Multiplex aggregation combining sample reweight composite network for pathology image segmentation.

Area of Science:

• Digital Pathology
• Medical Image Analysis
• Computer Vision

Background:

• Nuclei segmentation is crucial for pathological image analysis, impacting diagnosis and research.
• Existing methods face challenges including blurred boundaries, domain shifts, and uneven nuclei distribution.
• These obstacles hinder accurate and reliable pathological image segmentation.

Purpose of the Study:

• To propose an innovative network, CDNet, for robust nuclei segmentation in digital pathology.
• To address limitations of current segmentation techniques, enhancing accuracy and generalization.
• To improve pathological image analysis for better diagnostic and research outcomes.

Main Methods:

• Introduced CDNet, incorporating Diversified Aggregation Convolution (DAC) for boundary clarity.
• Integrated a Causal Inference Module (CIM) for enhanced cross-domain generalization.
• Developed a Stable-Weighted Combined loss function addressing uneven nuclei distribution.

Main Results:

• CDNet demonstrated superior performance on MoNuSeg, GLySAC, and MoNuSAC datasets.
• Achieved significant improvements in mean Intersection over Union (mIoU) and Dice Similarity Coefficient (DSC).
• Showcased strong generalization capabilities across diverse pathological image datasets.

Conclusions:

• CDNet effectively overcomes key challenges in nuclei segmentation for digital pathology.
• The proposed DAC, CIM, and loss function contribute to improved segmentation accuracy and robustness.
• CDNet offers a promising advancement for pathological image analysis and clinical applications.