A team at UCLA figured out how to cram two completely different light-detection methods into a single piece of paper, point a neural network at it, and get results that rival the quarter-million-dollar machines sitting in hospital basements. The test costs under four bucks, runs in 23 minutes, and needs a single drop of blood serum smaller than a teaspoon.

The Problem With Testing Hearts in a Hurry

When someone stumbles into an ER clutching their chest, doctors need to know fast whether it's a heart attack, heart failure, or a really aggressive burrito. The gold standard? Blood tests for cardiac biomarkers - proteins that leak out of damaged heart muscle like confessions from a bad poker player. The big three are cardiac troponin I (cTnI), creatine kinase-MB (CK-MB), and NT-proBNP.

Here's the catch: the machines that measure these accurately are the size of washing machines, cost a fortune, and take about an hour to spit out results. Existing point-of-care tests? They can usually only measure one biomarker at a time, with the sensitivity of someone reading an eye chart without glasses. Miss a low-level troponin signal and you might send a heart attack patient home with antacids.

Two Detection Modes, One Strip of Paper

The researchers built something called a dual-mode xVFA - a vertical flow assay that reads biomarkers two different ways on the same cartridge. Think of it like a camera that shoots in both daylight and infrared simultaneously.

Colorimetric mode handles the abundant biomarkers. Gold nanoparticles bind to the target proteins and produce visible color changes - like a pregnancy test, but for your heart. This works great for CK-MB and NT-proBNP, which show up in relatively high concentrations.

Chemiluminescent mode tackles the sneaky ones. For cTnI, which matters at absurdly low concentrations (we're talking picograms per milliliter - that's like detecting a single grain of sand in an Olympic swimming pool), the system triggers a chemical reaction that emits light. A portable reader captures this glow in the dark, literally.

Together, the two modes span about six orders of magnitude in dynamic range. That's the difference between measuring a whisper and a jet engine with the same microphone.

Where the Neural Network Earns Its Keep

Raw images from paper-based tests are messy. Uneven flow, variable lighting, batch-to-batch differences in the paper itself - it's the kind of noisy data that makes traditional curve-fitting cry in the corner. So the team trained neural networks to interpret the test images, essentially teaching the AI to read the coffee stains.

The models were blindly tested on 92 real patient serum samples, and the correlation with FDA-cleared clinical analyzers hit a Pearson's r above 0.96 for all three biomarkers. That's not "good enough for a screening test" territory - that's "maybe we don't need the washing machine" territory (Han et al., 2026).

This builds on years of work from Aydogan Ozcan's lab at UCLA, which has been steadily upgrading paper-based assays: first fluorescence detection (Goncharov et al., 2023), then nanoparticle-amplified colorimetric sensing that achieved 0.2 pg/mL troponin detection (Han et al., 2024), then standalone chemiluminescence (Han et al., 2025). The new paper is the mashup album - combining the greatest hits into one multiplexed platform.

Why This Matters Beyond the Lab Bench

Cardiovascular disease kills more people globally than anything else - about 17.9 million per year, which is roughly the population of the Netherlands just... gone. Heart attacks and heart failure often travel together like the worst buddy comedy imaginable, with patients bouncing between hospitalizations.

The commercial point-of-care landscape is heating up. Abbott's i-STAT hs-TnI recently got FDA clearance, and Siemens has its Atellica VTLi platform. But these are still single-analyte systems, and they don't come cheap. A $3.86 paper cartridge paired with a ~$170 portable reader could bring multiplexed cardiac diagnostics to rural clinics, ambulances, and low-resource settings where that washing-machine analyzer is never arriving (Wang et al., 2025).

The approach also isn't locked to heart markers. The dual-mode detection architecture could theoretically be retooled for any combination of high- and low-abundance biomarkers - cancer panels, infectious disease screens, you name it.

The Fine Print

Ninety-two patient samples is a solid proof-of-concept, but it's not the thousands you'd need for regulatory approval. The chemiluminescent mode requires a dark enclosure during imaging, which adds a design constraint to the portable reader. And while the neural network performed well in blind testing, deploying AI-driven diagnostics in clinical settings comes with its own regulatory labyrinth that makes the DMV look efficient.

Still, the core achievement is real: two optical detection modes on a single disposable strip, read by a pocket-sized device, interpreted by a neural network, delivering lab-quality measurements of three cardiac biomarkers in under half an hour. For the price of a fancy coffee.

Not bad for a piece of paper.

References

1. Han, G.-R., Eryilmaz, M., Goncharov, A., et al. (2026). Deep learning-enhanced dual-mode multiplexed optical sensor for point-of-care diagnostics of cardiovascular diseases. Light: Science & Applications, 15, 190. DOI: 10.1038/s41377-026-02275-9 | PMID: 41951574

2. Goncharov, A., Joung, H.-A., Ghosh, R., et al. (2023). Deep Learning-Enabled Multiplexed Point-of-Care Sensor using a Paper-Based Fluorescence Vertical Flow Assay. Small, 19(51), e2300617. DOI: 10.1002/smll.202300617 | PMID: 37104829

3. Han, G.-R., Goncharov, A., Eryilmaz, M., et al. (2024). Deep Learning-Enhanced Paper-Based Vertical Flow Assay for High-Sensitivity Troponin Detection Using Nanoparticle Amplification. ACS Nano, 18(41), 28424-28436. DOI: 10.1021/acsnano.4c05153 | PMCID: PMC11483942

4. Han, G.-R., Goncharov, A., Eryilmaz, M., et al. (2025). Deep Learning-Enhanced Chemiluminescence Vertical Flow Assay for High-Sensitivity Cardiac Troponin I Testing. Small. DOI: 10.1002/smll.202411585 | arXiv: 2412.08945

5. Wang, M., et al. (2025). Machine learning-assisted point-of-care diagnostics for cardiovascular healthcare. Bioengineering & Translational Medicine. DOI: 10.1002/btm2.70002 | PMCID: PMC12284442

6. Du, J., Cao, C., Xue, Z., et al. (2025). AI-Enhanced Lateral Flow Assay Enables 3-Minute Quantitative Detection with Laboratory-Grade Accuracy. Analytical Chemistry, 97(43), 24196-24208. DOI: 10.1021/acs.analchem.5c05108 | PMID: 41124618

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.