Deep Learning–Based Segmentation of Coronary Arteries and Stenosis Detection in X-Ray Coronary Angiography
Authors' conclusion
The models achieved performance levels on par with experts in coronary artery segmentation and detection of significant stenoses in the main arteries.
Study context
- Population
- Invasive Coronary Angiography (ICA) images from Amsterdam University Medical Centers and Emory University HospitalICA images for training
- Participants
- 10,573 images
- Comparison
- Deep learning models vs expert variability
- Outcome
- Coronary artery segmentation (Dice coefficient) and detection of ≥50% stenoses on ICA
Methodological detailsretrospective, 2 centres, external validation
- Retrospective
- 2 centres
- External validation
Performance
As reported in the abstract
| Test or model | Segmentation median Dice coefficient (IQR 0.81-0.88) | Stenosis detection rate (95% CI 0.63-0.71) | Stenosis detection agreement (95% CI 0.63-0.68) |
|---|---|---|---|
| Both centers model | 0.86 | 0.67 | |
| Experts | 0.65 |
Segmentation median Dice coefficient (IQR 0.81-0.88)
Both centers model0.86
Experts— not stated in the abstract
Stenosis detection rate (95% CI 0.63-0.71)
Both centers model0.67
Experts— not stated in the abstract
Stenosis detection agreement (95% CI 0.63-0.68)
Both centers model— not stated in the abstract
Experts0.65
Further findings
Stenosis detection, model trained on the data with the most stenoses
- Highest stenosis detection rate
- 0.67 (95% CI 0.64-0.71)
Models matched for data set size and proportion of stenoses
- Models trained on both centers performed similarly
Original publication
Deep Learning–Based Segmentation of Coronary Arteries and Stenosis Detection in X-Ray Coronary Angiography. JACC Adv. Published December 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →