Coronary stenosis assessment: AI-based CT quantification, visual analysis of invasive angiography, and quantitative coronary angiography.
Authors
Affiliations (8)
Affiliations (8)
- Department of Radiology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
- Faculty of Medical Imaging Technology, College of Health Science and Technology, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
- Department of Radiology, Affiliated Zhoupu Hospital, Shanghai University of Medicine and Health Science, Shanghai, China.
- Shukun Technology Co., Ltd, Beijing, China.
- Department of Radiology, Beijing Anzhen Hospital, Beijing Institute of Heart, Lung, and Blood Vessel Diseases, Capital Medical University, Beijing, China.
- Department of Radiology, The First Hospital of China Medical University, Shenyang, China. [email protected].
- Department of Radiology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China. [email protected].
- Faculty of Medical Imaging Technology, College of Health Science and Technology, Shanghai Jiao Tong University School of Medicine, Shanghai, China. [email protected].
Abstract
To evaluate the diagnostic performance for coronary stenosis of artificial intelligence (AI)-based CT quantification, manual CT quantification, and visual invasive coronary angiography (ICA) assessment against quantitative coronary angiography (QCA). This retrospective study included patients who underwent both coronary CT angiography (CCTA) and ICA within 1 month. Diameter stenosis (DS) was quantified on CCTA images both automatically by a deep-learning model and manually by radiologists, and was assessed visually on ICA. Diagnostic performance for detecting obstructive stenosis (DS ≥ 50% and ≥ 70%) was assessed against QCA using receiver operating characteristic (ROC) curve analysis on per-patient, per-vessel, and per-segment levels. Agreement with QCA was evaluated using Bland-Altman analysis. A total of 368 patients were included. AI-based CT quantification demonstrated high diagnostic accuracy. For DS ≥ 50%, area under the curves (AUCs) were 0.93 (per-patient), 0.94 (per-vessel), and 0.96 (per-segment). For DS ≥ 70%, AUCs were 0.85, 0.89, and 0.90, respectively. AI-based CT quantification significantly outperformed manual CT and visual ICA assessments at most levels in terms of AUC (all p < 0.05), except vs manual quantification by a senior radiologist at the segment level for ≥ 70% stenosis (p = 0.060). Bland-Altman analysis showed superior agreement between AI-based CT quantification of DS and QCA (mean differences: 1.5% per-patient, -0.7% per-vessel, -1.2% per-segment) compared with manual CT quantification and visual assessment of ICA. AI-based CT quantification demonstrated high diagnostic performance for obstructive stenosis with good agreement against QCA, outperforming manual CT quantification and visual assessment of ICA in most scenarios, with the exception of segment-level assessment of ≥ 70% DS compared with manual quantification by senior radiologists. AI-aided CT quantification outperforms both manual CT quantification and visual ICA analysis in diagnosing obstructive stenosis using QCA as the reference standard, providing an objective tool to reduce diagnostic uncertainty and guide subsequent patient management. Manual coronary stenosis quantification is subject to significant inter-observer variability. AI-based CT quantification outperforms both manual CT quantification and visual ICA analysis. Discordance between AI results and the angiographer's visual estimation should prompt a physiological evaluation.