Consistency and adversarial semi-supervised learning for medical image segmentation.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computer Vision

Background:

• Deep learning significantly advances medical image segmentation.
• Current methods require large labeled datasets, which are costly and time-consuming.
• There is a need for efficient semi-supervised methods to overcome data limitations.

Purpose of the Study:

• To propose a novel semi-supervised medical image segmentation method.
• To leverage adversarial training and collaborative consistency learning to improve performance with limited labeled data.
• To enhance the exploitation of unlabeled data for more accurate segmentation.

Main Methods:

• Integration of adversarial training and collaborative consistency learning into a mean teacher model.
• Utilizing a discriminator to generate confidence maps for unlabeled data.
• Employing an auxiliary discriminator to enhance the quality of supervised information during adversarial training.

Main Results:

• The proposed method was evaluated on three challenging tasks: skin lesion, optic cup/disk, and lower-grade glioma tumor segmentation.
• Experimental results demonstrated the superiority of the proposed method over existing state-of-the-art semi-supervised techniques.
• The method effectively utilizes unlabeled data to improve segmentation accuracy.

Conclusions:

• The novel semi-supervised approach significantly improves medical image segmentation performance.
• Adversarial training and collaborative consistency learning are effective strategies for reducing reliance on labeled data.
• The method shows strong potential for various clinical applications requiring accurate image segmentation.