Multi-task contrastive learning for semi-supervised medical image segmentation with multi-scale uncertainty...

Area of Science:

• Medical image analysis
• Computer vision
• Machine learning

Background:

• Automated medical image segmentation is crucial for disease diagnosis and treatment.
• Class imbalance in medical data poses challenges for accurate segmentation, particularly for underrepresented classes.
• Existing semi-supervised methods struggle with noisy pseudo-labels and unclear boundaries due to data imbalance.

Purpose of the Study:

• To develop a novel framework for semi-supervised medical image segmentation that effectively handles class imbalance.
• To improve the segmentation accuracy of tail classes in medical images.
• To enhance the robustness and quality of pseudo-labels generated in semi-supervised learning.

Main Methods:

• A multi-task contrastive learning framework employing a student-teacher model.
• Global image-level contrastive learning in the encoder to mitigate class imbalance.
• Local pixel-level contrastive learning in the decoder for intra-class aggregation and inter-class separation.
• Multi-scale uncertainty-aware consistency loss to reduce pseudo-label noise.

Main Results:

• The proposed method significantly outperforms state-of-the-art semi-supervised segmentation techniques on ACDC, LA, and LiTs datasets.
• Demonstrated superior segmentation performance compared to existing approaches.
• Achieved higher accuracy in segmenting challenging tail classes.

Conclusions:

• Multi-task contrastive learning effectively addresses class imbalance, leading to improved segmentation outcomes.
• Multi-scale uncertainty estimation enhances the reliability of pseudo-labels, boosting overall performance.
• The framework offers a promising solution for accurate semi-supervised medical image segmentation.