Topological-Preserving Membrane Skeleton Segmentation in Multiplex Immunofluorescence Imaging.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Machine Learning

Background:

• Multiplex immunofluorescence (MxIF) is crucial for advanced cellular analysis.
• Effective cell segmentation in MxIF data is challenging due to marker variability and complex cellular structures.
• Current watershed-based methods struggle with intricate marker relationships and stain quality issues.

Purpose of the Study:

• To develop a deep learning-based method for accurate cell membrane segmentation in MxIF imaging.
• To aggregate complementary information from multiple MxIF markers for improved segmentation.
• To enhance topology preservation in segmented tubular membrane structures.

Main Methods:

• Proposed a deep learning model integrating global (z-stack projection) and local (individual markers) information.
• Investigated state-of-the-art 2D deep networks and volumetric loss functions.
• Conducted a comprehensive ablation study and introduced a novel volumetric metric for skeletal structures.

Main Results:

• The deep learning model significantly improved segmentation performance.
• Achieved a 20.2% increase in the novel volumetric metric and a 41.3% increase in Adjusted Rand Index (ARI) compared to baseline methods.
• Results were statistically significant (p<0.05) based on Wilcoxon signed rank test.

Conclusions:

• Deep learning offers a promising direction for advancing MxIF cell segmentation.
• The proposed method effectively segments tubular membrane structures by leveraging multi-marker data.
• The developed tools are publicly available for further research and application.