A regularization-driven Mean Teacher model based on semi-supervised learning for medical image segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Semi-supervised learning is crucial for medical image segmentation but often requires extensive labeled data.
• Existing methods struggle with generalization due to limited labeled samples and parameter tuning challenges.

Purpose of the Study:

• To develop a novel regularization-driven Mean Teacher model for semi-supervised medical image segmentation.
• To enhance segmentation performance and model robustness while minimizing the need for labeled data.

Main Methods:

• Introduced a regularization-driven strategy incorporating virtual adversarial training.
• Optimized unsupervised loss and regularization terms using entropy minimization to refine decision boundaries.
• Evaluated the model on the ISIC2017 and COVID-19 CT segmentation datasets.

Main Results:

• Achieved more accurate results in challenging 2D semi-supervised medical image segmentation tasks.
• Demonstrated significant performance improvements compared to state-of-the-art semi-supervised segmentation methods.
• Showcased superior generalization and robustness on diverse medical imaging datasets.

Conclusions:

• The proposed regularization-driven Mean Teacher model offers a robust and accurate solution for semi-supervised medical image segmentation.
• This approach can be extended to various medical segmentation tasks, potentially reducing physician workload.
• The method effectively addresses the limitations of existing techniques by improving generalization with less labeled data.