NuSEA: Nuclei Segmentation With Ellipse Annotations.

Area of Science:

• Digital Pathology
• Computational Biology
• Medical Image Analysis

Background:

• Accurate nuclei segmentation is essential for quantifying the pathological microenvironment.
• Manual annotation of nuclei for deep learning models is labor-intensive and costly.

Purpose of the Study:

• To introduce NuSEA, an efficient tool for accurate nuclei segmentation using simple ellipse annotations.
• To develop a lightweight deep learning model (U-Light) for real-time nuclei segmentation.
• To propose novel loss functions for improved edge segmentation and smoothness.

Main Methods:

• NuSEA utilizes a lightweight U-Light network (0.86 M parameters) for efficient nuclei segmentation.
• Ellipse annotation is employed for rapid and effective nucleus outlining.
• Elliptical Field Loss and Texture Loss are introduced to enhance segmentation accuracy and smoothness.

Main Results:

• NuSEA demonstrates superior performance compared to state-of-the-art methods on public datasets (MoNuSeg, CPM-17, CoNSeP).
• The tool outperforms existing annotation methods based on point, rectangle, and text inputs.
• A new dataset, NuSEA-dataset v1.0, with 118,857 annotated nuclei from 12 organs, was created.

Conclusions:

• NuSEA offers a rapid and effective solution for nuclei annotation in histopathological images.
• The tool facilitates future advancements in deep learning for pathology.
• The released NuSEA-dataset v1.0 will aid further research in the field.