Uncertainty in automated stenosis quantification using multiview x-ray coronary angiography videos

Abstract:

The visual interpretation of X-ray coronary angiography, the primary imaging modality for coronary stenosis evaluation, is a difficult task and requires experience and expert knowledge. Automating stenosis assessment can improve confidence in stenosis identification and severity estimation, facilitating decisions regarding revascularization strategies. However, existing methods are predominantly limited to static images or single-view videos, which increases the risk of missing crucial information due to the complex structure of the coronary tree and movement of the heart. We propose a five-step workflow for automated stenosis detection, localization and severity estimation in X-ray angiography videos. For evaluation at the patient-level, multiple videos per patient, captured from different views, were considered. The workflow achieved an overall sensitivity of 58.98% and specificity of 84.15% for stenosis prediction per coronary segment. Sensitivity and specificity for stenosis severity classification were 62.75% and 59.72%, respectively. To assess the impact of multi-view analysis, we compared severity estimation performance for stenoses detected in single- and multi-view projections, demonstrating that only one view is associated with the highest uncertainty. Our findings encourage further refinement and development of the workflow and highlight the importance of multi-view consideration for accurate stenosis evaluation.