Dual consistency regularization with subjective logic for semi-supervised medical image segmentation.

Area of Science:

• Computer Vision
• Medical Image Analysis
• Machine Learning

Background:

• Semi-supervised learning is crucial for reducing data labeling costs in medical imaging.
• Existing methods like consistency regularization and pseudo-labeling can be misled by poor awareness of unlabeled data.

Purpose of the Study:

• To propose a novel dual consistency regularization method using subjective logic for semi-supervised medical image segmentation.
• To enhance model guidance by addressing the limitations of current approaches in handling unlabeled data.

Main Methods:

• Introduced subjective logic to estimate uncertainty in semi-supervised medical image segmentation.
• Developed dual consistency regularization under weak and strong perturbations based on the consistency hypothesis.
• Evaluated the method on ACDC, LA, and Pancreas datasets.

Main Results:

• The proposed method demonstrated improved performance compared to existing state-of-the-art (SOTA) techniques.
• Effectively guided the model's learning process using unlabeled data by estimating uncertainty.

Conclusions:

• Dual consistency regularization with subjective logic offers a promising approach for semi-supervised medical image segmentation.
• The method enhances the utilization of unlabeled data and improves segmentation accuracy.