Diversity-Enhanced Collaborative Mamba for Semi-Supervised Medical Image Segmentation.

Area of Science:

• Medical image analysis
• Computer vision
• Machine learning

Background:

• High-quality annotated medical data is expensive and time-consuming to acquire.
• Semi-supervised learning reduces annotation burden by using unlabeled data for pseudo-labeling.
• State space models like Mamba excel at capturing long-range dependencies.

Purpose of the Study:

• To explore the potential of advanced state space models in semi-supervised medical image segmentation.
• To propose a novel framework, DCMamba, enhancing diversity across data, network, and features.

Main Methods:

• Developed patch-level weak-strong mixing augmentation tailored for Mamba's scanning.
• Introduced a diverse-scan collaboration module to leverage directional prediction discrepancies.
• Implemented uncertainty-weighted contrastive learning to enrich feature representation diversity.

Main Results:

• DCMamba demonstrated superior performance in semi-supervised medical image segmentation.
• Achieved a 6.69% improvement over the latest SSM-based method on the Synapse dataset with 20% labeled data.

Conclusions:

• DCMamba effectively utilizes data, network, and feature diversity for improved semi-supervised segmentation.
• The proposed framework offers a promising direction for efficient medical image segmentation with limited annotations.