Factors affecting the performance of a novel artificial intelligence-based coronary computed tomography-derived ischemia algorithm against PET perfusion imaging
European Heart Journal - Cardiovascular Imaging, vol. 25
Abstract
Abstract Background An artificial intelligence-based quantitative coronary computed tomography angiography (CCTA) algorithm (AI-QCTischemia) employs 38 morphological features from CCTA aiming to identify ischemic coronary artery disease (CAD) defined as invasive FFR≤0.80. Recent studies have demonstrated good diagnostic accuracy and prognostic value of this algorithm. However, discrepancies between AI-QCTischemia and positron emission tomography (PET) myocardial perfusion have been observed. Purpose To detect the potential factors that are associated with discrepant findings between AI-QCTischemia and PET perfusion. Methods 662 patients, who had suspected obstructive CAD on CCTA and underwent [15O]H2O PET perfusion, were analysed. Demographic and imaging data were compared between the groups with concordant and discrepant findings. Multivariable logistic regressions were used to identify factors independently associated with discrepancy. In addition, PET-based relative flow reserve (RFR) was used to measure blood flow heterogeneity per patient. Results 453 (68%) patients had concordant findings by both methods. Out of the 209 (32%) patients with discrepant findings, 62 patients (9% of total) had normal AI-QCTischemia but abnormal PET perfusion ("false negative", FN group). In 147 patients (22% of total), AI-QCTischemia was abnormal despite having normal PET perfusion ("false positive", FP group). The patients in the FN group as compared to patients with true negative findings were more often males, smokers, had less often good CCTA image quality and had more advanced CAD by means of total calcium score, diameter stenosis, and percent atheroma volume (PAV). However, none of these parameters was statistically significant in multivariable analysis. In addition, more homogenous myocardial perfusion by means of RFR was observed in the FN group as compared to true positive findings (median±IQR: 0.68±0.15 vs. 0.56±0.23, p<0.001). In the FN group, 21 (34%) patients had globally decreased perfusion. The patients in the FP group as compared to patients with true positive findings were more often females, older, and had less often typical angina and less advanced CAD. In multivariable analysis, age per 1 year increase (OR, 95% CI: 0.955 (0.923–0.989), p=0.010), typical angina (1.796 (1.015–3.179), p=0.044), diameter stenosis per 1% increase (1.058 (1.036–1.080), p<0.001) and PAV per 1% increase (1.103 (1.051–1.158), p<0.001) were significantly predicting true positive findings. Conclusion Discrepancy between AI-QCTischemia and PET perfusion was found in 32% of patients. Age, typical angina, and features in AI-QCTischemia algorithm, such as diameter stenosis and PAV are associated with normal perfusion in abnormal AI-QCTischemia ("false positive") patients. On the other hand, normal AI-QCTischemia in abnormal PET perfusion ("false negative") patients was associated with higher RFR suggesting microvascular disease, which was found in 34% of these patients.Scatter plot
Authors 9
-
Affiliation as printed
Turku PET Centre , Turku , Finland
-
Affiliation as printed
Turku PET Centre , Turku , Finland
-
Affiliation as printed
Turku PET Centre , Turku , Finland
-
Takeru Nabeta Aachen
Leiden University Medical Center
Affiliation as printed
Leiden University Medical Center , Leiden , Netherlands (The)
-
Affiliation as printed
Cleerly inc. , New York , United States of America
-
Affiliation as printed
Cleerly inc. , New York , United States of America
-
Affiliation as printed
Cleerly inc. , New York , United States of America
-
Affiliation as printed
Turku PET Centre , Turku , Finland
-
Affiliation as printed
Turku PET Centre , Turku , Finland
Cited by 0 stored of 0
No patents citing this paper on Lens.org (checked 2026-10-11).