An anthropomorphic deformable phantom of the vaginal wall and cavity (2025)

Cervical cancer accounts for an annual 604,127 new cases and 341,831 deaths worldwide. This is equivalent to 7.7% of deaths in women due to cancer [1]. Early stage cervical cancer is typically treated by surgery, including conisation and (radical) hysterectomy. The extent of surgery depends on the tumour stage, defined by the Féderation Internationale de Gynécologie et d'Obstétrique (FIGO) and the Union for International Cancer Control (UICC) TNM systems. In locally advanced and lymph node positive cervical cancer, chemoradiation consisting of external beam radiotherapy combined with weekly cisplatin chemotherapy followed by brachytherapy (BT) is the standard of care. BT plays an important role in the radiotherapy management as it delivers a high and conformal dose to the tumor with maximal sparing of surrounding normal tissue [2]. Brachytherapy is a complex and rapidly developing field, branching to disciplines such as image analysis and radiation physics. Examples of developments include automatic tumour segmentation [3], automated treatment planning [4], in-vivo dose monitoring [5], and 3D-printing of patient-specific applicators [6–8].

To minimize patient burden during technological development and evaluation, some steps can and should take place in a preclinical setting. This can be done using either computer-based models, physical patient models or phantoms. For gynaecological BT, a model of the female pelvis is needed that includes the vaginal cavity, vaginal walls, cervix, uterus, and possibly surrounding organs (bladder, rectum, sigmoid) and ligaments. Model complexity may depend on assessment objectives. Although female pelvis phantoms are commercially available, they tend to be expensive, focus on medical imaging training, and present a single idealised anatomical configuration. In addition, they often contain thermoplastic or thermosetting polymers that do not match biomechanical tissue properties. As a result, these phantoms are particularly ill-suited to test personalised health solutions and cases in which inter-patient variability in topology or deformation responses are of importance.

The female reproductive system is built to sustain astonishing deformations, especially during pregnancy and labour [9]. Biomechanical tissue properties are complex and hormone-regulated and vary as a result of pregnancy, age, menopause, BMI, and pelvic floor disorders, e.g. pelvic organ prolapse (POP) [10–15]. For the vaginal wall, it has also been shown that smooth muscle content decreases in women with POP [16]. Properties of healthy vaginal wall tissue were examined in a limited number of studies. Typically, stress-strain relations were derived by means of uniaxial tensile testing, results of which are shown in figure 1. These stress-strain relations were often summarized with a Young's modulus. Reported values of healthy vaginal tissue ranged between 6.7–10.5 MPa [13, 17]. However, this linear biomechanical metric is inadequate [12], as the stress-strain relations are hyperelastic.

Related work in phantom design provided physical models of the male pelvis [18–20]. In addition, non-deformable female pelvis models were made for optimisations in radiomics [21]. Nattagh et al developed a phantom to train transrectal ultrasound image-guided needle insertion and suturing of the cervix [22]. Kadoya et al developed a deformable phantom of the vaginal cavity, uterus, bladder and rectum to assess deformable image registration [23]. Campelo et al made a phantom for BT training based on direct 3D phantom printing [24]. Although this is a convenient production method, mimicking biomechanics can be difficult as printable elastomers available today are still limited. The last two articles evaluated Shore hardness values via indentation tests. No articles were found in which large material strains were assessed and used to develop female pelvis phantoms.

Deformable materials used in pelvis models included silicone and polyvinyl alcohol (PVA) [18], agarose gel and silicone [19], carrageenan gel [20], gelatin and rubber [22], urethane and silicone [23] and 3D printable photopolymers [24]. Silicone is a widely used tissue-mimicking material with a high durability [25]. In addition, it has been shown that PVA can be suitable to mimic needle-tissue interactions [26], which may become relevant in BT tasks when interstitial needles are used. We therefore decided to evaluate both silicone and PVA materials.

The vaginal cavity topology was derived from anonymised MRI data, see figure 2. Treatment planning software (Oncentra-Brachy, Elekta, Sweden) was used to segment the vaginal cavity. Contours were saved as DICOM RT-structure files and a surface mesh model was constructed using the SlicerRT extension of 3D Slicer [27]. This model was exported as an STL file and processed further in a computer-aided design program (SolidWorks, Dassault, Systemes, USA). Based on the 3D model of the female pelvic described by Kraima et al [28], the outer vaginal wall contour was constructed by non-uniform scaling, using a 2 mm wall thickness. The mould was created using the SolidWorks cavity function. At this stage, surrounding tissue support was simplified by suspending the model in mixtures of PVA and silicone with a lower stiffness.

