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

Abstract

Aims: AI-QCT_ischaemia is an FDA-cleared novel artificial intelligence-guided method that utilizes features from coronary computed tomography angiography (CCTA) to predict myocardial ischaemia.

Objective: To identify factors associated with discrepancy between AI-QCT_ischaemia and positron emission tomography (PET) perfusion.

Methods and results: Six hundred and sixty-two patients with suspected obstructive coronary artery disease (CAD) on CCTA and undergoing [¹⁵O]H₂O PET were analysed using AI-QCT_ischaemia. Multivariable logistic regression identified factors associated with discrepancy. Perfusion homogeneity was measured by relative flow reserve. A total of 209 (32%) patients showed discrepancies: 62 (9%) exhibited normal AI-QCT_ischaemia but abnormal perfusion (false negative AI-QCT_ischaemia), whereas 147 (22%) had abnormal AI-QCT_ischaemia despite normal perfusion (false positive AI-QCT_ischaemia). False positive AI-QCT_ischaemia patients (vs. true positive) were more often females, older, with less typical angina, and less advanced CAD. In multivariable analysis, typical angina [OR 95% CI: 1.796 (1.015-3.179), P = 0.044], diameter stenosis per 1% increase [1.058 (1.036-1.080), P < 0.001], and percent atheroma volume per 1% increase [1.103 (1.051-1.158), P < 0.001] significantly predicted true positive, while age was inversely associated [0.955 (0.923-0.989), P = 0.010]. False-negative AI-QCT_ischaemia patients (vs. true negative) were more often males, smokers, with less good CCTA image quality, and more advanced CAD. However, none was significant in multivariable analysis. Furthermore, false-negative AI-QCT_ischaemia showed more homogenously reduced perfusion by relative flow reserve compared to true positive (median ± IQR: 0.68 ± 0.15 vs. 0.56 ± 0.23, P < 0.001) and 21 (34%) of false negative showed globally reduced perfusion.

Conclusion: For abnormal AI-QCT_ischaemia, younger age, typical angina, more severe stenosis, and more extensive atherosclerosis predicted abnormal PET perfusion. With false negative AI-QCT_ischaemia, perfusion abnormalities were partly explained by microvascular disease.