contrast-enhanced lesions have yet to be validated on a clinical dual-energy system to accurately create a subtraction image. Future work may be directed toward refining those lesion tasks, such as to create spiculated masses²⁵ or to create fixed calcification patterns using 2-D inkjet printing¹⁷ to integrate calcifications into the breast anatomy.

5.

Conclusion

This study represents significant improvements to our current phantom design. The commercial product VeroPureWhite and the custom, tungsten-doped Jf Flexible resin were presented to represent fibroglandular tissue. The commercial TangoPlus and the third party Jf Flexible resin without dopant were used to represent low-density adipose tissue. The modular approach allows for maximum versatility of the phantom including healthy tissue, lesions, contrast-enhanced lesions, and calcifications within the anthropomorphic region as well as a space for standard QC metric evaluation in the uniform chest wall region. Our fundamental goal is to create techniques that can provide flexible, customizable, and affordable options to fabricate phantoms that suit the needs of different researchers.