Cell segmentation methods for label-free contrast microscopy: review and comprehensive comparison.

Area of Science:

• Biomedical Imaging
• Cell Biology
• Computational Biology

Background:

• Label-free imaging offers non-destructive biological process study.
• Traditional microscopy (phase contrast, DIC) has artifacts hindering automatic cell segmentation.
• Developing robust segmentation for label-free microscopy is essential.

Purpose of the Study:

• To compare the efficacy of segmentation workflow steps for label-free microscopy.
• To evaluate image reconstruction, foreground segmentation, cell detection, and instance segmentation.
• To identify optimal segmentation methods for various label-free contrast modalities.

Main Methods:

• Collected cell image data from phase contrast, DIC, Hoffman modulation contrast, and quantitative phase imaging.
• Implemented and compared various established and learning-based segmentation algorithms.
• Validated segmentation performance across different microscopy techniques.

Main Results:

• Image reconstruction is critical for applying advanced segmentation methods to label-free images.
• Comprehensive comparison of thresholding, feature-extraction, level-set, graph-cut, and learning-based foreground segmentation.
• Evaluation of seed-point extraction and single-cell segmentation methods for label-free data.
• Validated specific method sets for each microscopy modality.

Conclusions:

• Image reconstruction significantly expands the applicability of segmentation techniques for label-free microscopy.
• Provided a validated framework and open-access resources (data, code) for label-free cell segmentation.
• Facilitated automated and accurate cell analysis in label-free imaging studies.