A neural network approach for real-time particle/cell characterization in microfluidic impedance cytometry

Area of Science:

• Biophysics
• Microfluidics
• Machine Learning

Background:

• Real-time particle classification is crucial for microfluidic applications like sorting and enrichment.
• Existing methods often lack the speed and accuracy required for dynamic analysis.

Purpose of the Study:

• To develop a fast, label-free particle characterization technique using neural networks and microfluidic impedance cytometry.
• To enable real-time multiparametric analysis of particle properties within microfluidic devices.

Main Methods:

• A recurrent neural network was designed to process impedance data from a novel microfluidic chip.
• The network was trained and validated using both synthetic and experimental datasets.
• The system performs label-free characterization of particle size, velocity, and cross-sectional position.

Main Results:

• The trained neural network achieved accurate characterization of beads, red blood cells, and yeasts.
• Unitary prediction time was as low as 0.4 milliseconds, enabling real-time analysis.
• The approach demonstrated effective multiparametric analysis of impedance data streams.

Conclusions:

• The combination of microfluidic impedance cytometry and recurrent neural networks offers a powerful tool for real-time particle analysis.
• This method can be extended to various cell types and device designs for electrical parameter extraction.
• The approach serves as a foundation for advanced real-time single-cell analysis and sorting applications.