DOSTA-Net: Domain-Shuffle Temporal Attention Network for Vessel Extraction in X-Ray Coronary Angiography Using...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Diagnosis

Background:

• Accurate artery extraction from X-ray coronary angiography (XCA) is crucial for diagnosing coronary artery diseases.
• Image quality in XCA is often degraded by superimposed tissues and cardiac motion, hindering traditional methods.
• Lack of large, annotated datasets limits deep learning for vessel extraction.

Purpose of the Study:

• To develop a novel deep learning framework for training-free vessel extraction from XCA images.
• To address the challenge of limited annotated data by leveraging synthetic temporal XCA data.
• To improve the robustness and accuracy of coronary artery segmentation.

Main Methods:

• A pipeline for synthesizing large-scale, realistic temporal XCA data with anatomical variations and artifacts was developed.
• A DOmain-Shuffle Temporal Attention Network (DOSTA-Net) was introduced to enhance temporal feature learning by integrating synthetic and real data.
• Pseudo-labeling of real data and an annealing loss function were employed to minimize domain discrepancies and utilize unlabeled real data.

Main Results:

• The proposed framework demonstrated superior vessel segmentation performance on two datasets.
• DOSTA-Net effectively utilized temporal information and mitigated domain gaps between real and synthetic data.
• A reader study confirmed the method's effectiveness through subjective image quality assessment.

Conclusions:

• The novel framework successfully enables training-free deep learning models for artery extraction from XCA images.
• Leveraging synthetic temporal data and domain adaptation techniques significantly enhances vessel segmentation accuracy.
• The approach offers a promising solution for improving coronary artery disease diagnosis and treatment planning.