Recursive Additive Complement Networks for Cell Membrane Segmentation in Histological Images.

Area of Science:

• Digital pathology
• Medical image analysis
• Computational biology

Background:

• Accurate cell membrane segmentation is crucial for calculating cell areas and Nuclear/Cytoplasmic (N/C) ratios.
• The N/C ratio is a key biomarker for diagnosing early hepatocellular carcinoma.
• Traditional methods struggle with precise segmentation, especially in unstained histological images.

Purpose of the Study:

• To introduce a novel Recursive Additive Complement Network (RacNet) for segmenting cell membranes as closed lines in histological images.
• To enhance the accuracy of cell area and N/C ratio estimation for improved diagnostic capabilities.
• To evaluate RacNet's performance across different imaging modalities.

Main Methods:

• Development of RacNet, integrating a complement network and an element-wise maximization (EWM) process.
• Recursive application of the network to refine segmentation outputs.
• Comparative analysis of RacNet performance using bright-field, dark-field, and phase-contrast imaging on unstained hepatic sections.

Main Results:

• RacNet significantly improved the accuracy of segmenting cell membranes as closed lines.
• The EWM process effectively mitigated the complement network's tendency to delete segmented membrane parts.
• Phase-contrast imaging yielded the highest accuracy among the tested methods.

Conclusions:

• RacNet offers a robust solution for accurate cell membrane segmentation in digital pathology.
• The method holds promise for improving the diagnostic accuracy of early hepatocellular carcinoma through enhanced N/C ratio estimation.
• Phase-contrast imaging is recommended for optimal results with RacNet on unstained hepatic tissues.