Is Artificial Intelligence-Based Quantitative Coronary Angiography Ready for Clinical Adoption?

Artificial intelligence-based quantitative coronary angiography (AI-QCA) is an innovative tool designed to automatically assess coronary artery lesions during angiography. By analyzing images in real-time, AI-QCA aims to assist clinicians in evaluating stenosis severity and planning interventions with greater efficiency and consistency. ¹ However, as with any new technology, close examination of its accuracy and reliability is essential, especially in preparation for its potential integration into routine clinical practice.

One concern is the use of a single angiographic frame for lesion identification, which raises questions about potential biases. Relying on just one frame may inadequately capture the complexity of coronary anatomy, increasing the risk of missed diagnoses. ² Such an approach could lead to underestimation of disease extent, particularly in cases involving diffuse or complex lesions.

In a study by Kim et al., ³ AI-QCA failed to detect 11% of lesions, particularly in distal coronary segments. This is concerning, as missed lesions could result in incomplete revascularization or suboptimal management of coronary artery disease. The impact on clinical practice is significant: undetected lesions in patients with multivessel disease or distal vessel involvement could compromise patient outcomes.

Despite the promise of AI-QCA, recent studies reveal discrepancies when comparing AI-QCA assessments with intravascular ultrasound (IVUS) measurements, which may affect clinical outcomes. For example, a study by Moon et al. found that AI-QCA demonstrated moderate to strong correlation with IVUS in assessing coronary lesions but notably underestimated vessel size and lesion length, particularly at distal reference points. ⁴ Such discrepancies are critical, as precise assessment of lesion dimensions is essential for optimal stent sizing and placement during percutaneous coronary intervention. Underestimation of lesion dimensions could lead to the selection of stents that are too small or too short, potentially increasing the risk of restenosis or stent thrombosis.

Another limitation arises from AI-QCA’s tendency to misclassify tandem lesions as single lesions, which complicates clinical decision-making. ⁵ This misclassification can result in the undertreatment of critical segments, especially in complex multivessel disease, where comprehensive lesion detection is crucial. Additionally, AI-QCA’s reliance on two-dimensional imaging for lesion characterization lacks the precision of IVUS in identifying plaque characteristics, such as calcification or vessel tortuosity, both of which are crucial for procedural planning. Studies show that highly calcified lesions pose challenges for AI-QCA, which struggles to accurately assess calcified plaque burden, an important limitation in cases involving complex lesions.

Moreover, findings underscore the importance of manually adjusting lesion margins in AI-QCA to improve its correlation with IVUS measurements, indicating that a fully automated system may currently lack the accuracy required for clinical adoption. ⁴ Geographic mismatches between AI-QCA and IVUS especially at distal lesion margins suggest that further refinement in lesion boundary detection is necessary before AI-QCA can reliably guide clinical interventions without supplementary imaging.

In conclusion, while AI-QCA offers a promising advancement in automated coronary angiography, observed discrepancies in comparative studies underscore the need for further validation and refinement. Ensuring accurate lesion characterization and dimension assessment is crucial before AI-QCA can be fully integrated into routine clinical practice. For now, AI-QCA should be viewed as a complementary tool to traditional imaging modalities, especially in cases involving complex coronary anatomy.

Author’s Contribution

H.K.R. conceived this idea predominantly, wrote the first draft of the letter, and critically revised and edited successive drafts of the manuscript.