Multiparameter Mechanical Phenotyping for Accurate Cell Identification Using High-Throughput Microfluidic...

Area of Science:

• Biophysics
• Cell Biology
• Computational Biology

Background:

• Single-cell mechanical phenotyping is crucial for understanding cellular behavior.
• Existing methods often rely on single parameters from static images, limiting comprehensive analysis.
• Characterizing cell mechanics provides insights into cell type, state, and function.

Purpose of the Study:

• To develop a quasi-real-time, multiparameter approach for cell mechanical property analysis.
• To enhance cell identification accuracy using a combination of mechanical parameters and machine learning.
• To validate the method's effectiveness for distinguishing cell lines and clinical samples.

Main Methods:

• Utilized high-throughput adjustable deformability cytometry to extract 12 deformability parameters from cell contours.
• Employed machine learning, including a back propagation (BP) neural network, for cell identification.
• Applied time-series analysis of cell deformation and transfer learning for clinical sample identification.

Main Results:

• Multiparameter analysis achieved over 80% accuracy in identifying cells with cytoskeletal modifications.
• Time-series analysis of mechanical parameters correlated with cell type, leading to over 90% accuracy in cell line detection.
• Transfer learning with the BP neural network model enabled approximately 95% accuracy in identifying clinical samples.

Conclusions:

• Multiparameter mechanical phenotyping offers a robust method for high-throughput cell analysis.
• The integration of machine learning and time-series analysis significantly improves cell identification accuracy.
• This approach shows promise for applications in cell biology research and clinical diagnostics, especially for rare cell detection.