Preterm infants are among the most vulnerable patients in medicine. Monitoring their condition often requires numerous sensors as well as repeated blood sampling to assess critical health parameters. An international research team, including scientists at Helmholtz Munich, has now developed an ultrathin, silk-based sensor patch capable of simultaneously measuring multiple biomarkers through skin fluids. The findings were published in the journal ACS Sensors.
“The newborn is the most demanding patient we have,” said senior author Anne Hilgendorff, a neonatologist and researcher at Helmholtz Munich and Carl von Ossietzky University Oldenburg. “What we’ve built is designed around that reality: no needles, no wires, nothing that pulls or irritates the skin. Just a small patch that listens to the body.”
Why One Number Is Never Enough
Most existing noninvasive monitors for babies, pulse oximeters, skin thermometers, read just one signal at a time. Doctors then have to piece together what those isolated numbers mean in context.
But the human body doesn’t work in isolated numbers. A baby’s sodium level only makes sense alongside their hydration, their temperature, and their metabolic state. A drop in glucose has very different implications if the pH is also shifting. The dangerous events that neonatal teams race to catch, such as sepsis, dehydration, or metabolic crises, almost always show up as patterns across multiple signals, not a single alarm on a single machine.
The new patch is, in effect, a miniature dashboard for a baby’s physiology with twelve different color-changing dyes printed onto a single biocompatible silk-and-paper disc, each tuned to a different biological question.
How It Works
The sensor is built in layers, each only fractions of a millimeter thick: (1) a silk fibroin base, derived from silk moth cocoons, stabilizes delicate biological molecules, including enzymes that normally need refrigeration, so the patch is shelf-stable and rugged. (2) a wax-printed paper layer acts as a tiny plumbing system, drawing in microscopic volumes of fluid (as little as 3 microliters per spot or about one-twentieth of a single raindrop) and routing it to each sensing dot, and (3) a waterproof medical adhesive seals the whole thing against the warm humidity of an incubator and lets the patch flex with a baby’s skin.
When sweat, saliva, or interstitial fluid touches the dyes, they change color: yellow to deep red for glucose, blue to purple for sodium, and so on across the temperature and pH ranges that matter for newborn care.
Color changes are notoriously hard to read reliably. They look different under fluorescent lights, daylight, shadows, or the soft glow of an incubator lamp, and they shift again if a camera is at an angle or partly blocked by a sheet. The team built an AI deep-learning model that automatically corrects for lighting, angle, and movement and converts the patch’s colors into precise measurements.
In testing, the system achieved a mean error of about half a degree Celsius for temperature, less than half a pH unit for acidity, and can detect clinically critical thresholds (including hypoglycemia, hypernatremia, and the salt levels used to diagnose cystic fibrosis) with accuracy above 91% across the board, and above 98% for low blood sugar. A second AI model can find and follow the patch on a moving baby through an incubator’s plastic walls, even with cables, blankets, and clinicians’ hands moving in and out of frame.
A Milestone for Non-Invasive Care
Crucially, the patch is designed around what preterm babies actually do: they lose fluid through their skin at high rates because their skin barrier hasn’t finished developing, turning a vulnerability into a diagnostic opportunity. That same fluid loss provides a continuous, painless sample. In about 15 to 40 minutes, the patch collects enough fluid to read all four biomarkers, comparable to the volume gathered by standard cystic fibrosis sweat testing, but without the mild electrical stimulation that test requires.
In pilot tests on adult saliva and sweat across fasting, exercise, and post-meal conditions, the sensor tracked the same trends as gold-standard laboratory assays and finger-prick blood glucose tests, demonstrating it responds to real physiology and not just controlled solutions in a lab.
“We’re not replacing the lab,” says senior author Dr. Benjamin Schubert, who is leading a DZL-supported research group at Computational Health Center, Helmholtz Munich. “We’re catching the things that happen between lab tests, the slow drift toward a problem that no one sees until it becomes an emergency. That’s where continuous, non-invasive monitoring saves lives.”
Built on a Growing Platform
The neonatal patch is the latest chapter in a longer arc of work coming out of the Silklab and its collaborators, who have spent years building a toolkit for turning silk into a Swiss-army knife for biosensing. Earlier publications from the group introduced paper-based silk patches that track lactate in sweat in real time, useful for monitoring muscle stress and metabolic state. Other work, in collaboration with the Institute for Protein Design at the University of Washington, has shown that computationally designed “de novo” proteins, molecular switches that don’t exist in nature, can be printed into silk to detect viruses, toxins, and cancer markers, with sensitivities comparable to advanced clinical lab assays. And the same printable silk inks have already been shown to work on everyday clothing, gloves, and masks, opening the door to distributed sensing woven directly into a T-shirt.
Stacked together, these advances point toward a future where biomedical sensing isn’t a piece of expensive hospital equipment in a single room, but something quiet and ambient, printed into a baby’s incubator patch today, into a runner’s shirt tomorrow, into a face mask during the next outbreak.
What Comes Next
The team is careful to call the current work a proof-of-principle. The next steps include larger studies in real neonatal units, pairing patch readings with traditional blood samples to confirm how closely skin fluid mirrors what’s happening inside a baby’s bloodstream, and broadening the AI’s training data across different incubators, lighting setups, and the full range of skin tones it will encounter in clinics worldwide. Toxicology and regulatory work will follow.
Longer term, the same platform could be extended to oxygen saturation, carbon dioxide, and other parameters and, given that the sensor itself costs cents to manufacture and needs no power, no wires, and no refrigeration, it is uniquely suited to low-resource settings, where neonatal mortality remains stubbornly high and high-end monitoring is often out of reach.
Original Publication
Alejandra Castelblanco et al., 2026: Artificial Intelligence-Supported Colorimetric Multibiomarker Sensor to Enable Critical Neonatal Monitoring. ACS Sensors. DOI: 10.1021/acssensors.5c04171
