DL-CSPF: deep-learning-based cell segmentation with a physical framework for digital holographic microscopy.

Area of Science:

• Biomedical Imaging
• Cell Biology
• Computational Biology

Background:

• Digital holographic microscopy (DHM) provides label-free, quantitative phase imaging of live cells.
• Accurate cell segmentation is essential for quantitative analysis but challenging due to complex cellular states and imaging conditions.

Purpose of the Study:

• To introduce DL-CSPF, a deep-learning-based method for high-precision live-cell segmentation.
• To address challenges in cell segmentation accuracy for long-term quantitative analysis.

Main Methods:

• DL-CSPF employs two neural networks for foreground-background segmentation and cell detection, generating foreground edges and seed points.
• A marker-controlled watershed algorithm utilizes these features for cell segmentation.
• The method focuses on low information entropy features, reducing dataset requirements and eliminating manual parameter tuning.

Main Results:

• DL-CSPF demonstrated accurate segmentation across diverse datasets (HeLa, pollen, COS-7 cells).
• Validated feasibility and generalization in various imaging conditions.
• Reliably characterized and quantitatively analyzed morphological metrics during long-term live-cell imaging.

Conclusions:

• DL-CSPF offers a robust and accurate solution for live-cell segmentation using DHM.
• The method facilitates precise quantitative analysis of cellular morphology throughout the lifecycle.
• DL-CSPF is a promising tool for advancing biomedical research.