Image-guided interventions rely on real-time imaging and the fusion of data sets with previous images acquired by CT and MRI. Image registration algorithms have to take up various challenges such as varying resolutions, slice thicknesses and altered patient positions between the different acquisitions. Anthropomorphic phantoms for validation of algorithms under near real-world conditions serve as an important additional stage between geometric phantoms and in vivo studies.

The anthropomorphic thorax phantom consists of four main modules placed in an acrylic case that can be flooded with water for MR imaging (Fig. 1).

A lung module is built from two breathing bags (volume of 2 l each) filled with cubical natural rubber foam (side length 5 or 20 mm). Inflation can be performed manually by a connected resuscitator. Inferiorly, a movable diaphragm is retracted by rubber bands and causes deflation of the breathing bags.

Two balloon-shaped liver modules (weight 1500 g each) are filled with distilled water. Approximate tissue parameters are achieved by further adding 2.37 ppm gadoterate meglumine (T₁ modifier), 1.3 % agarose (T₂ modifier), 3 % carrageenan (mechanical stabilizer) and 0.03 % sodium azide (preservative) based on Hattori et al.¹ Na⁺ concentration was set to 20 mM.² Several spherical inserts representing tumors (30 mm diameter) were added to one of the liver modules. Na⁺ concentrations of the inserts were varied in a physiological range (30/40/80/154 mM Na⁺).

Two rib cage modules can be inserted on top of lung and liver modules. They either consist of artificial ribs (dimension 15×10×200 mm³) made of epoxy (structural support) and 19.6/39.2 % CaCO₃ (HU modifier) or porcine ribs.

Isometric tracking spheres consisting of an MR visible sphere (30 mm diameter) enclosed by a CT visible shell (40 mm diameter) were also designed (Fig. 2).

CT data were acquired with a Somatom Force (Siemens Healthcare, Erlangen) using a convolution kernel Br36d (kVp=110 kV). A ROI-based HU analysis with ImageJ was conducted. CT images were rendered to 3D volume data sets using ITK software.³ The motion of the diaphragm was evaluated at two positions during five breathing cycles.

MR imaging was performed on a 3 T whole-body scanner (Magnetom Skyra, Siemens Healthcare, Erlangen) with a 16-channel body coil using turbo spin echo sequences (TE=[15, 30, 40, 50, 60, 120, 240] ms, TI=[250, 500, 800, 1000, 2000, 5000] ms, 2.3 mm resolution, 300×300 mm² FOV). T₁ and T₂ relaxation times were obtained by pixel-wise fitting. ²³Na imaging was performed with a double-resonant transmit receive array using a density-adapted three-dimensional radial gradient echo sequence (TR=100 ms, TE=0.49 ms, 4.0 mm resolution, 348×348 mm² FOV, 10000 spokes) as presented by Haneder et al.⁴

The HU values of the lung filling (-801 to -668 HU) are comparable to human lung tissue⁵ and depend on the state of compression of the foam cubes similar to human lungs. The diaphragm moves 26 mm on average during breathing cycles which is in accordance with literature values (18.8 mm to 38.2 mm).⁶ The liver filling has a T₁ relaxation time comparable to human liver (measured 790±28 ms, literature 812 ms ⁶) with a relative standard deviation of 3.5 % (Fig. 3). The T₂ relaxation time is larger than human liver values (measured 65±1 ms, literature 42 ms ⁷, Fig. 3). However, the T₂ relaxation time is adjustable during future construction by increasing the agarose concentration. Spherical inserts representing tumors are well distinguishable in ¹H and ²³Na MR imaging (Fig. 4). HU values of the artificial ribs (218±56 HU/339±121 HU) are comparable to human bone⁴ and offer a simple geometric structure suitable for registration algorithm validation. Porcine ribs are geometrically more complex and essential for near-realistic bone imitation. The tracking spheres are well detectable in both CT and MRI and serve as multimodal landmarks. The parameters of the tracking speres can be adjusted in the following ranges to receive a distinct signal: HU value from 150 HU to 900 HU, T₁ relaxation time from 550 ms to 2000 ms, T₂ relaxation time from 40 ms to 200 ms.

This work proposes an anthropomorphic multimodal thorax phantom which fulfills the demands of a simple, inexpensive system with interchangeable modules. It enables experimental studies for quantitative evaluation of multimodal, dynamic and deformable image registration as well as multi-nuclear MR imaging techniques. The modular design permits to complement the present setup with additional modules. In the future, this phantom will be used for validation of quantitative MR measurements (perfusion, density, diffusion) which is in line with current developements for reliable and comparable data in addition to diagnostic images.