Keywords

1 Introduction

Nowadays the use of breast physical phantoms is related to the applications in diagnostic radiology related to optimization of patient dose, quality control of developed diagnostic systems (Breast Tomosynthesis, breast CT, contrast-enhanced mammography units), as well as for proof of concept and development of new breast imaging techniques [1, 2]. Until recently, the exploited phantoms in the routine clinical work were mainly represented by a homogeneous mixture, incorporating standard test objects dedicated for system performance evaluation.

Recent advanced research in the field of 3D printing materials and technology for printing phantoms dedicated to x-ray breast imaging showed the attempts to produce realistic anthropomorphic breast phantoms from patient images [3] and computational breast models [4]. These are also used in education approach for teaching purposes at advanced classes. An example is the advanced educational course in Medical Physics at European level EUTEMPE-RX, where at several modules the participants are trained to perform virtual studies with anthropomorphic phantoms [5,6,7,8].

One major problem with the 3D printing approach is related to the limitations in the current 3D technology, which, however, has made enormous progress during the last two years in terms of technology and materials focused on printing 3D anthropomorphic phantoms. For example, phantoms may now be printed by using inexpensive parchment paper and a radiopaque ink [9].

We have recently studied the absorption and phase-contrast characteristics of several materials commonly used with fuse deposition modelling and stereolithography 3D printers [10, 11]. These studies showed that most of the resins, Nylon, Hybrid, PET-G have absorption properties in the mammography range similar to these of the glandular tissue, while ABS shows absorption characteristics close to these of the adipose tissue. Based on these findings, we further developed a physical phantom, based on these materials. This paper reports on the process, related to the manufacturing of such phantoms and its evaluation for mammography purposes. The evaluated phantoms will be used for assessment of new breast imaging techniques such as phase contrast mammography.

2 Materials and Methods

2.1 Physical Breast Phantom

The main elements in manufacturing a physical breast phantom are depicted in Fig. 1a, while Fig. 1b shows the complete phantom, filled with water. Specifically, the external shape is a semi-cylinder container with sizes: radius of the semi-cylinder base 59 mm, and 49 mm height.

Fig. 1.
figure 1

Physical breast phantom: (a) main parts, (b) the manufactured physical breast phantom

Full size image

Two 3D printing technologies are used: fused deposition modelling (FDM) and stereolithography (SLA). The materials used for the production of the phantom are summarized in Table 1. Their properties are taken from Ivanov et al. [10].

Table 1. Materials used for manufacturing of the breast phantom.
Full size table

The semicylinder container was modelled with the help of DesignSpark Mechanical 2.0 (https://www.rs-online.com/designspark/mechanical-software), while the glandular tree is modelled with the BreastSimulator [12, 13]. The adipose compartments were based on segmented adipose models from tomography images of the breast [14, 15]. As there are no requirements for breast phantoms to contain pathology, the inclusion of such was not considered at this stage of the development process.

2.2 Experimental Setup

X-ray images of the physical phantom shown in Fig. 1 were acquired with a Siemens digital mammography unit featuring a detector with a pixel size of 0.085 mm × 0.085 mm. Two projection sets, each one comprising of ten images, were taken at tube voltages set to 22 kVp and 28 kVp, respectively. Two such mammography images are shown in Fig. 2.

Fig. 2.
figure 2

Mammography images of the phantom shown in Fig. 1b acquired with (a) 22 kVp (b) 28 kVp.

Full size image

The source to detector distance was 660 mm, while the incident exposure depended from the phantom thickness.

2.3 Evaluation

The evaluation has been carried out with an in-house developed tool dedicated for feature extraction from x-ray images [16]. Figure 3 shows the main window of the tool. The user interface is implemented in MATLAB 2013a environment using the GUIDE module [17]. In short, the tool is divided into five regions related to different activities and functionalities, such as (a) various possibilities to load DICOM and other types of images, (b) possibilities to adjust contrast and brightness by using sliders, to zoom in/out the image and examine the details of specific pixel, (c) functionalities related to manually or automatically defining the regions of interest (ROI) in the selected image, (d) use of tools for automatic creation of ‘mask’ image of the area of the breast, where the ROIs are populated, (e) calculation of features based on the selected ROIs.

Fig. 3.
figure 3

A screenshot of the main window of the tool, used in the evaluation.

Full size image

Texture of a breast in mammogram images plays significant role in tissue classifications. This tool is capable of evaluating more than 20 statistical and other feature types. In this study, we used six features of interest, which description is shortly summarized in Table 2.

Table 2. Feature description.
Full size table

The evaluation of the phantom has been performed by comparison of features extracted from x-ray mammography images compared to those extracted from 20 mammography patient images. The described features are performed over randomly selected ROIs in the analyzed images and further averaged and expressed by a mean value and standard deviation.

3 Results and Discussion

For the calculations of the features, we used ROIs of size 128 × 128 pixels and 50% overlap of the regions. All ROIs are placed “inside” the projected breast, excluding skin, muscle and background. Table 3 summarizes the results for the calculated features for each set of ten images, acquired at 28 kVp and 22 kVp, respectively in comparison to the features extracted from twenty patient mammography images, free of lesions.

Table 3. Comparison of features extracted from phantom and patient images.
Full size table

The comparison of features shows a good coincidence between features from phantom and patient images. Better agreement is observed for the phantom images obtained with 28 kVp, which is expected since this energy is traditionally used with mammography units. In addition, comparison of distributions of selected features is shown in Fig. 4a–c.

Fig. 4.
figure 4

Distributions of the computed (a) kurtosis, (b), skewness and (c) fractal dimension distributions for the studied images.

Full size image

4 Conclusions

The conducted comparative evaluation the created physical phantom and the patient mammograms shows similarity between features extracted from their mammography images. From the studied features extracted from mammography images obtained from the two phantom sets of images (taken at 22 kVp and 28 kVp), the use of 28 kVp with the proposed phantom - glandular tree, printed from Gray resin; Adipose compartments, printed from Clear resin and container, printed from PLA - showed close results to these from patient mammography images.