Related Experiment Video
Updated: May 1, 2026
10:30
Analysis of Tubular Membrane Networks in Cardiac Myocytes from Atria and Ventricles
Published on: October 15, 2014
20.4K
TSSP-UNet: A Two-Stage Weakly Supervised Pathological Image Segmentation With Point Annotations
Shaoqiang Wang1, Guiling Shi1, Yuchen Wang1
1Qingdao University of Technology, Qingdao, China.
IET Systems Biology
|March 3, 2026
Summary
This study introduces TSSP-UNet, a novel two-stage weakly supervised segmentation method. It effectively improves cell nucleus segmentation accuracy using pseudo-labels and refined learning, outperforming baseline approaches.
Area of Science:
- Medical image analysis
- Computer vision
- Machine learning
Background:
- Deep convolutional neural networks excel at image segmentation but struggle with complex instances and annotation costs.
- Weakly supervised learning offers a solution by utilizing less precise annotations or algorithm-derived supervision.
Purpose of the Study:
- To develop an effective two-stage weakly supervised segmentation approach for complex image segmentation tasks.
- To address the challenges of high-precision data annotation in medical imaging.
Main Methods:
- Proposed TSSP-UNet, a two-stage segmentation network incorporating constraint and attention mechanisms on pseudo-labels.
- Utilized boundary and superpixel information, along with contour enhancement via pseudo-labels and binary masks.
- Employed a feature aggregation network for foreground segmentation and a confident learning algorithm for pseudo-label refinement.
Main Results:
- TSSP-UNet demonstrated strong performance in weakly supervised cell nucleus segmentation.
- The approach showed significant improvements compared to baseline methods on the MoNuSeg and TNBC datasets.
Conclusions:
- The proposed TSSP-UNet effectively handles complex instances and reduces annotation dependency in image segmentation.
- This method offers a promising solution for accurate cell nucleus segmentation in challenging datasets.
