Laser-made SERS chip sorts cells with 96.2% accuracy

Researchers in China and Japan developed a femtosecond laser-fabricated microfluidic chip that uses surface-enhanced Raman scattering and deep learning to identify and sort single cells without labels. The system reached 96.2% cancer-cell identification accuracy and could improve non-invasive cell sorting for diagnostics and precision medicine.

Why it matters: - The new chip targets a major bottleneck in precision biomedicine: isolating rare or mixed cells without fluorescent or magnetic labels. - Better label-free sorting could improve cancer research, drug screening, gene-expression analysis, and future clinical diagnostics. - The approach also matters because cell-sorting accuracy directly affects downstream single-cell analysis and cell-based therapies.

What happened: - Researchers developed a microfluidic surface-enhanced Raman scattering, or SERS, chip for deep-learning-assisted, label-free single-cell identification and sorting. - The work was made available online July 29, 2026, in the Early View section of Opto-Electronic Advances. - The team was led by Professor Koji Sugioka and Dr. Shi Bai from RIKEN's Advanced Laser Processing Research Team. - The chip was built using femtosecond laser processing. - In proof-of-concept tests, the device sorted cancer and non-cancer cells in a Y-shaped microchannel.

The details: - The chip uses plasmonic ring-shaped nanostructure arrays formed by laser near-field reduction of gold ions. - The fabrication method introduces cetyltrimethylammonium bromide into the precursor solution. - The gold nanoparticles are about 28 nm in size. - The number of gold nanoparticles and their gaps in each ring depend on reduction time, which affects SERS performance. - Raman mapping showed ultrahigh spatially uniform Raman enhancement, with a relative standard deviation of 2%. - The chip achieved an experimentally confirmed spatial resolution of about 185 nm. - That spatial resolution improved cell-identification accuracy from 75% to 96.2%. - The researchers said the high spatial resolution was essential for deep-learning label-free cell identification. - The preliminary sorter separated cells without biomarker or Raman-tag labeling. - The original paper is titled "Deep-learning-assisted label-free single-cell identification and sorting using femtosecond laser fabricated microfluidic surface-enhanced Raman scattering (SERS) chips with plasmonic ring-shaped nanostructure arrays." - The paper was published in Opto-Electronic Advances under DOI 10.29026/oea.2026.260071. - Opto-Electronic Advances launched in March 2018 and is an open-access, peer-reviewed SCI journal indexed in SCI, EI, and Scopus. - The journal's website is More information.

Between the lines: - The result points to a key tradeoff in label-free sorting: deep learning can boost performance, but the sensing hardware still needs enough sensitivity and spatial resolution to produce reliable inputs. - The reported jump from 75% to 96.2% suggests the chip design itself, not just the algorithm, was a major driver of performance. - The findings also underscore that SERS-based systems may be most useful when the substrate, cell type, and detection workflow are carefully matched.

What's next: - The researchers said the platform could inform future non-invasive, efficient cell-sorting techniques. - The next step is likely broader validation across more cell types and more complex biological samples. - If that testing holds up, the chip could move closer to clinical diagnostics and precision medicine use cases.

The bottom line: - A laser-fabricated SERS chip paired with deep learning delivered high-precision, label-free cell sorting and showed that spatial resolution is a key factor in accurate single-cell identification.