Masked pre-training and pseudo-vessel enhancement-based semi-supervised cerebral vascular segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Cerebrovascular segmentation in CT angiography (CTA) and CT perfusion (CTP) is crucial for diagnosing cerebrovascular diseases.
• Fully supervised deep learning methods struggle with limited annotated data and complex small vessel structures.

Purpose of the Study:

• Develop a semi-supervised learning framework for high-accuracy cerebrovascular segmentation with minimal labeled data.
• Improve robustness and enhance small vessel continuity in segmentation models.

Main Methods:

• A novel semi-supervised framework integrating dual-scale masked pre-training for vessel continuity.
• Pseudo-vessel augmentation to improve discrimination between true vessels and false positives.
• Multi-level consistency losses to enhance learning under low supervision.

Main Results:

• Achieved 98.7% of fully supervised performance with only 5% labeled data.
• Reached Dice scores of 0.8379 (CTP) and 0.9177 (CTA) under full supervision.
• Outperformed the second-best method by 1.12% (CTP) and 0.87% (CTA).

Conclusions:

• The semi-supervised framework enables accurate cerebrovascular segmentation with minimal labeled data.
• Effectively enhances small vessel continuity and reduces over-segmentation.
• Demonstrates strong robustness and generalizability for clinical applications.