Self-Supervised X-Ray Coronary Angiography Segmentation with Vessel-Aware Synthesis Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Interventions

Background:

• Percutaneous coronary intervention (PCI) demands precise, real-time coronary vessel segmentation for safety and efficacy.
• Current self-supervised methods excel in topological structure but neglect vessel texture and geometric details.
• Existing frameworks have object model defects in simulating X-ray attenuation.

Purpose of the Study:

• To develop an advanced self-supervised method for coronary vessel segmentation that incorporates texture and geometric details.
• To improve the accuracy and detail capture in vascular segmentation for PCI.
• To reduce the reliance on manual annotations in medical image analysis.

Main Methods:

• Utilizes mixed real and synthetic data for training.
• Simulates X-ray brightness attenuation using the product of ray attenuation rates, unlike traditional addition methods.
• Incorporates vessel texture synthesis and geometric contour bending into the self-supervised segmentation framework.

Main Results:

• Vessel Aware Synthesis Learning significantly reduces segmentation errors, with a 9.6% increase in Jaccard index.
• The method captures more vascular details, showing a 13.8% increase in Recall.
• Effectively eliminates the need for labor-intensive manual annotations.

Conclusions:

• The proposed method offers a practical approach to enhancing AI-driven medical image analysis for PCI.
• It provides a cost-effective solution for developing reliable AI tools by minimizing annotation requirements.
• Improves the accuracy and detail of coronary vessel segmentation in a self-supervised manner.