Keep it accurate and robust: An enhanced nuclei analysis framework.

Area of Science:

• Digital pathology
• Computational biology
• Medical image analysis

Background:

• Accurate nuclei segmentation and classification in histology images are crucial for disease diagnosis but are hindered by data variability and tissue complexity.
• Current methods often fail due to limited dataset diversity, leading to suboptimal model performance.

Purpose of the Study:

• To develop a robust framework for histological nuclei analysis that overcomes limitations of existing methods.
• To improve the accuracy and reliability of nuclei segmentation and classification in complex histology images.

Main Methods:

• A novel framework employing dual-level ensemble modeling (intra-ensembling and inter-ensembling) to address limited dataset variation.
• Enhancements to the HoVer-Net architecture, including updated encoders, nested dense decoding, and a regularization strategy.

Main Results:

• Achieved state-of-the-art results on public benchmarks, securing 1st place in nuclear composition prediction and 3rd place in segmentation/classification at the 2022 CoNIC Challenge.
• Demonstrated the effectiveness of the proposed dual-level ensemble modeling and HoVer-Net enhancements for accurate histological nuclei analysis.

Conclusions:

• The proposed framework significantly advances the state-of-the-art in histological nuclei analysis, offering a robust solution for segmentation and classification.
• The study's findings provide insights into optimal network design and training techniques, with code and models to be released for community use.