Design considerations

We list the following considerations in designing a phantom, gathered from literature and from experience designing phantoms in our research group.

Phantom measurands

There are many measurands (measurement targets) for MRI applications and systems. They can be loosely categorized into system-related targets that assess how well an MRI system is functioning, and subject-related targets that measure properties of the scan subject. Examples of system-related targets include radiofrequency field uniformity, gradient amplitude, gradient linearity, geometric distortion, spatial resolution, slice thickness, slice position, image contrast, SNR, coil loading, and temperature. Examples of subject-related targets include relaxation (T1, T2, T1/T2, T2*, T1ρ), diffusion, fat fraction, fat suppression, elastography, proton density, magnetization transfer, susceptibility, chemical shift, flow, musculoskeletal imaging, conductivity, permittivity, and multimodal imaging signals (MR-PET, radiomics). There is overlap between system-related and subject-related targets: if the system is not functioning correctly, then the subject-related measurements will be affected.

Measurand materials

Phantom materials should be physiologically relevant for the target application(s). Ensuring that materials are representative of target tissues is vital to accurately evaluate scan performance for subject-related measurement targets. Even for many system-related targets, physiologically representative materials are necessary for assessing system performance for clinical use. Additionally, the properties of the phantom materials should cover the entire measurement range expected for the application. Finally, the materials should have a sufficient measurement resolution so that the results are useful for evaluating performance.

Beyond physiological relevance, phantom materials should be stable, and their properties should not degrade over their lifetime of use. Finally, phantom materials should not be toxic or harmful to the patient or operator. They should be affordable and easy to manufacture, which are both important for open-source designs.

A variety of materials have been used in MRI phantoms, each with its own properties and purpose. Some common materials are given in Table 1.

Table 1 Example materials used for MRI phantoms, categorized by their general applicationsPhysical properties

The physical design of a phantom includes the size, shape, and material. Many phantoms consist of multiple components. The arrangement of these components is also a design consideration. Finally, there are physical design choices that can improve phantom usability and safety.

Size and shape

A necessary requirement for any phantom is that it is small enough to fit into the imaging coil. The phantom should also load the coil similarly to how the human body would, so that the impedance matching circuitry and pre-scan functions can operate and be tested under similar conditions to human scanning. The electrical permittivity and conductivity of the phantom materials will also affect coil loading.

The phantom’s shape should minimize air interface susceptibility artifacts. These artifacts are often minimized by creating a phantom that consists of a liquid-filled container, thereby reducing the surface area that interfaces with air. Other design choices for phantom shape include whether it is symmetric or asymmetric, and whether it is designed to mimic the anatomy of a specific organ or tissue.

Shell material

The material of the phantom shell or casing is distinct from the measurand materials in that the primary task is to encase the measurand materials. An encasing material should not produce artifacts in the MR image. As with the measurand materials, the shell material can be chosen to target MR properties. For example, the magnetic susceptibility of the shell can be selected to mimic in vivo conditions [21, 22]. The outside of the phantom can be made flexible by using a material that is pliable, such as silicone or rubber [5], allowing the phantom to conform to the coil shape and to mimic pliable anatomy. Additive manufacturing can be used to create the phantom shell.

Phantom components

MRI phantoms are generally made up of components, some targeting desired measurands and some required for usability. For tissue mimicking samples, the sample size depends on the available space and desired image resolution. Samples can be arranged in a grid to offer the ability to do geometric assessment. Tissue mimicking samples are often placed in a fillable volume. When designing a fillable phantom, two fill ports, large enough for easy filling and releasing air, should be included.

Positioning aids for orienting the phantom in the scanner include 3D-printed serial numbers or text visible in the MR image, and levels or crosshairs to orient the phantom when it is placed in the scanner. It is desirable to have some MR visible asymmetry for image orientation, as well as features that extend through the sagittal, axial, and coronal imaging planes to help with image registration.

For temperature regulation and monitoring, a thermometer can be placed through a fill port, or an MRI-readable thermometer can be included within the phantom [23]. Often ice water or flow jackets are used to fix the temperature, but these techniques make imaging more complex.

Physical design for usability

To make the phantom easy to handle, it should be as light as possible and easy to move. If the phantom is spherical, a holder can help with scanner positioning. A holder can also help ensure the phantom is always in the same position in the scanner, which will improve measurement repeatability. The phantom should be easy to assemble, and the bolts that hold it together should be easy to tighten and loosen. Phantoms with modular designs may provide greater usage flexibility.

