Artificial intelligence quantification and experienced reader computed tomography analysis for differentiating normal from minimally and mildly diseased coronary arteries: an early real-world compatibility study.
Journal:
The international journal of cardiovascular imaging
Published Date:
Feb 28, 2025
Abstract
Differentiating normal from minimally and mildly diseased coronary arteries on coronary computed tomographic angiography (CCTA) is crucial, impacting treatment decisions due to the extremely low coronary artery event risk associated with the former. Artificial intelligence quantitative computed tomographic (AI-QCT) can potentially identify subclinical atherosclerosis in cases deemed normal by reader interpretation. We aimed to evaluate AI-QCT's ability to distinguish reader-determined normal coronary arteries from those with minimal and mild diseased on CCTA. We screened 849 consecutive patients without coronary artery stents or bypass grafts who underwent CCTA and AI-QCT for suspected coronary artery disease between October 2022 and February 2023. Clinical reads were blinded to AI-QCT results. 411 patients (mean age 60, 63% women) with qualifying results were categorized into normal coronary arteries (NORMAL: calcium score of 0 and reader CAD-RADS 0), minimal (MINIMAL: coronary calcium score of ≤ 10, CAD-RADS score of 1, and 1 or 2 segments with plaque), and mild (MILD: coronary calcium score > 10 and < 100, CAD-RADS 1 or 2, and 1-3 segments with plaque) disease based on reader interpretation. AI-QCT results were compared among the categories and Youden index directed area-under-curve (AUC) analysis was employed to determine the optimal total plaque volume threshold distinguishing NORMAL from the other categories. Among the 411 patients, there were 235 NORMAL, 46 MINIMAL, and 130 MILD cases. AI-QCT detected no total plaque in 61/235 (26.0%) NORMAL cases. From NORMAL to MINIMAL to MILD, AI-QCT showed significant stepwise increases in total plaque volume (mean 7.7 mm vs. 22.5 mm vs. 40.5 mm, p < 0.001 all pairwise comparisons) and noncalcified plaque volume (mean 6.7 mm vs. 17.3 mm vs. 24.4 mm, p < 0.01 all pairwise comparisons). An AI-QCT total plaque volume of < 12.3 mm identified 189/235 (80.4%) NORMAL cases and excluded 136/176 (77.3%) MINIMAL and MILD cases, with an AUC of 0.86. AI-QCT revealed significantly higher total plaque volume in reader-determined MINIMAL and MILD compared to NORMAL cases, showing promising concordance with reader interpretation. Our analysis suggests that an AI-QCT total plaque volume of < 12.3 mm may serve as a useful initial cut-off for CCTA likely to be interpreted as normal by an experienced reader.
Authors
Keywords
Aged
Artificial Intelligence
Asymptomatic Diseases
Computed Tomography Angiography
Coronary Angiography
Coronary Artery Disease
Coronary Vessels
Female
Humans
Male
Middle Aged
Observer Variation
Plaque, Atherosclerotic
Predictive Value of Tests
Radiographic Image Interpretation, Computer-Assisted
Reproducibility of Results
Retrospective Studies
Severity of Illness Index
Vascular Calcification