Related Experiment Video
Updated: Jul 5, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Curriculum Consistency Learning and Multi-Scale Contrastive Constraint in Semi-Supervised Medical Image Segmentation
1Department of Computer and Information Engineering, School of Science and Engineering, The Chinese University of Hong Kong (Shenzhen), Shenzhen 518000, China.
Bioengineering (Basel, Switzerland)
|January 22, 2024
Summary
This study introduces a novel curriculum consistency approach for semi-supervised medical image segmentation, improving neural network learning with sparse data. The method enhances feature representation and model generalization, significantly boosting segmentation accuracy.
Area of Science:
- Medical Image Analysis
- Artificial Intelligence
- Computer Vision
Background:
- Data scarcity is a major hurdle in medical image segmentation.
- Semi-supervised learning offers a solution by utilizing limited labeled data.
- Existing methods often struggle with optimal convergence and generalization.
Purpose of the Study:
- To introduce a curriculum consistency constraint for semi-supervised medical image segmentation.
- To enhance neural network learning by simulating human learning processes.
- To improve feature representation and model generalization.
Main Methods:
- Implemented a curriculum consistency constraint by dynamically adapting patch sizes.
- Employed multi-scale contrast learning to extract richer semantic and point-wise features.
- Utilized features from multiple network layers for comprehensive representation.
Main Results:
- Achieved a 9.2% increase in mean intersection over union (mIoU) on the Kvasir-SEG dataset.
- Outperformed state-of-the-art semi-supervised methods in medical image segmentation.
- Demonstrated improved convergence optima and generalization capabilities.
Conclusions:
- The proposed curriculum consistency constraint is effective for semi-supervised medical image segmentation.
- Multi-scale contrast learning enhances feature representation and model performance.
- The approach successfully addresses data scarcity challenges in medical imaging.