Physical design for safety

Ferromagnetic materials should be avoided in the phantom materials as well as in fasteners used in the phantom. Some materials may have unexpected or unknown ferrous compositions (e.g., epoxies or pigments). Safety data sheets or a description of safe consumption-approved components should be provided with the phantom. A short safety response plan should be provided in case the phantom is damaged and leaks, even if it is composed of safe nontoxic materials. Pressure release or expansion ports may be required if the phantom is designed to operate over a wide range of temperatures to ensure the phantom does not crack or pose an over-pressure safety hazard.

Software and scan protocols

Sequences used in the phantom scan protocol should be the same as in in vivo imaging for subject-related measurands. Longer “gold standard” sequences may be appropriate for system-related measurands. The range of acceptable scan parameters should be identified for each sequence. The positioning of the phantom in the scanner should be pre-defined in the design process.

Ideally, data analysis tools would be provided so that the analysis is standardized across users and sites [8]. Some key aspects of a data analysis toolbox include how to define regions of interest (ROI), how to calculate statistics such as accuracy, variation, and bias, and how to determine confidence in measured values. A toolbox to measure image characteristics between different hardware or software versions (e.g., image scaling calculations) would be useful. A toolbox could also record variations in temperature, field strength, or other quantities that could affect measurements.

Finally, publishing numerical three-dimensional descriptions (e.g., computer-aided design (CAD) files, recommended ROI locations, property maps) of open-source phantoms allow for experimental simulations using the phantom. These numerical descriptions constitute a corresponding digital phantom. Creating a public database or forum where users can share information on protocols and measurements can further aid in widespread adoptability of phantom use.

Open-source phantom designPhantom measurands

The goal for the phantom design was to create an accessible phantom using off-the-shelf materials that exhibit contrast in T1-weighted and T2-weighted images, as T1-weighted and T2-weighted scans are very common and can be found on most MRI systems. A second aim for the phantom design is the ability to target system-related measurands, in this case geometric distortion along one axis, SNR, and CNR.

Measurand materials

For open-source phantoms, it is important that the materials are accessible and affordable, in addition to the requirement that they are sensitive to the measurands. This phantom design uses deionized water, olive oil (Whole Foods, Austin, TX, USA), and Chelated Zinc plus Copper mineral supplements (Whole Foods, Austin, TX, USA), which were bought at a local store. Oil is a common material used as an approximate fat mimic, whereas copper is a T1 and T2 modifier. Copper supplements were dissolved in deionized water in 15 ml falcon tubes (Fisher Scientific, Hampton, NH, USA), at varying concentrations. These solutions are straightforward to manufacture, as opposed to more process-involved materials (e.g., materials with gelling agents or PVP). Three tubes were filled with one, two, or three copper supplement capsules, respectively, resulting in sample concentrations: 15 mg Zn and 2 mg Cu; 30 mg Zn and 4 mg Cu; and 45 mg Zn and 6 mg Cu.

Physical properties

The phantom was designed to be a fillable cylinder with an 18 cm outer diameter and 14.8 cm inner diameter. The 18 cm diameter of the cylinder was chosen to be compatible with the OSI2 ONE [24] open-source MRI system, which has a 20-cm-inner diameter RF coil. Two tops to the cylinder were designed: a flat top resulting in a phantom length of 16 cm, and a domed top resulting in a phantom length of 21.4 cm.

The phantom can hold up to 24 samples of target material (in this case, water, oil, or dissolved supplements). Two plates were designed to hold samples, each of which can slot into the phantom volume and has holes that sample tubes press fit into. The first plate is a rectangular plate oriented along the length of the cylinder, and it can hold 15 of the 15 ml Falcon tubes and 9 smaller 5 ml cryogenic tubes (Wheaton Science Products, Millville, New Jersey, USA). The second plate is a circular plate oriented coaxially with the cylinder, and it can hold 18 of the 15 ml Falcon tubes and 6 of the 5 ml cryogenic tubes.

Phantom fabrication

The phantom (Fig. 1) was designed using the CAD software Fusion 360 (Autodesk, San Francisco, CA, USA). STL and STEP files of the design have been made available. The outer shell of the phantom was printed on an Original Prusa 3D printer (Prusa Research, Prague, Czech Republic) using gray polylactic acid (PLA). A fused deposition 3D printer was used since they are common and inexpensive. Two test plates were printed on the Prusa printer using black PLA material. Duplicates of the test plates were printed on a Form 3 printer (Formlabs, Somerville, MA, USA) in a stereolithography (SLA) material. After printing and before filling, the fill port and screw holes were drilled out using appropriate drill bits for each hole and tapped with a threading tap.

