Related Experiment Video
Updated: Aug 5, 2026
09:03
Exploiting Live Imaging to Track Nuclei During Myoblast Differentiation and Fusion
Published on: April 13, 2019
NS-GUSL: Green U-Shaped Learning for Nuclei Segmentation from Histopathology Images
Catherine Aurelia Christie Alexander1, Vasileios Magoulianitis1,2, Jiaxin Yang1
1Ming Hsieh Department of Electrical and Computer Engineering, University of Southern California (USC), Los Angeles, CA 90089, USA.
Journal of Imaging
|July 27, 2026
Summary
We developed a lightweight Green U-Shaped Learning (NS-GUSL) model for nuclei segmentation in digital pathology. This efficient model achieves high performance and generalizability, offering a low-complexity alternative for edge devices.
Area of Science:
- Digital pathology
- Computational biology
- Medical imaging analysis
Background:
- Nuclei segmentation is crucial for cancer evaluation in digital histopathology.
- Current deep learning methods are computationally complex and struggle with generalization due to staining and acquisition variability.
Purpose of the Study:
- To introduce a transparent, lightweight, and efficient nuclei segmentation model (NS-GUSL).
- To improve generalization capabilities for unseen organs and slide preparations.
Main Methods:
- Developed a Green U-Shaped Learning (NS-GUSL) model with a multi-scale architecture.
- Utilized novel low-confidence sample binarization (LCSB) and morphological post-processing.
- Employed unsupervised representation learning and supervised feature selection.
Main Results:
- NS-GUSL achieved top panoptic segmentation performance and competitive detection quality on the MoNuSeg dataset.
- Demonstrated strong generalizability in external validation experiments.
- The model is compact, computationally efficient, and has a minimal carbon footprint.
Conclusions:
- NS-GUSL offers a high-performance, efficient, and generalizable solution for nuclei segmentation.
- Its low computational complexity makes it suitable for deployment on edge devices in digital pathology.
