Most coffee drinks have experienced the dreaded coffee mug stain. (Credit: by Anna Kondratiuk-Swiacka / Shutterstock)
Table of contents
In A Nutshell
- What’s new? UC Berkeley researchers have developed a smartphone-based biosensor that uses the coffee-ring effect and gold nanoparticles to detect tiny amounts of disease proteins in saliva or buffer samples.
- Why it matters: The prototype test showed lab-based detection limits up to 30× better than current hospital ELISA tests and over 100× more sensitive than common lateral flow tests, potentially enabling much earlier intervention for conditions like sepsis, COVID-19, and cancers.
- How it works: A small liquid sample dries on a special membrane, concentrating proteins into a visible ring pattern. A second droplet with gold nanoshells binds to these proteins, forming patterns read by AI from a smartphone photo.
- Caveats: So far, the technology has only been tested in laboratory settings with prepared samples, not real patient blood. It will need extensive clinical trials and regulatory approval before any real-world use.
Biosensor Uses Smartphone Camera To Spot Disease Proteins In Minutes
BERKELEY, Calif. — A breakthrough technology turns smartphone cameras into powerful diagnostic tools, potentially revolutionizing how doctors detect diseases. Berkeley researchers developed a biosensor that’s over 100 times more sensitive than current rapid tests, and it works by mimicking the physics of those annoying coffee-ring stains that mugs often leave behind.
Using just a tiny sample and a smartphone photo, the new biosensor can detect disease proteins at incredibly low concentrations. For sepsis, a condition that kills nearly 20% of patients worldwide, this enhanced sensitivity could mean the difference between life and death.
Current rapid tests often miss the critical window when treatment works best. As the researchers note in their Nature Communications paper, “By starting the antibiotic administration just one hour earlier can increase the chance of survival by 7.6%.”
Why Coffee Rings Inspired Study Authors
The Berkeley team solved a puzzle hiding in plain sight: coffee rings. When spilled coffee dries, it naturally concentrates particles at the edges, creating those stubborn stains on countertops. Scientists realized they could use this same effect to magnify disease markers in samples.
“We thought this could be an interesting method to enhance the sensitivity of a sensor. We tried a couple of different approaches, and eventually, the first version of the coffee-ring sensor emerged,” study co-author Kamyar Behrouzi tells StudyFinds. Our full Q&A with Behrouzi can be found at the end of the article.
Their system works in two steps. First, a sample dries on a specially designed membrane, concentrating disease proteins into a ring pattern. Next, tiny gold particles with attached antibodies flow over these concentrated proteins, creating distinct visual patterns that reveal what diseases are present.
The entire process takes less than 12 minutes. Artificial intelligence then analyzes smartphone photos of these patterns, distinguishing between different diseases with remarkable accuracy.
“We knew the popular machine learning schemes have been used to analyze and distinguish various images, including those for medical applications,” co-author Liwei Lin tells StudyFinds. “As such, we thought a smartphone camera could be used to take images of the experimental images for improved sensing results. The ‘Eureka!’ moment was things did work out after we tried the scheme.”
AI Reads Disease Patterns
Two AI systems work together to interpret test results. One acts like a screener, identifying whether disease markers are present. The other measures exact concentration levels to track disease progression over time.
During laboratory testing, the AI correctly identified positive cases across four different protein markers: SARS-CoV-2 nucleocapsid protein for COVID-19, procalcitonin for sepsis, and cancer markers PSA and CEA. The system maintained its accuracy even when tested with samples spiked into pooled human saliva.
Laboratory Results Show Exceptional Sensitivity
Laboratory testing revealed extraordinary sensitivity levels. For prostate-specific antigen (PSA), the test achieved a detection limit of 3 picograms per milliliter, about 30 times better than current ELISA tests used in hospitals. For COVID-19 protein detection, sensitivity surpassed lateral flow rapid tests by over 100 times.
Sepsis detection proved equally impressive. The test identified procalcitonin biomarkers at concentrations present just hours after infection begins, when current rapid methods typically fail. Since sepsis can progress rapidly to organ failure, this enhanced sensitivity could enable much earlier intervention.
The technology worked reliably across an enormous range, from trace amounts to high concentrations, spanning five orders of magnitude.
Accessible Technology Design
Unlike existing diagnostic methods requiring expensive laboratory equipment and trained technicians, this system needs only a smartphone. Researchers designed the system for accessibility, though specific costs weren’t provided in the study.
The most sensitive component — the gold nanoparticle solution — remains stable for about six months when stored at 4°C. The team developed a complete testing kit using 3D-printed components and a battery-powered heating element.
The technology goes beyond simple yes-or-no answers, providing precise measurements. The platform’s design allows new disease markers to be added by training the AI on additional proteins.
Laboratory testing with human saliva samples showed the system maintained its performance when detecting spiked proteins, demonstrating potential for real-world applications. However, the study was conducted entirely in laboratory settings using prepared samples rather than clinical testing with actual patients.
While the research shows promising laboratory results, the technology would need extensive clinical validation and regulatory approval before reaching patients. The authors acknowledge that different proteins require individual optimization, and the quality of commercial antibodies can affect performance.
StudyFinds Q&A With Co- Authors Liwei Lin and Kamyar Behrouzi
What inspired you to look at coffee-ring physics as a way to detect disease markers?
LL: Coffee-ring is a naturally occurring process to concentrate nanoparticles in our daily life. We thought about using this physics in this work from our earlier efforts in developing the COVID-19 at-home test kits, as it doesn’t need any equipment for possible low-cost applications.
