01 / Why it matters
Body chemistry,
without a blood sample?
Interstitial fluid surrounds the cells within tissues, including the skin. It provides access to biomarkers such as glucose. Our longer-term ambition was to investigate whether radiofrequency measurements through the skin could support monitoring of the body’s chemistry.
Glucose was the starting motivation. The broader ambition extended to metabolic and stress monitoring, including insulin, adrenaline and cortisol as potential targets. Detecting these molecules with this technique remained a research goal.
Human studies require ethical review and appropriate approvals, and a living finger does not let us freely set the fluid composition. We therefore needed a controllable tissue substitute: a phantom that reproduced relevant electrical properties and let us change what was inside.
Our contribution was to develop that testbed. We planned a multilayer finger, then built and measured a simpler model of the tissue between the thumb and index finger: two skin-like layers around a liquid compartment.
Team project at CROMA · Julien Le Boulch · Maxime Raffort · Andrea Diserò · Mathilde Faucher
02 / The materials
Match the electrical response,
then make it workable.
The planned finger combined materials with different dielectric properties—the way they store and dissipate energy in an electric field. PVDF was selected for the bone analogue, polyepoxide for fat, and water-rich polyvinyl alcohol (PVA) formulations for skin and muscle.
The fabrication idea was to work inward from a silicone mould, reserving space for each new layer. Explore the intended sequence below; the last step shows the complete design.
The 3D concept
Built up, layer by layer.
Planned fabrication sequence
Schematics from the project · not to scale
01 / Planned assembly
Start with the finger geometry.
A silicone mould would define the outer shape. It provides the reference surface for casting the tissue analogues from the outside inward.
Planned tool: silicone mould in a beaker.
02 / Planned assembly
Leave a thin outer layer.
A removable tube would reserve space for the inner tissues. PVA and water would fill the gap against the mould to form the skin analogue.
Critical dependency: the PVA must stabilise before the next layer.
03 / Planned assembly
Keep the liquid compartment separate.
A thin plastic-bag layer would contain the interstitial-fluid analogue. Its role was to keep a changeable fluid compartment within the tissue model.
The membrane itself would also need an electromagnetic control.
04 / Planned assembly
Cast around a smaller former.
Polyepoxide resin would form the fat-equivalent region around an 8 mm tube. Removing the tube would leave room for the remaining inner components.
Constraint: resin handling and curing heat complicated this step.
05 / Planned assembly
Finish with muscle and bone analogues.
A PVDF tube would represent the rigid bone core. A PVA-water formulation would fill the surrounding muscle-equivalent region.
This is the intended assembly. The full 3D phantom was not completed.
- Skin
- PVA + water
- Fluid compartment
- Sealed plastic membrane
- Fat
- Polyepoxide resin
- Muscle / bone
- PVA / rigid PVDF core
03 / Making the skin
From a formulation
to a physical sample.
We first compared PVA concentrations and sample masses. Once a skin formulation was selected, preparation followed three main stages: dissolve, degas and dry. Each stage could change the sample that the antenna would eventually measure.
The selected skin formulation
16%
PVA by mass
The team selected this formulation as a practical balance between the intended dielectric response and handling.
The comparison guided fabrication; it did not independently validate equivalence to human skin.
- 0140 min
Dissolve & mix
Heat PVA with water while stirring and monitoring temperature.
- 0290 min
Remove the bubbles
Degas the mixture to reduce trapped air before forming the sample.
- 031 week*
Form the skin
Leave the sample to dry slowly in a covered dish, then measure it.
*Timings shown in the preparation protocol. One week was the illustrated drying step; actual drying varied from days to weeks and did not reliably produce a stable sample.
04 / The redesign
The drying step
changed the design.
Passive drying took days to weeks and left uncertain water content and uneven thickness. Both affect the electromagnetic response. The 3D route also depended on supervised resin handling that was not available during the project.
We first tested folded solid PVA around a saline-filled bag, then replaced it with two sealed bags of liquid PVA around the saline compartment. This avoided the drying step and kept the starting composition defined.
Tested configuration 01
Folded solid skin
The bottleneck: making the same sample twice.
Tested configuration 02
Sealed liquid layers
The tradeoff: more interfaces to control.
Inside the liquid prototype
Change the fluid.
Keep the surroundings defined.
The middle bag contains the saline solution. The two outer bags provide the skin analogues.
A more controllable composition does not automatically mean a tissue-equivalent material. Liquid and dried PVA still require separate characterisation.
05 / The measurement
Measure the response.
Check what repeats.
A vector network analyser (VNA) measures the magnitude and phase of reflected and transmitted radiofrequency signals. We used it to compare the sample configurations and examine how changing the saline compartment affected the response.
Repeated measurements helped separate stable features from weak signals and changes at the interfaces. Reflection was generally more repeatable; transmission required care near deep signal minima.
- S11 · Reflection
- How much of the incident signal returns to the input. It helps assess how the antenna couples to the sample.
- S21 · Transmission
- How much signal reaches the receiving port. A very weak signal makes differences between measurements harder to interpret.
Why the reference frequency matters
We compared the experimental response over 1–4 GHz with a published hand model designed around the 2.4 GHz band. The resonances did not align. Our sensing target was around 5.5 GHz, so this comparison could not validate tissue equivalence at the intended operating frequency. A meaningful validation needs a reference and measurements in the same band.
06 / The result
Salt concentration did not
give a linear readout.
We varied salt concentration as a controlled change in the fluid compartment. At 5.5 GHz, transmission first decreased, then increased: it did not consistently move in one direction as concentration rose. The straight-line fit was weak, with R² = 0.0468.
The practical consequence was clear: this configuration did not provide a straightforward concentration readout. A changing signal alone was insufficient to infer concentration. These were saline experiments; glucose and hormone detection remained untested.
This configuration did not offer a simple route toward the sensing ambition. A usable concentration measurement would require substantial further development.
07 / What we learned
A testbed built.
A harder sensing path.
The phantom made it possible to test a sensing idea under controlled changes in fluid composition. The measured response suggests this configuration may not be the easiest path toward the long-term objective: even the saline experiment yielded no simple calibration. Nonlinearity alone does not rule out RF sensing, but these results do not establish a usable sensor.
Before investing in a more realistic 3D phantom, the priority is to establish a repeatable, interpretable response in the simpler model. That means controlling hydration, thickness, contact pressure and membrane effects, then comparing antenna configurations against a reference at the same frequency. Glucose-specific experiments would be a separate next stage.
The academic foundations