Related Experiment Video
Updated: Apr 25, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
993
ZScribbleSeg: A comprehensive segmentation framework with modeling of efficient annotation and maximization of
Ke Zhang1, Bomin Wang2, Hangqi Zhou2
1School of Data Science, Fudan University, Shanghai, 200433, China; Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, USA.
Medical Image Analysis
|April 23, 2026
Summary
This study introduces ZScribbleSeg, a novel framework for medical image segmentation using efficient scribble annotations. It achieves competitive performance by maximizing supervision and incorporating spatial priors, reducing manual annotation efforts.
Area of Science:
- Medical Imaging
- Computer Vision
- Machine Learning
Background:
- Medical image segmentation is crucial but requires extensive manual annotation.
- Weakly supervised methods using scribble annotations offer a less labor-intensive alternative.
- Existing scribble-based methods struggle with inaccurate segmentations due to limited supervision.
Purpose of the Study:
- To develop an efficient scribble annotation strategy for medical image segmentation.
- To introduce a framework that integrates efficient scribbles with spatial priors for improved segmentation accuracy.
- To reduce the burden of manual data curation in medical imaging.
Main Methods:
- Investigated principles for efficient scribble annotation forms (supervision maximization, randomness simulation).
- Developed regularization terms for spatial relationships and shape constraints.
- Utilized the Expectation-Maximization (EM) algorithm for estimating label class mixture ratios.
- Integrated efficient scribble supervision with spatial priors into the ZScribbleSeg framework.
Main Results:
- ZScribbleSeg achieves competitive performance on six diverse medical image segmentation tasks.
- The method demonstrates effective segmentation using only scribble annotations.
- The framework successfully incorporates spatial priors and shape constraints for improved accuracy.
Conclusions:
- ZScribbleSeg offers a promising solution for efficient and accurate medical image segmentation.
- The proposed method alleviates the need for fully annotated datasets.
- This approach enhances the feasibility of applying deep learning to medical image segmentation challenges.
