Enhancing quantitative coronary angiography (QCA) with advanced artificial intelligence: comparison with manual QCA...

Area of Science:

• Cardiovascular Imaging
• Medical Artificial Intelligence
• Interventional Cardiology

Background:

• Manual quantitative coronary angiography (QCA) has limitations in reproducibility and correction.
• Artificial intelligence-based QCA (AI-QCA) was developed to overcome these challenges.
• An updated AI-QCA solution (MPXA-2000) requires performance assessment.

Purpose of the Study:

• To evaluate the performance of an updated AI-QCA solution (MPXA-2000) for coronary lesion detection and quantification.
• To compare AI-QCA performance against manual QCA as the reference standard.
• To demonstrate AI-QCA's superiority over visual estimation in lesion assessment.

Main Methods:

• Multi-center retrospective analysis of 1,076 coronary angiography images from 420 patients.
• Comparison of AI-QCA and visual estimation against manual QCA.
• Lesion detection defined by manual QCA minimum lumen diameter (MLD) falling within AI-QCA or visual estimation boundaries.
• Evaluation of diameter stenosis (DS), MLD, and lesion length (LL) for detected lesions.

Main Results:

• AI-QCA achieved 93% sensitivity in lesion detection, with strong correlations for DS (R²=0.65), MLD (R²=0.83), and LL (R²=0.71) compared to manual QCA.
• Undetected lesions by AI-QCA were <4% in major vessels and >92% sensitivity in side branches.
• AI-QCA significantly outperformed visual estimation in lesion detection (93% vs. 69%, p<0.001) and identifying all lesions in multi-lesion images (86% vs. 33%, p<0.001).

Conclusions:

• The updated AI-QCA solution demonstrates robust, operator-independent performance in detecting and quantifying coronary artery lesions.
• AI-QCA offers reproducible vessel analysis, potentially optimizing angiography-guided interventions through quantitative metrics.
• Automated AI-QCA analysis provides a reliable alternative to manual methods and visual estimation.