Cell detection and segmentation using correlation clustering.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Machine Learning

Background:

• Accurate cell detection and segmentation are crucial for high-throughput biological research.
• Existing methods struggle with transparent cells or varying imaging modalities.

Purpose of the Study:

• To present a novel, versatile learning-based method for cell segmentation in microscopy.
• To address challenges in segmenting transparent cells and improve quantitative analysis.

Main Methods:

• A learning-based classifier detects cell boundaries.
• Superpixels and a weighted region adjacency graph are generated.
• Graph partitioning using correlation clustering segmentation, including a novel length-constrained variant.

Main Results:

• The proposed method shows high performance across different microscopy types (bright field, phase contrast).
• The length-constrained correlation clustering variant achieves state-of-the-art results on several datasets.
• Effective segmentation of transparent and challenging cell types was demonstrated.

Conclusions:

• The developed method offers a robust solution for cell segmentation in diverse microscopy applications.
• This technique enhances the accuracy and efficiency of quantitative biological studies.
• The novel length-constrained approach pushes the boundaries of current segmentation performance.