Tissue mimicking phantoms are important tools to study the impact of trabecular bone structure on data interpreted from MRI such as the apparent trabecular relaxation time^1,2 and bone integrity. Bone phantom construction is challenging due to the small size of the structures and their tight spacing; normal trabeculae are 80-200 µm thick and demonstrate 400-700 µm marrow spacing³ that increases with age and in osteoporotic subjects. Prior phantoms have used glass rods or polyethylene strings to represent trabecular bone structures inside a Gd doped saline bath^1,2. In recent years, 3D printing for medical applications has expanded rapidly and provides a new means for phantom realization⁴. We exploited the flexibility and resolution provided by the 3D printing platform to study the impact of several variables including trabecular thickness, spacing, and orientation, along with pulse sequence parameters.

Trabecular rods were organized in a 5x5 grid whose subsections consist of rods with variable thickness and spacing (Fig. 1a). In order to replicate the effect of different trabecular bone structures, four tilted versions of rod ensembles were built into a composite phantom (0° to 45°, Fig. 1b). A 3D model of the structure was designed using FreeCad v0.15 software and manufactured with ABS material on a Fortus 360mc printer (Stratasys, Eden Prairie, MN). The printed model was secured inside a two compartment enclosure (Fig. 1c) which was filled with peanut oil (inside) to represent bone marrow and water (outside, 1.25 g/L NiSO4.6H2O and 4/L NaCl) to represent muscle tissue.

Imaging experiments were performed to investigate the effect of slice thickness -partial volume averaging- and sequence parameters (3T Skyra scanner, Siemens, Erlangen; with 15 channel knee coil, QED, Mayfield, OH). We performed balanced SSFP image acquisitions (constructive interference in steady state: CISS) with two different flip angle phases (0° and 180°) and combined acquired images using maximum intensity projection to reduce the banding artifacts. The following parameters were used: TE=5.23ms, TR=12.2ms, flip angle=50°, bandwidth=130Hz/Px. The inplane resolution (0.24x0.24 mm²) and slab thickness was kept the same in multiple acquisitions (slice thickness ranged between 0.24 mm to 1.5 mm).

Figure 1c shows the 3D printed structure inside a two compartment phantom. Coronal and axial representative images of the phantom are shown in Figure 2a. Axial CISS images are displayed on Figure 2b with different slice thickness and rod tilt angle with respect to the axis normal to the acquisition plane. Slice thickness played an important role on the image quality, which was reduced by blurring of the segments with greater tilt angle (normal to the acquisition plane). This effect is more prominent for acquisitions with slice thicknesses more than 500μm. Separation distance between rods decreases blurring and enabled acquisition with 500μm slice thickness to resolve rod structures with more than 800μm separation for all tilt angles. As seen on Figure 2b, there are missing rods on the manufactured structure which was printed based on the 3D CAD model (e.g. rod thickness=80μm and spacing=800μm). This could have happened either during printing or more probably during the support material removal process, which can be mitigated by optimizing the process for fine structures.We developed and manufactured a resolution phantom with fine structures using 3D printing technology. The effects of the slice thickness on the resolving power of MR scans of trabecular structure were investigated. Our results indicate that the 500μm slice thickness could still resolve rod structures with studied tilt angles in case there is enough separation between rods (which is expected in osteoporotic trabecular bone structure). In the future, such phantoms could potentially be widely mass produced with different types of trabecular microarchitecture (plates, rods, varying levels of connectivity), which would permit not only sequence optimization for high-resolution MRI scanning, but also potential cross-scanner and cross-vendor calibration of high-resolution data for multicenter clinical studies.