AI Chip Helps Find Rare Cancer Cells in Blood in New Liquid Biopsy Study

Source: Drug Today, by Drug Today Bureau (September 30, 2026). AI-generated summary by biochip.com, published October 1, 2026. Not independently reviewed.

Researchers at Taiyuan University of Technology and Shanxi Bethune Hospital have paired a spiral microfluidic chip with artificial intelligence to find rare cancer cells in blood samples. The laboratory system enriches circulating tumor cells, or CTCs, by their physical properties and then identifies them in standard microscope images without fluorescent staining at the final recognition step. In tests using breast cancer cells added to white blood cells, the chip recovered 89.2% of tumor cells and the AI model reached 96.0% accuracy. The work could simplify a difficult step in liquid biopsy research, although it has not yet been validated using blood from cancer patients.

A Physical Filter for Rare Cells

CTCs are cells that detach from a tumor and enter the bloodstream. They may carry useful information about cancer progression, but they are extremely difficult to locate: a blood sample can contain only one to 10 CTCs among roughly a billion blood cells.

The team, led by Prof. Xiaochun Li, built a spiral microfluidic chip to narrow that search. Microfluidics means guiding very small volumes of fluid through tiny channels. Like a river bend that sorts material by how it moves through a current, the chip uses inertial forces in its curved channel to separate cells according to physical features such as size.

That approach matters because it does not initially rely on molecular labels attached to cell surfaces. CTCs are generally larger than most blood cells, giving the device a physical way to enrich the tumor-cell fraction before image analysis begins.

What the Chip Recovered

The researchers tested the chip with MCF-7 breast cancer cells mixed with white blood cells. It recovered 89.2% plus or minus 3.1% of the tumor cells and removed 86.9% plus or minus 1.4% of white blood cells.

After processing, tumor cells represented 38% of the collected cells, compared with 9% before separation. That does not mean the device finds every cancer cell in blood, but it substantially increases the share of cells that may deserve closer inspection.

Teaching AI to Recognize Tumor Cells

Cell enrichment alone cannot establish which individual cells are cancerous. The team therefore trained YOLOv8, a deep-learning model built to locate and classify objects in images, using ordinary bright-field microscope images of cells.

Bright-field microscopy is the familiar imaging approach in which cells are viewed using transmitted light, rather than made visible with fluorescent dyes. The trained model distinguished tumor cells from non-tumor cells with 96.0% accuracy, 94.6% precision and recall, and 95.4% specificity.

Using label-free imaging at the recognition stage could make CTC workflows more flexible. It may also leave cells available for later work, such as studying individual tumor cells or testing drug responses.

Why This Matters

Liquid biopsy aims to extract clinically useful cancer information from a blood sample, but rare-cell detection remains a major obstacle. Combining a size-based chip with automated image recognition addresses two parts of that problem: concentrating a sparse target population and distinguishing likely tumor cells from the remaining cells.

The results remain early-stage. The experiments used a cancer cell line in artificial blood samples, not blood drawn from people with cancer, and the enrichment and AI components have only been partially integrated into a single automated system.

The next test is whether the platform can handle the biological variety of real patient samples and detect CTCs across different cancer types. Those studies will determine whether this chip-and-AI workflow can progress from a controlled laboratory demonstration toward a reliable clinical liquid biopsy tool.

Topics: Microfluidics · Cancer

Start here: Microfluidics explained