Related Experiment Video
Updated: Jun 25, 2025
Author Spotlight: Advancing Cardiovascular Imaging - Introducing the Spatially Weighted Calcium Score for Early Disease Detection
Published on: September 22, 2023
Real-time coronary artery segmentation in CAG images: A semi-supervised deep learning strategy
Chih-Kuo Lee1, Jhen-Wei Hong2, Chia-Ling Wu2
1Division of Cardiology, Department of Internal Medicine, National Taiwan University Hospital Hsin-Chu Branch, No. 25, Lane 442, Section 1, Jingguo Rd, North District, Hsinchu City 300, Taiwan; Graduate Institute of Clinical Medicine, College of Medicine, National Taiwan University, No.1, Chang-Te St., Taipei 100, Taiwan; Department of Internal Medicine, College of Medicine, National Taiwan University, No.1, Jen Ai road section 1, Taipei 100, Taiwan.
Deep learning segmentation aids coronary artery disease treatment by creating a real-time roadmap for percutaneous coronary intervention (PCI). This semi-supervised approach reduces contrast and radiation exposure, improving patient care.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiology
Background:
- Coronary artery disease (CAD) patients with renal issues require careful management during percutaneous coronary intervention (PCI).
- Minimizing contrast dye and radiation exposure is crucial for these patients.
- Accurate vascular visualization is essential for effective PCI guidance.
Purpose of the Study:
- To develop a deep learning (DL) based semi-supervised method for real-time vascular segmentation.
- To create a real-time roadmap for guiding PCI procedures.
- To reduce the need for extensive manual data labeling in DL models.
Main Methods:
- Utilized a U-Net based teacher-student architecture for segmentation.
- Employed semi-supervised learning with a large dataset of unlabeled coronary angiography (CAG) images.
- Implemented quality control strategies including consistency regularization and RandAugment for model optimization.
Main Results:
- Achieved an average Dice similarity coefficient of 0.9003 for vascular segmentation.
- Outperformed traditional supervised learning methods with limited labeled data.
- Demonstrated superior performance even with as little as 5% labeled data compared to fully supervised methods.
Conclusions:
- Semi-supervised learning effectively overcomes the high cost of data labeling in DL.
- DL-assisted segmentation can provide a real-time roadmap for PCI.
- This approach has the potential to significantly reduce radiation and contrast doses in PCI procedures.
