Label-Free Holographic Imaging Flow Cytometry With Deep-Learning-Based Detection and Classification of Thousands of...

Area of Science:

• Biomedical Engineering
• Computational Biology
• Optical Imaging

Background:

• Label-free imaging flow cytometry is crucial for real-time cell analysis.
• Current methods often require chemical staining or lack speed for high-throughput applications.
• Digital holography offers detailed cellular structure information without staining.

Purpose of the Study:

• To introduce a novel end-to-end neural network for real-time, label-free cell detection and classification.
• To achieve high-speed imaging flow cytometry using digital holography and quantitative phase imaging.
• To demonstrate the method's adaptability across different cell types and conditions.

Main Methods:

• Developed a custom two-stage neural network integrating fixed convolution layers with image processing filters for detection.
• Utilized two subsequent convolutional layers for classifying detected cells.
• Employed label-free quantitative imaging flow cytometry based on digital holography for cell imaging during flow.

Main Results:

• Achieved cell detection and classification in 0.44 milliseconds, enabling real-time analysis.
• Demonstrated over 10x speedup compared to existing methods like YOLOv8n.
• Successfully validated the method on T-cells and cancer cells, showing adaptability.
• Enabled imaging, detection, and classification of thousands of cells per second.

Conclusions:

• The proposed neural network approach facilitates high-throughput, label-free imaging flow cytometry.
• This method offers a significant speedup for real-time cell analysis, reducing computational complexity.
• Potential applications include real-time cell monitoring, early disease detection, and high-speed diagnostics.