Automated stenosis estimation of coronary angiographies using end-to-end learning
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
The initial evaluation of coronary stenosis during coronary angiography is typically performed by visual assessment. The visual assessment of coronary angiographies has limited accuracy compared to quantitative methods like fractional flow reserve and quantitative coronary angiography. Quantitative methods are also more time-consuming and costly.
Objectives
To test whether applying deep-learning-based image analysis to coronary angiographies might yield a faster and more accurate stenosis estimation than visual assessment.
Methods
We developed deep learning models for predicting coronary artery stenosis using 332,582 multi-frame x-ray images (cine loops) from 19,414 patients undergoing coronary angiography. The curated dataset for model development included 13,840 patients, with 62,165 cine loops of the left coronary artery and 31,161 cine loops of the right coronary artery.
Results
For identification of significant coronary stenosis (visual assessment of diameter stenosis >70%), our model obtained a receiver operator characteristic (ROC) area under the curve (ROC-AUC) of 0.903 (95% CI: 0.900-0.906) on the internal test set with 5,056 patients. The performance was evaluated on an external test set with 608 patients against visual assessment, 3D quantitative coronary angiography, and fractional flow reserve (≤ 0.80), obtaining ROC AUC values of 0.833 (95% CI: 0.814-0.852), 0.798 (95% CI: 0.741-0.842, and 0.780 (95% CI: 0.743-0.817), respectively.
Conclusion
For assessment of coronary stenosis during invasive coronary angiography a deep-learning-based model showed promising results for predicting visual assessment (ROC AUC of 0.903). Compared to previous work, our approach demonstrates performance increase, includes all 16 segments, does not exclude revascularized patients, is externally tested, and is simpler using fewer steps and fewer models.
