Label-free microfluidic chip and AI identify blood-borne tumor cells

Finding a few tumor cells hidden among billions of blood cells is one of the biggest challenges in liquid biopsy. Circulating tumor cells (CTCs), which break away from primary tumors and enter the bloodstream, can provide important clues about cancer progression and treatment response. However, their extremely low abundance-often only one to ten cells among a billion blood cells-makes them difficult to isolate and accurately identify.

To address the challenges outlined above, the research team led by Prof. Xiaochun Li from the Institute of Biomedical Precision Testing and Instrumentation, College of Artificial Intelligence, Taiyuan University of Technology, in collaboration with Director Lizhong Zhang from Shanxi Bethune Hospital, published a research article entitled "Label-free Enrichment and Identification of Circulating Tumor Cells Integrating Inertial Microfluidics and Deep Learning" in Biomedical Analysis. Dr. Weizhi Liu and Prof. Xiaochun Li from Taiyuan University of Technology, together with Director Lizhong Zhang from Shanxi Bethune Hospital, are the co-corresponding authors of the paper. Junyi Ouyang, a master's student at Taiyuan University of Technology, is the first author, and Dr. Haiqin Li from Taiyuan University of Technology also contributed to this study. In this study, the researchers developed a label-free platform for circulating tumor cell (CTC) analysis that integrates inertial microfluidic enrichment with YOLOv8-based deep learning for bright-field image recognition, enabling rapid enrichment and intelligent identification of tumor cells within the complex blood background. Notably, the cell separation process does not rely on specific cell-surface markers, while the identification step is performed solely using bright-field images. This integrated strategy provides a new technical approach for label-free CTC analysis.

A chip that separates tumor cells by size

Most current approaches for isolating CTCs rely on molecular markers found on tumor-cell surfaces. However, tumor cells are highly diverse, and different CTC populations may not carry the same markers. To overcome this limitation, researchers have explored label-free methods that use physical characteristics such as cell size and deformability. The team designed a spiral microfluidic chip containing specially engineered contraction-expansion structures. Because CTCs are generally larger than blood cells-with typical diameters of about 12–25 μm, compared with 7–12 μm for white blood cells and 6–8 μm for red blood cells-different cell populations experience different forces as they move through the microchannel. The chip uses these differences to guide tumor cells and blood cells into separate flow paths, allowing tumor cells to be collected without relying on specific surface markers.

Computer simulations and experimental studies confirmed that the specially designed channel could separate particles and cells according to size. In artificial blood samples containing MCF-7 breast cancer cells and white blood cells, tumor cells were enriched at the central outlet, while most blood cells were directed toward the side outlets. The system achieved an MCF-7 cell recovery rate of 89.2 ± 3.1% and a white blood cell removal rate of 86.9 ± 1.4%. Meanwhile, the proportion of tumor cells increased from 9% before separation to 38% after enrichment, demonstrating the ability of the chip to reduce blood-cell background and improve tumor-cell enrichment.

AI helps identify tumor cells without fluorescent staining

Separating tumor cells is only half of the challenge. Researchers still need to determine which collected cells are truly tumor cells. In many existing workflows, this step requires fluorescent staining or antibody-based labeling, adding additional processing steps and limiting some downstream applications.To address this problem, the researchers introduced a YOLOv8 deep-learning model that can recognize tumor cells from brightfield microscope images. During model development, fluorescence images were used to establish the identity of different cell types and support image annotation. After training, however, the model performed recognition using brightfield images alone.

Unlike image-analysis approaches that require individual cells to be manually separated before classification, YOLOv8 can directly locate, classify, and count cells within complete microscope images. The model achieved 96.0% accuracy, with 94.6% precision and recall and 95.4% specificity in distinguishing tumor cells from non-tumor cells.These results show that artificial intelligence can extract useful morphological information from standard microscope images, reducing the need for additional fluorescent labeling during the identification process.

Combining microfluidics and AI for next-generation CTC analysis

By bringing together physical cell sorting and artificial intelligence, the new platform addresses two major bottlenecks in CTC analysis: finding rare tumor cells in a complex blood environment and recognizing them accurately after enrichment.The label-free workflow may simplify sample processing and help maintain cell integrity for downstream applications, including single-cell sequencing and drug susceptibility testing. More broadly, the study demonstrates how combining microfluidic technology with deep learning can create new possibilities for analyzing rare biological cells.

The researchers note that the current work remains a proof-of-concept study. The platform was validated using MCF-7 cell lines and artificial blood samples rather than patient-derived clinical samples, and the enrichment and recognition modules have not yet been fully integrated into a single automated system. Future studies will focus on evaluating additional tumor types and patient samples, while improving system integration and automation.

The challenge of circulating tumor cell analysis is not only finding these rare cells, but also identifying them accurately after separation. By combining microfluidics with artificial intelligence, we hope to provide a simpler and more flexible approach for label-free CTC analysis and future downstream applications."

Xiaochun Li, corresponding author