REU-Net: Region-enhanced nuclei segmentation network.

Area of Science:

• Computational pathology
• Medical image analysis
• Deep learning for biomedical imaging

Background:

• Nuclei segmentation is crucial for automated pathological screening.
• Challenges include clustered nuclei, appearance variability, and complex backgrounds.
• Existing methods struggle with these complexities.

Purpose of the Study:

• To develop a novel multi-task region-enhanced nuclei segmentation network (REU-Net).
• To improve the accuracy and robustness of nuclei segmentation in pathological images.
• To address challenges posed by nuclei clustering and complex backgrounds.

Main Methods:

• Proposed a multi-task region-enhanced nuclei segmentation network (REU-Net).
• Employed a U-shaped architecture with serial and parallel U-structures for multi-task learning.
• Integrated auxiliary tasks (contour extraction, rough segmentation) with attention gates to enhance saliency regions and features.

Main Results:

• REU-Net demonstrated superior performance in nuclei segmentation.
• The multi-task approach with region enhancement improved fine segmentation accuracy.
• The method effectively refined nuclei and contour details using aggregated spatial and texture features.

Conclusions:

• REU-Net offers a significant advancement in nuclei segmentation for pathological screening.
• The proposed architecture effectively handles nuclei clustering and complex backgrounds.
• REU-Net outperforms state-of-the-art methods on multiple benchmark datasets (HUSTS, MoNuSeg, CoNSep, CPM-17).