Dual-Student Adversarial Framework With Discriminator and Consistency-Driven Learning for Semi-Supervised Medical...

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Semi-supervised medical image segmentation reduces manual annotation costs but faces challenges with unreliable pseudo-labels and confirmation bias.
• Existing methods often exhibit unstable optimization and performance degradation due to these limitations.

Purpose of the Study:

• To propose a novel dual-student adversarial framework for robust semi-supervised medical image segmentation.
• To address limitations of existing methods by improving pseudo-label quality and training stability.

Main Methods:

• Introduced a dual-student adversarial framework incorporating an adversarial learning-based segmentation refinement (ALSR) module for prediction diversity and pseudo-label refinement.
• Employed a residual exponential moving average (R-EMA) within uncertainty estimation with inter-instance consistency measurement (UIM) for a stable teacher model and uncertainty-based filtering.
• Developed a Contrastive Representation Stabilization (CRS) module for enhanced voxel-level semantic alignment using contrastive learning on confident regions.

Main Results:

• The proposed method consistently outperformed state-of-the-art approaches in extensive experiments on benchmark datasets.
• Demonstrated improved segmentation accuracy and stability compared to existing semi-supervised methods.

Conclusions:

• The dual-student adversarial framework offers a robust solution for semi-supervised medical image segmentation.
• The integrated ALSR, R-EMA, UIM, and CRS modules effectively enhance pseudo-label reliability, training stability, and feature discriminability.