Anthropomorphic liver phantom with flow for multimodal image-guided liver therapy research and training

Regarding the commercially available liver phantoms, the Triple Modality 3D Abdominal Phantom (CIRS Inc., Norfolk, VA, USA) [5] provides a training model for ultrasound-guided needle insertion and navigation technology. This represents a slice of the abdominal cavity that includes parts of the liver, tumor models and blood vessel structures. The CIRS phantom allows for multimodal medical imaging (US, CT and MR). Another commercial phantom, representing the upper abdominal organs is the IOUSFAN\({\circledR }\) (Kyoto Kagaku Co., Ltd, Kyoto, Japan) [6]. This is designed as a practice and demonstration tool for abdominal intraoperative and laparoscopic ultrasound. It has realistic appearance and provides tumor models and blood vessels structures.

Most of the liver phantoms developed for research purposes have been intended for CT imaging applications [7‐13]. Fewer phantoms have been developed for MRI [14, 15] and ultrasound [16‐18]. Some of the latter include vascular structures [15‐17] and have a realistic appearance [16, 17]. Multimodal liver phantoms are still scarce [3, 4, 19, 20]. To the best of our knowledge, the present study is the first describing a multimodal liver phantom based on real human liver anatomy, which includes tumor models, vascular structures and blood flow functionality.

In the literature, a multitude of techniques and tissue-mimicking materials have been proposed to prepare phantoms. The bulk matrix materials for soft tissue mimicry have been based on aqueous suspensions, agarose, gelatin, magnesium-silicate, oil gel, polyacrylamide gel, polyurethane resin, polyvinyl alcohol (PVA), polyester resin, epoxy resin, polysaccharide gels TX-150 and TX-151, polyacrylamide and room-temperature-vulcanizing (RTV) silicone [21‐23]. Agarose- and gelatin-based gels (hydrogels) are the most widely used alternatives for soft tissue modeling described in the literature [21, 22]. They are well-characterized and can be prepared with a range of acoustic properties [21], but they usually have a limited life span [23]. Oil gel-based phantoms have a propylene glycol or ethylene glycol base. The main advantages of propylene glycol are its resistance to bacterial colonization and its linear increase of speed of sound and attenuation with the proportion of propylene glycol. They also contain a gelatinizer and polymethyl methacrylate microspheres [24]. Ethylene glycol-based oil gels tend to be too dense, with low attenuation. Polyurethane, polyester and epoxy resins have good mechanical characteristics for mimicking soft tissue [24]. Standardization of the polyurethane gel-based phantoms production, however, is problematic due to complex molecular design of the gels. PVA-based models (cryogels) are of low cost and indefinite longevity, but the production process is highly demanding [25, 26]. Polysaccharide gels are inexpensive, conveniently moldable and temporally stable [27], but it may be difficult to avoid air bubbles, which constitute an issue for ultrasound imaging. Polyacrylamide gel-based tissue substitutes are made by polymerization of highly toxic acrylamide monomer [28], therefore, rendering special precautions mandatory. RTV silicone is easy to work with [29] and provides a soft rubbery texture similar to that of tissue. The major shortcomings of this material are high cost and extended hardening time.

The four most common materials used for ultrasound scattering particles are lipid microparticles, polymer microparticles, metal oxide powders and quartz glass microspheres. Lipid microparticles of 10–500 nm are biologically analogous to bilipid membranes of cells and organelles, which are believed to cause scattering in tissue. Commercially available lipid-based scatterers are milk [30, 31], fat/oil/lipid [32] and Intralipid/Nultralipid [33‐35]. Polymer microspheres of 50–100 nm are produced in well-controlled sizes, which means that repeatability and predictions of spectra are good [23]. Wide availability of TiO\(_{2}\) powder, 20–70 nm, makes titanium dioxide one of the most commonly used scatterers. However, it settles when not stirred, which is a problem when fabricating aqueous suspensions. Therefore, TiO\(_{2}\) powders should preferably be used with gelatin- or agarose-based, RTV and resin phantoms. The use of quartz glass microspheres (250 nm) is less well established [23].

All organs in the human body have a vascular blood supply. There is at least one arterial blood vessel leading blood into the organ, and there is at least another taking blood away from it. The connection between the arterial and venous vascular trees is through a capillary network, on a microscopic level.

Due to its accurately preserved shape, a cadaveric human liver specimen was used as the starting point (Fig. 1). After careful preparation and dissection, a three-part silicone mold was made from the liver using Elasturan 4503 (BASF Polyurethanes GmbH, Lemförde, Germany).

Polyurethane was chosen to develop the parenchyma due to its durability and liver-like consistency. To this end, Elasturan 6005/264 (BASF Polyurethanes GmbH) -100 pbw (parts by weight), and ISO 136/131 -20 pbw- with 0.6% Sephadex G2580 (Sigma-Aldrich, San Luis, MO, USA) bw (by weight) were used as a volume spreader. Components were mixed by hand, and subsequently vacuumed at 15 mBar pressure to eliminate air bubbles.

Three different types of tumor models were created and integrated in the phantom: metastatic lesions, hepatocellular carcinoma and benign cysts. The basic material for the tumors was the same as for the parenchyma with the addition of 5% calcium carbonate bw (0.5 g calcium carbonate powder/10 g of polymer). Cyst walls were made using parenchyma material only.