Three types of moulds were created from polylactic acid (PLA) with a 3D filament printer (Ultimaker 3, Ultimaker, The Netherlands). To determine stress-strain curves, modified dog-bone samples were made, as shown in figure 3(A). Four Silicone 940 samples, ten Silicone 940+thinner samples, and ten PVA 12% samples were made. The cross-sectional surface area of these samples at the strained region was 30 mm² (10×3 mm). To evaluate material visibility in MRI, blocks (length, width, height=50, 50, 40 mm) with cylindrical holes (radius, height=10, 25 mm) were made, as shown in figure 3(C). Finally, anthropomorphic phantom moulds were created (figure 3(D)). The mould contained four interlocking parts, surrounding a core. In addition, outer support structures were printed for the floor and top of the mould, to keep parts together. The core had the shape of the vaginal cavity and was created using a water-soluble PVA 3D printing filament. The core was centred with four bolts. The mould contained four filling reservoirs at the top, with slanted channels running to the enclosed space. The mould was filled in sequence with silicone and PVA. Air was able to escape via the space between the mould and the central core. Figure 3(E) shows the fully assembled mould with silicone in the reservoirs. After curing of the silicone and PVA phantoms, the structure was suspended in an acrylic container (figure 3(F)). The container was filled with the softer supportive material to facilitate phantom handling and positioning. Finally, the insert was dissolved in water and the phantom was ready to use (figure 3(G)).

Data were recorded at a sample rate of 200Hz. Stored data were post-processed in Matlab (R2019b, Mathworks, USA) by means of a moving average filter with a kernel size of 50. Stress was defined as the force normalised per unit of cross-sectional area of the sample, in N m⁻² or Pa. Strain was defined as the change in length divided by the original length (L), i.e. ΔL/L or l-L/L (-), where l is the final sample length.

Tissue biomechanics were summarised using the Young's modulus and the area under the mean stress-strain curves. The slope of the stress-strain curve was determined with a linear least squares fit and evaluated along a running window of 1000 data points (5 s). This enabled a search for the largest slope in the 20%–60% strain region, which was considered to be the Young's modulus. The area under the stress-strain curve was determined using a trapezoidal numerical integration of the mean stress-strain curve for strains up to 60%.

Dosimetry plans in BT are often constructed with the use of MR images. We therefore evaluated the contrast of adjacent phantom layers using an Optima MR450w 1.5T scanner (GE Healthcare, Chicago, USA). Samples were placed on an MR-compatible tray (figure 4(A)). The second coil was positioned above the tray in accordance with standard pre-BT MRI treatment protocols. Standard MRI sequences used in gynaecological BT were used for imaging, of which the T2-weighted fast recovery fast spin echo scan (FRFSE) with TR/TEeff=2986ms/103ms and T1-weighted 2-point Dixon radio-frequency spoiled gradient echo scan (LAVA Flex) using TR/TE/flip angle=6.7ms/4.5ms/10° were selected to compare phantom layers, see figures 4(B) and 4(C).

In addition to the phantoms, standardised blocks of silicone and PVA were produced, see figure 3(C) and figure 4(A). These blocks contained a cylindrical cavity to quantify contrast with ultrasound gel (supragel, LCH medical products, France), which is used in clinical practice to distend the vaginal cavity and enhance contrast to facilitate contouring [29]. Image data were stored in DICOM format and contrast was analysed in Matlab. For each sample layer, the average of four regions of interest was determined, as shown in figure 4(C). Air pockets and shadows, which were mostly seen in the ultrasound gel filling, were avoided.

Contrast was determined in normalised FRFSE and LAVA Flex images, in which pixel intensity ranged between [0,1]. Contrast was defined as ∣Ĩ_FG-Ĩ_BG∣, where Ĩ_FG is the mean pixel intensity of the foreground layer and Ĩ_BG the mean pixel intensity of the background layer. To summarize material visibility, contrast values were averaged for samples that compared the same adjacent materials.

The results of the uniaxial tensile tests are shown in figure 5(A). In addition, the area of interest of stress-strain curves of vaginal wall tissue of healthy patients is shown, as derived from literature (see Introduction). The curves are furthermore summarised by means of the Young's modulus (YM) and area under the curve (AUC) in table 1. From the curves, it can be seen that the silicone samples resemble the vaginal wall biomechanics, in particular when low to moderate elongations (<50%) are applied. The PVA samples have a much lower stiffness. Figure 5(B) shows the effect of preconditioning for Silicone 940 with a thinner. It appears that preconditioning increases the non-linearity of the stress-strain curve. This is visible from the Young's modulus, which increases rapidly (table 1).

Data on the relative contrast of structures in multi-layered phantoms in MR images are presented in table 2. In addition, the contrast of these materials relative to ultrasound gel is shown. The multi-layered silicone and PVA structures could best be distinguished with the LAVA Flex sequence (contrast in silicone: 0.11, and in PVA: 0.12). The contrast was much lower in the FRFSE sequence (silicone: 0.01 and PVA: 0.04).