Detecting and segmenting cell nuclei in two-dimensional microscopy images.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Digital Pathology

Background:

• Cell nuclei are critical biomarkers for understanding cellular functions and disease states.
• Accurate segmentation of cell nuclei from microscopy images is essential for quantitative analysis.
• Variability in nuclei appearance across different tissues and staining methods necessitates robust segmentation techniques.

Purpose of the Study:

• To develop an unsupervised method for automatic cell nuclei detection and segmentation in 2D microscopy images.
• To improve the accuracy and robustness of nuclei segmentation, addressing challenges posed by diverse image characteristics.

Main Methods:

• An unsupervised approach utilizing matching-based detection and edge map generation across multiple blurring levels.
• Edge selection in polar space and iterative contour refinement within an edge pyramid.
• Validation on hematoxylin and eosin-stained liver and Papanicolaou-stained thyroid datasets.

Main Results:

• The proposed method demonstrates competitive nuclei detection efficiency compared to supervised template matching.
• Achieved superior segmentation performance against four state-of-the-art methods.
• Reported average segmentation errors of 10.34% (area error rate) and 0.33 (normalized sum of distances).

Conclusions:

• The developed method provides automatic and accurate cell nuclei segmentation from microscopy images, even with noisy backgrounds.
• The technique shows significant potential for application in clinical diagnostic settings.
• Highlights the effectiveness of unsupervised learning for complex image segmentation tasks.