Factors affecting the performance of a novel artificial intelligence-based coronary computed tomography-derived...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• AI-QCTischaemia is an FDA-cleared AI tool for predicting myocardial ischaemia from CCTA.
• Understanding discrepancies between AI-QCTischaemia and PET perfusion is crucial for clinical application.

Purpose of the Study:

• To identify factors associated with discrepancies between AI-QCTischaemia and [15O]H2O PET perfusion.
• To analyze patient characteristics and coronary artery disease features that influence AI-QCTischaemia accuracy.

Main Methods:

• Analysis of 662 patients with suspected obstructive coronary artery disease (CAD) undergoing CCTA and [15O]H2O PET.
• Utilized AI-QCTischaemia for ischaemia prediction and multivariable logistic regression to identify discrepancy factors.
• Measured perfusion homogeneity using relative flow reserve.

Main Results:

• 32% of patients (209/662) showed discrepancies between AI-QCTischaemia and PET.
• False positive AI-QCTischaemia (normal AI, abnormal PET) associated with female sex, older age, less typical angina, and less advanced CAD.
• False negative AI-QCTischaemia (abnormal AI, normal PET) associated with male sex, smoking, poorer CCTA quality, and more advanced CAD; partly explained by microvascular disease.

Conclusions:

• Factors like age, typical angina, stenosis severity, and atheroma volume predict true positive AI-QCTischaemia.
• Perfusion abnormalities in false negative cases may relate to microvascular disease, highlighting a limitation of AI-QCTischaemia in certain scenarios.