Fig. 1

CAD model of the open-source phantom where dimensions in millimeters are indicated by double-sided arrows. a The body of the phantom, which can be filled with a liquid. b The top of the phantom, used as a liquid-tight seal. c A domed version of the phantom top. d A plate that can be inserted into the phantom to hold sample tubes. This plate is referred to as the Cor/Sag plate because it is often oriented in the MR scanner such that coronal or sagittal images will show cross sections of tubes held in the plate. e A plate that can be inserted into the phantom to hold sample tubes. This plate is referred to as the Axial plate because it is often oriented in the MR scanner such that axial images will show cross sections of tubes held in the plate

To seal the phantom for liquid-tight performance, the inner lining of the outer shell was coated with two coats of white Plasti Dip (Plasti Dip International, Blaine, MN, USA). Alternatively, the shell can be printed using a photocured resin (SLA or digital light processing (DLP)), that is watertight, without the need for extra sealing. A 4-mm-wide, 155-mm-inner diameter oil-resistant Buna-N o-ring (part number 1302N303, McMaster-Carr, Elmhurst, IL, USA) was placed between the phantom body and top. This o-ring sizing affected the dimension of the phantom’s inner chamber, resulting in a phantom inner diameter of 14.8 cm. The top was bolted into place using ten 14-mm-long M4 × 0.70 mm thread nylon plastic socket head screws (part number 93640A146, McMaster-Carr, Elmhurst, IL, USA). Although nylon is typically avoided inside phantoms because it absorbs water and deforms, it can be used outside of the water volume. Finally, to plug the phantom after filling, a 1/8 NPT pipe plug (part number 45505K195, McMaster-Carr, Elmhurst, IL, USA) was used.

The phantom was assembled by press fitting the sample tubes into the selected tube-holder plate, placing the plate and tubes into the phantom, and filling the phantom with deionized water to the top of the cylinder. The top was bolted on, and the phantom was filled with more water through the fill port until it overflowed. Finally, the fill port was plugged.

Data acquisition

Axial and coronal T1-weighted and T2-weighted images of the phantom were acquired using a 64 mT Hyperfine Swoop (version 1.8, Hyperfine, Guilford, CT, USA). T1-weighted images were acquired using a multi-slice inversion-recovery spin-echo sequence with a repetition time of 1.5 s, inversion time of 0.3 s, echo time of 0.006 s, 24 signal averages, in-plane resolution of 1.6 mm × 1.6 mm, slice thickness of 5 mm, and fields of view of 22 cm (anterior/posterior) or 18 cm (superior/inferior and left/right). T2-weighted images were acquired using a three-dimensional fast spin-echo sequence with a repetition time of 2 s and an echo time 0.231 s (coronal imaging) and 0.203 s (axial imaging), and 80 signal averages. The in-plane resolution was 1.5 mm × 1.5 mm, and the slice thickness was 5 mm, with fields of view of 22 cm (anterior/posterior) or 18 cm (superior/inferior and left/right). The acquisition times for the T1-weighted images were 5:31 (coronal imaging) and 5:37 (axial imaging). For the T2-weighted images, the acquisition times were 5:25 (coronal imaging) and 5:55 (axial imaging).

To evaluate geometric distortion along one axis, the distance between sample vial centers was calculated for a central slice of the scans: T1-weighted coronal scan using the flat top; T2-weighted coronal scan using the flat top; T1-weighted coronal scan using the domed top; T2-weighted coronal scan using the domed top. The center of each vial was found by identifying each vial’s region of interest (ROI) through an automated process of first thresholding each image so that the edges of the vials were visible, and then finding representative circles for each vial edge. Once all ROIs were determined for an image, the inter-vial distance was calculated between circle centers for consecutive vials. Using all inter-vial distances, the average inter-vial distance was calculated for each image. To interpret the inter-vial distance metric, note that this phantom did not include external or internal level indicators, so it could have been placed at an angle to the imaging volume. Thus, the average inter-vial distance cannot be directly compared to the absolute inter-vial distance of the phantom. Instead, comparisons of the average inter-vial distances between scans with the phantom placement unchanged can be made to determine whether there is relative distortion between scan types. Here, average distance comparisons can be made between the T1-weighted and T2-weighted scans for either the flat top or domed top images.

To evaluate the SNR and CNR, two noise regions of 10 × 10 pixels near the edge of the images were selected, as well as one signal region of 5 × 5 pixels. For T1-weighted images the signal region was placed within the vial of highest signal intensity (oil). For T2-weighted images the signal region was placed in a centrally located portion of the background water fill, as the water signal had the highest intensity in the T2-weighted images. SNR was calculated as the ratio of the mean signal value to the standard deviation of the noise pixel values. CNR was calculated as the ratio of the difference of the mean signal value and the mean noise value to the standard deviation of the noise pixel values.