A two-part silicone mold was produced using an anatomic specimen as basis, modeling the body of the gallbladder, the cystic duct and the common bile duct. The right and left bile ducts run approximately in parallel to the right and left branches of the portal vein. Parenchyma material was used to create the gallbladder.

The two tubular vessel models were secured in place in the bottom two portions of the liver mold. The gallbladder was secured in place. The two vascular tree casts of the portal and the cava system, respectively, were inserted in the open end of the corresponding vessel model and fixed. The free ends of the two cast trees were joined together manually by softening/melting the paraffin. Care was taken to keep the corresponding bile ducts alongside the branches of the portal vein. The parenchyma substitute was poured into the mold in three layers, and tumor models were added at vessel junctions in layers 2 and 3 (Fig. 3). Before addition of the tumors, each layer was vacuumed for 7 min at 15 mBar pressure. Complete curing was allowed between each layer. Layer 3 was added after the 3rd, and top part of the mold was secured in place. The final phantom cured for 5–7 days.

Once cured, the phantom was placed in an 80\({^{\circ }}\)C water bath for 1/2 h to allow for the paraffin to begin melting. The phantom vascular tree was flushed with running water of the same temperature, first through the vena cava, allowing the paraffin to flow out through the portal vein. The same process was repeated starting with the portal vein, and continued until only clean water was seen to emerge from the vena cava. On the finished phantom, six depressions were carved on the anterior surface corresponding to the six front lobes of the liver. The depressions, 1.5 mm deep and wide, serve as positional markers for navigation. They are clearly visible by CT and MR. The depressions were stained, each with a different color, for accurate visual identification (Fig. 4). An electromagnetic (EM) tracking sensor was incorporated inside the parenchyma 5 mm below the surface for use with the navigation system (Fig. 4). The navigation platform, CustusX [38], is capable of importing preoperative data like CT, reconstructing these into 3D, segmenting structures from the CT data, reading in real time the tracked ultrasound probe, multimodal visualization techniques, registration of images to the phantom (or patient) and also image-to-image matching techniques. Navigation is further explained, and example images are shown in the Results section.

CT images of the phantom were acquired using Dyna CT (SIEMENS, Berlin, Germany), Siemens Artis Zeego and a liver scan protocol with 629 slices of 512 \(\times \) 512 pixels. The image resolution was 0.422 \(\times \) 0.422 mm, and the slice thickness was 0.6 mm with 0.3 mm between slices.

MR images were acquired using a 1.5 Tesla scanner (SIEMENS) with a T1-weighted Abdomen Liver scanning protocol. Thus, 160 images were recorded, each with 512 \(\times \) 512 pixels. The image resolution was 0.5 \(\times \) 0.5 mm, and the slice distance was 1.0 mm.

Ultrasound images of the phantom were acquired using the Ultrasonix MDP scanner (Analogic Ultrasound, Canada) with a laparoscopic probe (4-way flexible linear array probe, Vermon, Tours, France) (Fig. 9).

According to the difference in time of the received signal, sound speed of Elasturan was determined to be 1425 m/s (reference value of soft tissue is 1480 m/s). Attenuation at 5 MHz was 2 dB/cm/MHz, whereas at 2 MHz it was 1 dB/cm/MHz. Attenuation was not affected by the concentration of Sephadex. The concentration of calcium carbonate added to the parenchyma mixture to create the tumors caused realistic looking contrast between tumor and parenchyma with minimal shadowing. The benign cyst models appear anechoic in the ultrasound images, while metastatic lesions appear hyperechoic.

Figure 6 shows the Color Doppler ultrasound of internal vessels of the phantom with flow, and Fig. 7 shows the Power Doppler ultrasound of a vessel structure next to a cyst model. Larger and smaller vessels were identified. The outflow through the vena cava was regulated by using a clamp in order to provide enough pressure in the portal vein to ensure blood flow through the smaller vessels in the liver model.

The phantom was evaluated using the module for laparoscopic surgery in the research navigation platform CustusX (SINTEF, Trondheim, Norway, http://​www.​custusx.​org) [38].

CT images of the phantom were imported into the navigation system and visualized in both 2D cross-sectional slices and a 3D model. The digital ultrasound video stream was enabled by a connection to an Ultrasonix MDP scanner. A model of the ultrasound probe and the real-time ultrasound images were visualized using CustusX. The ultrasound image was tracked with an electromagnetic tracking (EMT) sensor (Aurora, NDI, Ontario, Canada) mounted at the tip of the probe (Fig. 10). The position of the ultrasound image in relation to the sensor was calculated through a calibration procedure [39]. The ultrasound probe and images were aligned in the corresponding position and orientation to the CT images through a landmark-based registration procedure. Three markers on the liver phantom surface (described in the phantom assembly section) were pinpointed both in the CT image and in the tracking system by using an EMT pointer. The registration matrix was obtained by minimizing the total distance between the corresponding points.

Figure 11 shows the navigation scene from the CustusX navigation research platform. After calibration, the real-time ultrasound and preoperative CT images were visualized together in the same images using the navigation pane. This included a 3D scene with the ultrasound overlaid onto the CT image. This produced a 2D slice of the CT image projected to the spatial positions of the ultrasound image sector, thus visualizing the part of the CT volume corresponding to the real-time tracked ultrasound image.