Related Experiment Video
Updated: Aug 25, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Minimizing Estimated Risks on Unlabeled Data: A New Formulation for Semi-Supervised Medical Image Segmentation.
IEEE Transactions on Pattern Analysis and Machine Intelligence
|October 17, 2022
Summary
This study introduces a new semi-supervised segmentation method for biomedical images, reducing costs by utilizing unlabeled data. The approach enhances segmentation accuracy by minimizing risks on both labeled and unlabeled images.
Area of Science:
- Medical image analysis
- Computer vision
- Machine learning
Background:
- Supervised segmentation is expensive due to the need for expert annotations.
- Semi-supervised segmentation offers a cost-effective alternative by leveraging unlabeled data.
- Existing methods often focus only on labeled data risks.
Purpose of the Study:
- To develop a novel semi-supervised segmentation framework for biomedical images.
- To improve segmentation accuracy by utilizing both labeled and unlabeled data effectively.
- To address the high cost of manual annotation in medical imaging.
Main Methods:
- A new risk minimization formulation is proposed.
- An unbiased estimator is developed to incorporate unlabeled data risks.
- A general framework for semi-supervised image segmentation is established.
Main Results:
- The proposed method was validated on cardiac, optic cup/disc, and 3D whole heart segmentation tasks.
- The unbiased estimator proved effective in improving segmentation performance.
- Superior results were achieved compared to state-of-the-art approaches.
Conclusions:
- The developed semi-supervised framework offers a powerful and efficient solution for biomedical image segmentation.
- The risk minimization approach effectively utilizes unlabeled data, reducing annotation costs.
- The method demonstrates significant potential for clinical applications.
