Automatic classification of circulating blood cell clusters based on multi-channel flow cytometry imaging.

Area of Science:

• Computational Biology
• Biomedical Imaging
• Hematology

Background:

• Circulating blood cell clusters (CCCs) are key biomarkers for diseases like thrombosis and inflammation.
• Flow cytometry with fluorescence staining is standard for analyzing CCCs, but automated analysis of clusters is lacking.
• Single-cell analysis tools do not address the complex shapes and heterogeneous cell types within CCCs.

Purpose of the Study:

• To develop a computational framework for automated analysis of circulating blood cell cluster images.
• To accurately identify cell clusters and their constituent cell types within flow cytometry data.
• To overcome challenges posed by irregular cluster shapes and multi-channel fluorescence staining.

Main Methods:

• A two-step computational framework was developed for CCC image analysis.
• The You Only Look Once (YOLOv11) model was fine-tuned for classifying images into cluster and non-cluster categories.
• Cell types within clusters were identified by overlaying cluster contours with multi-channel fluorescence staining data.

Main Results:

• The YOLOv11 model outperformed traditional CNNs like Vision Transformers (ViT) in image classification.
• The framework achieved over 95% accuracy in both cluster classification and cell phenotype identification.
• The method effectively minimized artifacts from cell debris and staining.

Conclusions:

• An automated framework for analyzing circulating blood cell clusters from flow cytometry data was successfully created.
• The framework leverages both bright-field and fluorescence imaging for comprehensive analysis.
• This approach shows potential for broader applications in analyzing immune and tumor cell clusters for various disease research areas.