HomeAIAI Science & HealthBerkeley GPCPI Reads Label-Free Optical Cell Fingerprints

Berkeley GPCPI Reads Label-Free Optical Cell Fingerprints

UC Berkeley engineers have used a new optical method, General Polarization Common-Path Interferometry, or GPCPI, to pick out distinct optical fingerprints in normal and cancerous skin cell lines without chemical labels or fluorescent dyes. In a 20 November college release, Berkeley Engineering says the findings were reported in the journal Optica and describes the work as a proof-of-concept demonstration, not a test ready for the clinic. Lead author Kamyar Behrouzi called it a “promising proof-of-concept result.”

The paper, “AI-Assisted Hyperspectral Interferometry and Single-Cell Dispersion Imaging,” is listed under doi 10.1364/optica.615458. A preprint of the same title was posted on arXiv on 2 January 2026. The Optica full text could not be opened for this report, so the account below follows the Berkeley release and the public abstract.

How GPCPI was tested on cell lines

GPCPI, the release says, divides light into two closely aligned beams: one as a reference, the other to probe the sample. When the beams are recombined, the interference patterns are used to read subtle changes in optical properties. Because the paths are nearly identical, they pick up similar environmental disturbances, which the team argues makes the measurement more stable than some earlier interferometers, and eases requirements on the polarisation of the material being studied.

Artificial intelligence is used in two ways. It helps to identify and correct irregularities in the measured phase spectra, and a deep-learning model analyses interference patterns to detect small optical changes without reconstructing the full phase spectrum for every measurement. Together, the researchers say, the architecture and the AI-assisted analysis let them measure phase variations “approximately 10 times smaller than a previously reported benchmark.”

The team first showed the platform on nanoscale gold structures, detecting changes in the optical properties of surrounding liquids, then used it for hyperspectral imaging of biological cells, measuring how their optical responses change across wavelengths. Those experiments, Berkeley says, revealed distinct optical fingerprints in normal and cancerous skin cell lines. Behrouzi, a postdoctoral researcher, said the result “suggests that our technique could help distinguish different cell types without requiring chemical labels or fluorescent dyes, which can alter biology or damage living samples.” Co-author Tanveer Ahmed Siddique, a PhD student in electrical engineering and computer sciences, said that “with further development and validation, such techniques could contribute to faster, less invasive and more informative biological analysis.”

The principal investigator is Boubacar Kanté, a professor of electrical engineering and computer sciences at UC Berkeley and a faculty scientist at Berkeley Lab. The release also names Megan Teng and Walid Redjem among the co-authors. Kanté described a “long-term vision” of making “highly sensitive, AI-assisted optical measurements accessible to a wider range of scientific, biomedical and industrial applications,” and the team says it now wants a more compact platform. Those are aims, not results already in hand.

Label-free imaging of cells sits close to other recent TSN science coverage in which a laboratory method is easily mistaken for a clinical tool, including Mass General Brigham’s THREADS pathology foundation model, a Singapore paper on a model that explains its reading of medical scans, and a UCL study of AI heart MRI readings.

A proof of concept on cell lines is not a diagnostic

The useful sentence in the Berkeley release is Behrouzi’s: this is a promising proof of concept. The experiments were done on skin cell lines, not in patients, and they show that two cultured lines can look different under a label-free optical measurement, not that a clinic can diagnose disease with GPCPI. Siddique’s remark about “further development and validation” is the condition, not a residual courtesy. Until that validation exists, the work should not be written up as a test, a screen or a diagnostic device.

The same caution applies to the other figures. A phase variation “approximately 10 times smaller than a previously reported benchmark” is a laboratory sensitivity figure, as given by the team, not a head-to-head in a hospital. The gold-structure experiment is a sensing demonstration. Kanté’s long-term vision, and the list of possible later uses in healthcare, drug discovery or quantum sensing, are directions of travel. They are not findings from this study. GPCPI, as published and described this week, is an optical-plus-AI method that, in a proof-of-concept test, read label-free fingerprints from normal and cancerous skin cell lines.

Sources

Related stories

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

  1. ECMWF Starts Forecast-in-a-Box Weather Pilot in Malawi
    AI Science & Health

More in this category

Latest stories

  1. ECMWF Starts Forecast-in-a-Box Weather Pilot in Malawi
    AI Science & Health

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.