KB: It happened rather suddenly. While I was developing another method for COVID detection (K. Behrouzi, L. Lin, Biosensors and Bioelectronics, 2022), I left a droplet of plasmonic nanoparticles on a substrate. After it evaporated, I noticed a series of ring patterns. Discussing this with Prof. Lin, I realized this phenomenon is known as the coffee-ring effect — a natural way of pre-concentrating particles. We thought this could be an interesting method to enhance the sensitivity of a sensor. We tried a couple of different approaches, and eventually, the first version of the coffee-ring sensor emerged.
Was there a ‘Eureka!’ moment when you realized this could work with a smartphone camera?
KB: I’d say the true Eureka moment was the very first time I saw a specific coffee-ring pattern form in the presence of the COVID-19 N-protein, visible to the naked eye, meaning a simple photo (captured by a smartphone) is enough to detect the level of a particular biomarker in a sample. However, when I discovered that the image of coffee-ring patterns correlated with the concentration of the sample, that was a second exciting moment.
LL: We knew the popular machine learning schemes have been used to analyze and distinguish various images, including those for medical applications. As such, we thought a smartphone camera could be used to take images of the experimental images for improved sensing results. The ‘Eureka!’ moment was things did work out after we tried the scheme.
Why did you choose these four specific proteins for your initial tests?
LL: SARS-CoV-2 Nucleocapsid (N) protein for COVID-19 is an obvious one, as most people have had the test experience during the pandemic. Procalcitonin (PCT) for sepsis is very important for early diagnosis – sepsis is estimated to cause 11 million deaths per year globally and is a leading cause of death in hospitals (1 in 3 deaths in hospitals). Carcinoembryonic antigen (CEA) and Prostate-specific antigen (PSA) for cancer diagnosis are the ones we picked up for cancer diagnosis that are commonly conducted in hospitals.
KB: As we wanted to demonstrate the application of our biosensor, we selected biomarkers for which our sensor’s sensitivity falls within the clinical range of the respective diseases. The N-protein of COVID-19 was chosen to showcase the application for infectious diseases. PSA and CEA are relevant biomarkers to highlight the potential for early cancer diagnosis, and PCT was chosen to show the biosensor’s application for rapid sepsis detection.
How did you ensure the AI could reliably read such tiny variations in the patterns?
KB: We tested our trained model on a separate test dataset to confirm it could reliably process the coffee-ring patterns. We collected a large dataset in a repeatable way and developed our model using transfer learning, a technique that builds on models already trained on millions of images.
LL: AI helps to differentiate the tiny variations in the patterns by going through the training process (more training data will result in better accuracy).
Can you describe what it was like to see the test work on human saliva samples for the first time?
LL: We were very excited to see the testing results on human saliva samples as this is the first milestone toward practical applications.
KB: From the beginning, we always envisioned testing our method on real human saliva samples. After demonstrating that the sensor worked across a wide range of biomarkers, we decided to begin processing human samples. I’ll never forget how much I was praying before seeing the very first results. Yet, something in my heart told me that the expected pattern would appear.
How do you imagine this technology helping people in low-resource or remote areas?
LL: This low-cost, at-home test technology would provide opportunities for people in low-resource or remote areas for early detection of health problems without using expensive or bulky equipment.
KB: The coffee-ring sensor can detect biomarkers with orders of magnitude better performance than lateral flow tests, without requiring complex tools or specialized training. We anticipate people using this as a simple way to get an initial indication of their health status. If they receive concerning results, they could discuss them with their primary care providers and pursue more precise testing. Consider a situation like sepsis, waiting hours for results can significantly endanger a patient’s health. Having access to a rapid diagnostic tool could reveal early warning signs, allowing doctors to begin basic care even before more accurate results arrive from complex tests.
Looking ahead, what’s the timeline you hope for to see this technology in everyday healthcare?
LL: We have filed a U.S. patent based on this technology and are looking for possibilities to start a company concentrating on the development of this technology.
KB: We’ve already patented this technology and are actively seeking funding opportunities to commercialize our coffee-ring sensor. We hope to begin a wide range of clinical tests with the at-home version of our sensor within a year, laying the groundwork for the next steps toward commercialization.
Paper Summary
Methodology
Researchers developed a biosensor using the coffee-ring effect, where samples evaporate on a special nanofibrous membrane to concentrate disease proteins. Gold nanoshells (150 nanometer particles) with attached antibodies create visible patterns. The team tested four proteins: COVID-19 (N-protein), sepsis (procalcitonin), and cancer markers (PSA and CEA). They used smartphone cameras to capture images and trained AI networks to analyze patterns. Tests included buffer solutions and human saliva samples spiked with target proteins.
Results
The biosensor achieved sensitivity as low as 3 picograms per milliliter for PSA—30 times better than ELISA tests and over 100 times more sensitive than lateral flow immunoassays. The AI correctly identified proteins across five orders of magnitude in concentration. Testing in human saliva showed no performance loss compared to buffer solutions. The process completed in under 12 minutes with smartphone-analyzable results.
Limitations
Different proteins required individual optimization of antibody interactions and test conditions. Commercial protein and antibody quality affected performance consistency. The AI system requires extensive training data for each new protein marker. The study was conducted in laboratory settings using prepared samples, not clinical environments, and needs regulatory approval before clinical use.
Funding and Disclosures
Research was supported by the 2020 Seed Fund Award from CITRIS and the Banatao Institute at the University of California. Authors declared no competing interests. Additional support came from UC Berkeley’s nanotechnology facilities and computational resources.
Publication Information
“Plasmonic coffee-ring biosensing for AI-assisted point-of-care diagnostics” was published in Nature Communications, volume 16, article number 4597, in 2025. The paper was received April 12, 2024, and accepted May 7, 2025. DOI: https://doi.org/10.1038/s41467-025-59868-y.
