English | EN 
Before any new imaging method reaches patients, it’s stress‑tested on “phantoms”, lab‑built models that mimic human tissues’ electrical behaviour. For microwave breast imaging, phantoms are indispensable: they enable safe, repeatable experiments, fair algorithm comparisons, and measured progress without exposing patients.
We have a research project in this domain. We don’t fabricate phantoms; we rely on high‑quality public datasets such as UMBMID built and measured by expert teams. Those datasets are the training ground for our AI modules to detect, localize, and estimate lesion size from microwave measurements.
What makes a good breast phantom?
A good phantom should “behave” like a breast when exposed to microwaves. It must emulate: (i) fatty tissue (adipose) which is more like oil, it contains less water, so it interacts less strongly with microwaves; (ii) fibroglandular tissue which has higher water content and interacts more strongly with microwaves. To be useful, a phantom should present realistic dielectric contrast across microwave frequencies (roughly 0.5–10 GHz):
- Adipose vs fibroglandular contrast: adipose (fat) has low permittivity and conductivity. Fibroglandular tissue is higher in both due to greater water content. Malignant tissue tends to be higher still, creating detectable contrast.
- Clinically meaningful targets: inclusions commonly span 5–30 mm diameter (size of a pea to a grape) to probe detectability around screening‑relevant sizes and are usually made more water‑rich than their surroundings to create a contrast that algorithms can pick up.
- Frequency dependence and stability: properties should follow known dispersion with frequency and remain stable over days to weeks.
Materials you’ll often find inside
Most breast phantoms use well‑characterized, “kitchen‑adjacent” ingredients tuned to match tissue targets:
- Water, oils, and gels: water represents water‑rich tissue; oils lower effective permittivity for fatty regions. Gelatine/agar/gelling agents set the structure and reduce settling.
- Surfactants: agents like Triton X‑100 act like soap, helping the emulsification of oil‑in‑water mixtures for uniformity over time.
- Salts and safe fillers: tiny amounts of salt and powders are added to fine‑tune conductivity and dispersion across GHz bands.
- “Skin” and housings: thin silicone (elastomers) can mimic skin; acrylic or polycarbonate cups maintain geometry with minimal microwave field distortion.
Why recipes matter
These choices aren’t arbitrary. Specific Absorption Rate (SAR) and exposure standards provide validated tissue‑simulating liquid targets and formulations across 300 MHz–6 GHz (IEEE/IEC 62209‑1528:2020), which phantom builders adapt for breast‑relevant targets. Keysight and other metrology notes outline best practices for measuring dielectric properties and verifying mixtures.
Why materials matter for AI, too
- Fair training and testing: If a phantom’s “fatty” region slowly separates from its “watery” region, the measurements drift and any model trained on day‑1 data might struggle on day‑7. Stable materials reduce this risk.
- Transfer to the real world: Public datasets document what’s inside the phantom and how it was measured. That transparency helps researchers understand what parts of a model’s performance may or may not carry over to real breasts.
- Better benchmarks: Shared phantoms and standardized recipes mean different research groups can compare results apple‑to‑apples. This is a big step toward trustworthy AI in medical devices.
A balanced view: the strengths and limits of phantoms
| What they do well | What they can’t fully capture |
| Control: Precise lesion size, contrast, and placement | Human variability in shape, motion, temperature, density and heterogeneity |
| Safety and scale: Large datasets without patient risk | Complex biology (e.g., vascularity, microstructures, etc.) |
| Standardization: Enables rigorous benchmarking | Workflow impact (recall rates, triage utility) that only clinical studies reveal |
That’s why a staged path makes sense: start with robust phantom datasets, then expand to carefully designed clinical studies. Our work follows that arc.
We’ll keep sharing what we learn as our research progresses, especially on making AI models robust across phantom generations and measurement setups.
Sources and further reading
Foundational tissue properties
- Gabriel, S.; Lau, R. W.; Gabriel, C. The dielectric properties of biological tissues: I–III. Phys. Med. Biol., 1996. https://doi.org/10.1088/0031-9155/41/11/001 | https://doi.org/10.1088/0031-9155/41/11/002 | https://doi.org/10.1088/0031-9155/41/11/003
- Lazebnik, M. et al. A large-scale study of the ultrawideband microwave dielectric properties of normal, benign and malignant breast tissues obtained from cancer surgeries. Phys. Med. Biol., 2007. https://doi.org/10.1088/0031-9155/52/20/002
- IT’IS Foundation Tissue Properties Database (frequency‑dependent tissue permittivity/conductivity and more). https://itis.swiss/virtual-population/tissue-properties/database/
Phantoms, standards, and measurement
- 62209-1528-2020 – IEC/IEEE International Standard – Measurement procedure for the assessment of specific absorption rate of human exposure to radio frequency fields from hand-held and body-mounted wireless communication devices – Part 1528: Human models, instrumentation, and procedures (Frequency range of 4 MHz to 10 GHz). https://doi.org/10.1109/IEEESTD.2020.9231298
- Hartsgrove, G.; Kraszewski, A.; Surowiec, A. Simulated biological materials for electromagnetic radiation absorption studies. https://doi.org/10.1002/bem.2250080105
- Keysight Technologies. Basics of Measuring the Dielectric Properties of Materials (App Note 5989‑2589EN). https://www.keysight.com/us/en/assets/7018-01284/application-notes/5989-2589.pdf