Artificial intelligence-based classification of peripheral blood nucleated cells using label-free imaging flow...

Area of Science:

• Biomedical Engineering
• Cell Biology
• Medical Diagnostics

Background:

• Identifying circulating rare cells, like circulating tumor cells (CTCs), in peripheral blood nucleated cells (PBNCs) is difficult due to their low abundance amidst white blood cells (WBCs).
• Existing methods often require cell labeling, which can be complex and potentially affect cell viability.

Purpose of the Study:

• To develop and validate a label-free image cytometry method for identifying and quantifying circulating rare cells.
• To assess the potential of this method as a diagnostic and monitoring tool for cancer patients.

Main Methods:

• Utilized quantitative phase microscopy and computer vision for label-free cell classification based on subcellular structures.
• Applied image recognition algorithms to cells in a flow cytometry microfluidic channel to differentiate WBCs from cancer cell lines.
• Analyzed blood samples from healthy volunteers and advanced cancer patients to determine non-WBC fraction rates (NWBC-FRs).

Main Results:

• The image recognition system achieved high accuracy in differentiating WBCs from cancer cell lines (AUC = 0.957).
• Significantly higher NWBC-FRs were observed in advanced cancer patients compared to healthy volunteers.
• Monitoring NWBC-FRs over therapeutic courses indicated potential for tracking clinical status during treatment.

Conclusions:

• Label-free image cytometry offers a promising approach for the morphological diagnosis of circulating rare cells.
• The non-WBC fraction rate (NWBC-FR) can serve as a valuable biomarker for cancer detection and therapy response monitoring.