Related Experiment Video
Updated: Sep 24, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Semi-supervised medical image segmentation via a tripled-uncertainty guided mean teacher model with contrastive
Kaiping Wang1, Bo Zhan1, Chen Zu2
1School of Computer Science, Sichuan University, Chengdu, China.
Medical Image Analysis
|May 5, 2022
Summary
This study introduces a novel semi-supervised learning method for medical image segmentation, utilizing auxiliary tasks and uncertainty guidance to improve accuracy with limited labeled data.
Area of Science:
- Medical Imaging
- Computer Vision
- Machine Learning
Background:
- Limited labeled data hinders medical image segmentation accuracy.
- Current semi-supervised methods rely on data-level and model-level consistency.
- Unlabeled data holds potential for improving segmentation models.
Purpose of the Study:
- To enhance medical image segmentation using semi-supervised learning.
- To leverage unlabeled data through auxiliary tasks and task-level consistency.
- To address teacher model bias in semi-supervised learning.
Main Methods:
- Implemented auxiliary tasks: foreground/background reconstruction and signed distance field (SDF) prediction.
- Utilized a mean teacher architecture with a tripled-uncertainty guided framework.
- Incorporated contrastive learning and an uncertainty weighted integration (UWI) strategy.
Main Results:
- Demonstrated effectiveness on the ACDC and PROMISE12 datasets.
- Showcased improved medical image segmentation performance.
- Validated the mutual promotion effect between auxiliary and segmentation tasks.
Conclusions:
- The proposed method effectively utilizes unlabeled data for medical image segmentation.
- Auxiliary tasks and uncertainty guidance enhance representation learning.
- The approach offers a promising solution for segmentation with scarce labeled data.
