NucleiMix: Realistic data augmentation for nuclei instance segmentation.

Area of Science:

• Pathology image analysis
• Computational biology
• Medical imaging

Background:

• Nuclei instance segmentation is crucial for pathology image analysis and downstream applications.
• Existing methods face challenges with imbalanced datasets, particularly concerning rare nuclei types.
• Public datasets have advanced research but haven't fully resolved data imbalance issues.

Purpose of the Study:

• To introduce NucleiMix, a data augmentation technique to address data imbalance in nuclei instance segmentation.
• To enhance the representation of rare nuclei types within pathology datasets.
• To improve the accuracy and robustness of nuclei segmentation and classification models.

Main Methods:

• NucleiMix employs a two-phase approach for data augmentation.
• Phase 1: Identifies candidate locations and inserts rare-type nuclei.
• Phase 2: Uses a diffusion model for progressive inpainting to integrate new nuclei.

Main Results:

• NucleiMix effectively synthesizes realistic rare-type nuclei.
• The method significantly enhances nuclei segmentation and classification quality.
• Evaluations on public datasets demonstrate superior performance with popular segmentation models.

Conclusions:

• NucleiMix offers a robust solution for data imbalance in nuclei instance segmentation.
• The technique improves the accuracy and reliability of automated pathology image analysis.
• This method holds potential for advancing diagnostic capabilities in digital pathology.