Artificial Intelligence Powered Real-Time Coronary Stenosis Recognition and Quantification in Angiography

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate, real-time measurement of coronary lesions is crucial for interventional decision-making.
• Current methods for automated coronary lesion analysis are limited in real-time application.

Purpose of the Study:

• To develop and validate a real-time deep learning system for automated detection and quantification of stenotic lesions in coronary angiography.

Main Methods:

• A deep learning model was trained on 2651 coronary angiographic images from 502 patients.
• The system integrates vessel classification, keyframe selection, lesion detection, vessel segmentation, and quantitative coronary angiography (QCA).

Main Results:

• The system achieved high accuracies in vessel type classification (up to 96.33%) and keyframe selection (up to 98.29%).
• Lesion detection and segmentation/QCA metrics demonstrated strong performance (recall/precision 0.93/0.89 for detection; accuracy >0.92 for segmentation/QCA).
• The complete system identified stenotic lesions within 2 minutes with over 80% clinical satisfaction.

Conclusions:

• The developed real-time deep learning system shows significant potential for improving diagnostic accuracy in coronary angiography.
• This automated approach can streamline clinical workflows, aiding timely decision-making during interventions.