Label-Free Platform Combines Microfluidics and AI for Circulating Tumor Cell Analysis

Source: Labmedica.com (October 1, 2026). AI-generated summary by biochip.com, published October 2, 2026. Not independently reviewed.

Researchers at Taiyuan University of Technology and Shanxi Bethune Hospital have built a label-free circulating tumor cell analysis platform that combines a spiral microfluidic chip with artificial intelligence image recognition. The system separates likely tumor cells from blood cells based on their physical behavior in fluid, then uses a YOLOv8 deep-learning model to identify cells in ordinary bright-field microscope images. In artificial blood samples seeded with MCF-7 breast cancer cells, the platform recovered 89.2% of tumor cells and identified tumor cells with 96.0% accuracy. The work could help address a central challenge in liquid biopsy: finding extremely rare cancer cells without relying on antibody labels or cell-surface markers.

Separating rare cells by how they move

Circulating tumor cells, or CTCs, are cancer cells that have detached from a tumor and entered the bloodstream. They can offer clues about cancer progression and treatment response, but they are difficult to isolate because there may be only one to 10 CTCs among a billion blood cells.

The team's enrichment device is a spiral microfluidic chip, a small platform that directs fluid through microscale channels. Like leaves and pebbles following different paths in a fast-moving stream, cells of different sizes experience different forces as they travel through the chip's contraction-expansion structures. Because CTCs are generally larger than blood cells, the device guides tumor cells toward a central outlet while directing most blood cells to side outlets.

Label-free enrichment results

The proof-of-concept experiments used artificial blood samples containing MCF-7 breast cancer cells and white blood cells. The chip concentrated tumor cells at its central outlet, increasing their proportion from 9% before separation to 38% afterward.

The platform achieved an MCF-7 recovery rate of 89.2 ± 3.1% and removed 86.9 ± 1.4% of white blood cells. Crucially, the approach does not depend on antibodies that bind particular cell-surface markers. That may avoid missing tumor cells whose surface features differ from the chosen marker and may leave recovered cells available for later testing.

AI identifies cells in bright-field images

Enrichment alone does not prove that a captured cell is a tumor cell. To handle that second step, the researchers trained a YOLOv8 deep-learning model, a type of AI system designed to locate and classify objects in images, to distinguish tumor cells from non-tumor cells.

Fluorescence images helped the team label cells during model development. After training, however, the model analyzed bright-field microscope images alone, avoiding the need for fluorescent labels in the identification workflow. It reached 96.0% accuracy, with 94.6% precision and recall and 95.4% specificity for distinguishing tumor cells from non-tumor cells.

Why This Matters

Liquid biopsy often focuses on tumor-derived material in blood, and intact circulating tumor cells could support more detailed follow-up studies than a simple molecular signal. A label-free workflow could preserve isolated cells for downstream applications named by the team, including single-cell sequencing and drug susceptibility testing.

The study, published in Biomedical Analysis, remains an early demonstration. The researchers tested MCF-7 cell lines in artificial blood rather than patient-derived specimens, and the separation and recognition modules have not yet been fully integrated into one automated system. Future work will examine additional tumor types, clinical samples, automation, and full system integration.

If those steps succeed, pairing physical cell sorting with image-based AI could provide a simpler route to rare-cell analysis from blood. The key next test is whether performance holds when the platform confronts the biological variation and far lower tumor-cell counts found in patient samples.

Topics: Microfluidics · Cancer

Start here: Microfluidics explained