Small U-Net for Fast and Reliable Segmentation in Imaging Flow Cytometry.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Cellular Imaging

Background:

• Accurate cell segmentation is crucial for imaging flow cytometry (IFC) and deformability cytometry (DC) in disease diagnostics.
• Traditional thresholding methods lack accuracy with low-contrast images, while standard neural networks are too slow for high-throughput applications.
• Existing methods struggle to balance speed and accuracy in IFC segmentation.

Purpose of the Study:

• To develop a rapid and accurate image segmentation method for IFC.
• To improve cell contour detection for enhanced cellular morphology analysis and cell counting.
• To address the limitations of traditional and standard neural network segmentation in high-throughput cytometry.

Main Methods:

• Developed a small U-Net model optimized for speed and accuracy.
• Trained the model on high-quality, curated, and annotated IFC data.
• Evaluated performance against traditional thresholding methods and standard U-Net on CPU.

Main Results:

• Achieved a 35× speed improvement on CPU compared to standard U-Net.
• Demonstrated superior accuracy in cell contour detection over traditional methods.
• Significantly reduced systematic measurement errors in blood sample analysis using DC.

Conclusions:

• The optimized U-Net model provides a fast and accurate segmentation solution for IFC.
• This method enhances the reliability of cellular morphology analysis and disease marker detection.
• The developed tools are adaptable for diverse IFC applications, supporting high-throughput analysis.