Semi-supervised aortic dissections segmentation: A time-dependent weighted feedback fusion framework.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate segmentation of true lumen (TL) and false lumen (FL) is crucial for diagnosing and treating aortic dissection (AD).
• Deep learning models achieve high performance but require substantial labeled data, posing a significant burden on medical professionals.

Purpose of the Study:

• To develop a novel semi-supervised learning framework for aortic dissection segmentation that effectively utilizes unlabeled data.
• To reduce the reliance on extensively labeled datasets for training deep learning models in AD segmentation.

Main Methods:

• Proposed a time-dependent weighted feedback fusion semi-supervised framework for aortic dissection segmentation.
• Integrated a feedback network to refine segmentation by fusing predicted outputs with original image features iteratively.
• Leveraged both labeled and unlabeled data through a feedback loop in the fused feature space to ensure consistency and improve accuracy.

Main Results:

• The proposed framework demonstrated improved segmentation accuracy through iterative refinement and feature fusion.
• The method effectively utilized unlabeled data, alleviating the need for large labeled datasets.
• Outperformed five state-of-the-art semi-supervised segmentation methods on both type-B AD and public datasets.

Conclusions:

• The novel semi-supervised framework offers a robust and accurate solution for aortic dissection segmentation.
• This approach significantly reduces the labeling burden for medical experts while maintaining high performance.
• The feedback fusion mechanism enhances segmentation accuracy and robustness in medical image analysis.