CM-UNet: A Self-Supervised Learning-Based Model for Coronary Artery Segmentation in X-Ray Angiography

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease

Background:

• Accurate coronary artery segmentation is crucial for diagnosing and managing coronary artery disease.
• Limited annotated datasets hinder the development of automated segmentation tools for radiologists.
• Existing methods struggle with segmentation accuracy due to data scarcity.

Purpose of the Study:

• To introduce CM-UNet, a novel approach for accurate coronary artery segmentation.
• To leverage self-supervised pre-training and transfer learning to minimize reliance on extensive manual annotations.
• To enhance diagnostic capabilities for coronary artery disease using AI.

Main Methods:

• Developed CM-UNet, incorporating self-supervised pre-training on unannotated data.
• Utilized transfer learning on limited annotated datasets for fine-tuning.
• Evaluated segmentation performance using Dice score on X-ray angiography images.

Main Results:

• Fine-tuning CM-UNet with 18 annotated images showed a 15.2% decrease in Dice score.
• Baseline models without pre-training experienced a 46.5% drop in Dice score under similar conditions.
• Self-supervised learning demonstrated superior performance and reduced data dependency.

Conclusions:

• Self-supervised learning significantly enhances coronary artery segmentation accuracy.
• CM-UNet reduces the need for large annotated datasets, making AI tools more accessible.
• The approach holds potential for improving clinical workflows and patient outcomes in cardiovascular disease diagnosis.